The evaluative information ecology: On the frequency and diversity of “good” and “bad”. Christian Unkelbach, Alex Koch & Hans Alves. European Review of Social Psychology, Volume 30, 2019 - Issue 1, Pages 216-270. Nov 24 2019. https://doi.org/10.1080/10463283.2019.1688474
Abstract: We propose the Evaluative Information Ecology (EvIE) model as a model of the social environment. It makes two assumptions: Positive “good” information is more frequent compared to negative “bad” information and positive information is more similar and less diverse compared to negative information. We review support for these two properties based on psycho-lexical studies (e.g., negative trait words are used less frequently but they are more diverse), studies on affective reactions (e.g., people experience positive emotions more frequently but negative emotions are more diverse), and studies using direct similarity assessments (i.e., people rate positive information as more similar/less diverse compared to negative information). Next, we suggest explanations for the two properties building on potential adaptive advantages, reinforcement learning, hedonistic sampling processes, similarity from co-occurrence, and similarity from restricted ranges. Finally, we provide examples of how the EvIE model refines well-established effects (e.g., intergroup biases; preferences for groups without motivation or intent) and how it leads to the discovery of novel phenomena (e.g., the common good phenomenon; people share positive traits but negative traits make them distinct). We close by discussing the benefits relative to the drawbacks of ecological approaches in social psychology and how an ecological and cognitive level of analysis may complement each other.
Keywords: Evaluation, ecology, halo effects, person perception, intergroup biases
Excerpts. Check the full paper for figures, tables, references, etc.
Implications
So far, we provided evidence and explanations for the higher frequency of
positive information relative to negative information (i.e., traits, experiences,
behaviours), and the higher similarity of positive information to other positive
information. In the remainder, we aim to back up our claim that the interaction of these properties with well-established social-cognitive principles within
the organism may lead to the discovery of novel phenomena and alternative
explanations for classic social psychological findings. We will address halo
effects, the relation of similarity and liking, the relation of frequency and liking,
as well as the field of intergroup biases. In the following review of our empirical
findings, anything that is reported as a difference is significant (i.e., probability
of the test statistic under the H0 is p < . 05), unless indicated otherwise; all
reported experiments had proper power considerations and reported all conditions, all data exclusions, and all variables. In addition, we predicted the
empirical findings from the assumed properties and did not derive the EvIE’s
properties from these studies; thus, the following experimental work supports
the EvIE as a general model for people’s social reality.
Halo effects: being honest makes you industrious, but lying does not make you lazy
Halo effects are among the best-established findings in psychology. Thorndike
(1920) coined the term when he observed a “constant error in psychological
ratings”: When army officers were evaluated by their superiors, theoretically
independent dimensions constantly correlated more highly than they should.
Thus, raters either used information on one dimension to rate another dimension or made inferences from a global impression about the to-be-rated target
(Cooper, 1981). Probably the most famous halo effect is from ratings of
physical attractiveness to ratings of intelligence or morality, famous under
the “What is beautiful is good” label (Dion, Berscheid, & Walster, 1972).
Based on our assumptions about the EvIE, an intriguing prediction
follows from the similarity property, namely that halo effects should be
most apparent given positive traits and rating dimensions, but less pronounced given negative traits. This is a strong prediction insofar as there is
consensus in the literature that negative information has more impact than
positive information on social evaluations (e.g., Kanouse & Hanson, 1972;
Peeters & Czapinski, 1990; Skowronski & Carlston, 1989).
To test this idea, Gräf and Unkelbach (2016) presented participants with
targets’ positive or negative traits as well as behaviours from two dimensions
of social perception (Bakan, 1966; see also Abele & Wojciszke, 2018), namely
communion (e.g., being honest) and agency (e.g., being industrious), and
asked participants to rate the targets on other traits either from the same or
the other dimension. Across three experiments, Gräf and Unkelbach investigated halo effects on 30 traits and 48 different behaviours. Participants
observed a target showing either a trait label or a behavioural description and
were asked how likely it was that the target would possess another trait
(Experiments 1 and 2) or would show another behaviour (Experiment 3).
Importantly, they varied the valence and the social perception dimension.
For example, participants saw a lying target (i.e., a negative communion trait)
and answered how likely this person was to also be lazy (i.e., a negative
agency trait), or in another trial, how likely this person was also to be egoistic
(i.e., negative communion trait). Similarly, they would see an honest target
and answer how likely this person was to also be industrious (or, in another
trial: helpful). Thus, the trials tested whether halo effects, an inference from
one behaviour/trait to another behaviour/trait, vary as a function of trait/
behaviour valence and as a function of within/between dimension inferences
on the two fundamental dimensions of social perception.
Figure 5 shows the data from these three experiments. As predicted from
the EvIE’s similarity property, positive traits and behaviours lead to substantially stronger halo effects, both within and across the dimensions of communion and agency (Gräf & Unkelbach, 2016; Exp. 1 to 3; see also, 2018, for
a conceptual replication). These findings are difficult to reconcile with classic
assumptions about the unconditional higher impact of negative information
on social evaluations, but they follow from the EvIE’s similarity property. The
results may also explain apparent features in the literature, namely why there
are few published studies showing “negative” halo effects (i.e., “horn” effects),
simply because they usually do not exist (i.e., lying does not make you lazy).
The EvIE’s frequency property also suggests an intriguing point; namely,
that the observed halo effects might not be an error in ratings (Thorndike,
1920), but a true property of the ecology. Similar to our argument concerning
how higher similarity follows from a higher frequency, the higher frequency
of occurrence of positive traits and behaviours also implies that any positive
trait or behaviour is more likely to co-occur. People should, therefore, learn
that positive traits and positive behaviours appear together on a person-level.
If our assumption about the EvIE’s frequency property is correct, then the
personality profile of being both honest and industrious is factually more
likely than the profile of being dishonest and lazy. From an ecological view,
the constant error in ratings observed by Thorndike might not be entirely an
error after all, but a generalisation of observed ecological co-occurrence to
a task involving trait ratings in a psychology experiment. Investigating this
alternative source for halo effects provides a fascinating venue for future
research.
Similarity and liking: your friends are all alike
The EvIE model states that positive information is more similar and less
diverse compared to negative information; as Figure 4 illustrates, there is
only one way (or fewer ways) to be good compared to the many ways
someone might be bad. One implication of this ecological property is that
liked people (i.e., someone’s friends) should be more similar to one another
compared to disliked people.
This is an interesting prediction, because, based on the hedonic sampling
principle discussed above, people should spend more time with other people
they like compared to people they do not like (Denrell, 2005). This increase
in spent time should lead to more knowledge about liked people, and thereby
to a more differentiated representation of these liked others. Smallman and
Roese (2008) explicitly stated this as follows: “to cherish a loved one is to
relish the fine nuances of his or her personality” while “the rejected and
forsaken are construed on a relatively surface level” (p. 1228). However, if we
assume that people like each other because they possess positive traits,
attributes, or qualities which makes them likeable, the EvIE’s assumed
similarity property predicts that these people should be very similar, particularly in comparison to disliked people. Their mental representation might
be highly differentiated as proposed by Smallman and Roese, but this differentiation does not make them dissimilar, just because the properties (i.e.,
traits and behaviours) that lead to liking are factually highly alike.
Alves, Koch, and Unkelbach (2016) conducted seven experiments to test
whether people see other people they like as more similar to one another
compared to people they dislike. We discuss five of these experiments in the
following. The basic paradigm was straightforward. Participants generated
names of target persons they liked and of targets they disliked. Then, they
used the spatial arrangement method described above (see Figure 1‘s right
panel; Hout et al., 2013) or pairwise similarity ratings (see Figure 1‘s left
panel) to arrange these targets on the screen according to the similarities of
their personalities. They also provided ratings of the time spent together with
these people and of how much they knew about them. As expected, participants reported having spent more time with liked compared to disliked
targets, and they reported knowing more about the liked compared to the
disliked targets. Yet, in line with the prediction from the EvIE, participants
consistently reported higher similarity for liked and disliked targets.
Figure 6 provides a summary of the similarity judgements from
Experiments 1, 3 and 5. Experiment 1 used target persons participants knew
personally with spatial arrangement to assess similarity. Experiment 3 used
target persons participants knew personally with pairwise comparisons to
assess similarity. Experiment 5 used celebrity targets with pairwise comparisons. As Figure 6 shows, participants consistently reported liked targets to be
more similar than disliked targets, despite spending more time with them. We
omit Experiments 2 and 6 here; Experiment 2 replicated Experiment 1 with
target valence manipulated between participants and Experiment 6 replicated
Experiment 5 with a larger set of celebrity targets.
Experiment 4 tested the underlying EvIE structure directly. Participants
generated as many traits as they could for each of the four liked and disliked
targets they named. First, in line with the assumed greater knowledge for liked
targets, participants generated on average 6.9 traits for liked, but only 3.9 traits
for disliked targets. Second, we computed the probability that a trait was shared
among the targets. Figure 7‘s left panel shows the relevant data. The probability
that participants generated shared traits among liked targets was substantially
higher compared to disliked targets. This was true within participants’ eight
targets, but also across participants; that is, even across participants, liked targets
were more likely to share traits and therefore be more similar, providing support
for the assumption that there are ecologically fewer ways to be liked than to be
disliked. This difference in shared traits also held when controlling for the
number of generated traits in a regression analysis.
Experiment 7 then flipped the paradigm and asked participants to generate the names of two people they personally knew without specifying
whether they had to be liked or disliked. Instead, we asked them to generate
either positive traits or negative traits that described each of the two targets.
After providing as many traits as they could, we asked participants to rate the
similarity of the two targets. First, as expected, participants showed the
reversed effect as well – generating positive traits made the two targets appear
more similar compared to generating negative traits. As the targets were
selected in both conditions before we asked for positive or negative traits, any
alternative explanation in terms of differential target generation is taken care
of. In addition, participants generated more traits in the positive traits
condition, 6.4 on average, compared to the negative traits condition, where
they generated only 3.8 traits on average. Replicating Experiment 4, as shown
in Figure 7‘s right panel, the probability that participants generated shared
traits among positive traits was substantially higher compared to negative
traits. This was again true within and also across participants, and also when
controlling for the absolute number of traits generated.
Across seven experiments, of which we summarised five here, we found that
positive traits are more frequently generated and these generated traits also are
more likely to be found across targets, leading to the conclusion that liked
people tend to be seen as alike. In particular, the within-participant comparisons might partially follow from intra-psychic mechanisms (e.g., motivated
reasoning to see your friends as similar and good); however, the effects acrossparticipants are difficult to explain without the presented EvIE model (see
Figure 7).
Frequency and valence: the common good in person perception
In another series of experiments (Alves, Koch, & Unkelbach, 2017b), we
tested a prediction from the frequency property discussed above: If positive
information is more frequent, then it should more likely co-occur with other
positive information compared to negative information. Across people, this
implies that people have positive traits in common, but their negative traits
make them distinct: “Those attributes that connect different people and that
define their similarities are usually good attributes. Those attributes that
distinguish different people and make them unique are often bad attributes.”
(p. 512). This prediction follows solely from the frequency property and does
not depend on the similarity of the information.
For illustration, let us again consider the formal relation of shared and
unshared positive and negative attributes, as we did above for personality
traits. For example, positive attributes may have the probability of being
present in any person of p(pos) = 0.6, and negative attributes may have
a probability of being present of p(neg) = 0.2. The probability of a shared
attribute (i.e., being simultaneously present in two persons) being positive is
then p(positive|shared) = p(pos)*p(pos) = 0.36, while the probability for the
negative attribute is p(negative|shared) = p(neg)*p(neg) = 0.04. In other
words, if a positive trait is three times more likely in the ecology than
a negative trait, it is nine times more likely to be shared than a negative
trait. This leads to two hypotheses: positive traits should be more likely to be
shared amongst targets compared to negative traits, p(shared|positive) > p
(shared|negative), and shared traits should be more likely to be positive
compared to negative traits, p(positive|shared) > p(negative|shared).
To test these hypotheses, Alves et al. (2017b) asked participants to
sample traits of target persons. Experiments 1a and 1b tested the first
prediction, p(shared|positive) > p(shared|negative). In Experiment 1a
(n = 41), participants generated two people they knew personally and
then generated four positive traits and four negative traits for one of the
two. Then, we asked them which of the eight traits also described the other
person. In line with our first prediction, participants assigned on average
3.4 positive traits (i.e., almost all) two both targets. Out of the four negative
traits, they assigned only 1.1 to both targets. Figure 8‘s left panel reports the
respective conditional probabilities for positive and negative traits.
To generalise this result, Experiment 1b (n = 82) asked participants to
generate 10 target persons. Then, we randomly sampled a given set of four
positive and four negative traits from Experiment 1a and participants had to
indicate to which of the 10 targets each of the traits applied. Replicating 1a,
participants assigned on average 3.1 of the positive traits to a target from
their own sample, but only 1.2 of the negative traits. Figure 8 shows the
resulting conditional probabilities. As the left panel shows, positive traits
were much more likely to be shared across participants compared to negative
traits. And as the trait and target generation were separated in Experiment
1b, this replication provides support for our ecological argument.
Experiment 2 in this series of “common good” experiments (Alves et al.,
2017b) tested the second prediction: if a trait is shared as opposed to
unshared, it should be more likely positive, and thus, p(positive|shared) > p
(negative|shared). Participants again generated two target names; then, we
asked them for either shared or unshared traits. We asked for four shared
traits in the former, and two traits that belonged uniquely to the first target,
and two traits that belonged uniquely to the second target, in the latter
condition. Then, participants rated the valence of the generated traits.
Figure 8‘s right panel shows the probabilities: Overall, participants generated
more positive traits than negative traits in both conditions, reflecting the
general positivity prevalence. Yet, in the shared condition, 3.5 traits were
positive on average, and only 0.2 traits were negative. In the unshared
condition, 2.3 traits were positive and 1.3 traits were negative. Thus, the traits
people have in common are usually positive.
Experiment 4a (n = 176) in Alves et al. (2017b) aimed to show that
searching for similarities (i.e., shared traits) amplifies the ecological default,
and searching for differences (i.e., unique traits) attenuates it. Thus, the
experiment replicated Experiment 2 but included a “natural” condition, in
addition to the “shared” and “unshared” conditions. The “natural” condition
asked participants to generate traits for two target persons without specifying
whether these should be shared or unshared traits. Again, across conditions,
participants generated substantially more positive traits: about 4.8 traits out
of six were positive. However, the probability of generating a positive trait
varied as a function of the traits being generated as “shared”, “unshared”, or
“natural” (i.e., without specific instructions). Figure 9 shows these probabilities of a trait being positive. The probability of a trait being positive was
smaller in the natural condition compared to the “shared” condition, and
smaller in the “unshared” condition compared to the “natural” condition.
Thus, as predicted, looking for similarities amplifies the prevalence of positive traits, while looking for differences attenuates it.
A basic drawback in the reported “common good” studies so far is that
participants self-generated targets, which makes the observed “common good”
effect less surprising, as most people might generate people they know and also
like, and the phenomenon might follow from the “my friends are all alike”
effect described above. However, the present approach is different as it is solely
based on the proposed EvIE’s frequency property. The similarity property
implies that positive information should always be more similar to other
positive information (again; there is only one way to be good), and thus, as
long as people have friends they like, these should be alike.
The present “common good” effect, however, follows only if the available
information is predominantly positive. This leads to the reverse prediction if
the available information is predominantly negative. Thus, in Experiments 5
and 6 in Alves et al. (2017b) “common good” series, participants did not
generate targets, but we provided liked and disliked targets for which the
available trait information should be either predominantly positive or negative,
respectively. To do so, Experiment 5 took advantage of the US’s bipartisan
political structure of Democrats and Republicans and recruited 310 US participants online. Half of the participants generated either shared or unshared
traits for Mitt Romney and George W. Bush, two well-known republicans, and
the other half did the same for Bill Clinton and Barack Obama, two wellknown democrats. To divide the sample, we asked participants how much they
liked these political figures; 160 participants reported liking the politicians in
their respective conditions, and 143 participants reported disliking them.
Seven participants reported neither liking nor disliking them and were
excluded from the analysis.
In Experiment 6 (n = 307), we sampled the target persons from a list of the
10 most popular and most unpopular people other participants generated.
The 10 most popular people for US citizens were Abraham Lincoln,
John F. Kennedy, Elvis Presley, Martin Luther King, Oprah Winfrey, Taylor
Swift, George Washington, Michael Jordan, Beyoncé Knowles, and Jesus
Christ. The 10 most unpopular people were Adolf Hitler, Donald Trump,
George W. Bush, Osama Bin Laden, Saddam Hussein, Joseph Stalin, Kim
Jong Un, Justin Bieber, Fidel Castro, and Kanye West. For example, participants generated four traits that Abraham Lincoln and Elvis Presley shared or
two traits that were unique to Lincoln and Presley, respectively. Each pairing
was randomly created for each participant. In the negative targets condition,
for example, participants generated traits that Adolf Hitler and Justin Biber
shared, or two traits that were unique to each of these targets.
Figure 10 shows the results for these two studies, plotting the frequency of
traits being positive and negative as a function of being shared or unshared
among the target persons. For liked targets, the trait frequencies replicate the
previous studies. Both for liked political figures of that time as well as
consensually liked persons, looking for similarities yielded many positive
traits, and few negative traits. Looking for differences yielded fewer positive
traits and more negative traits. However, when participants disliked the
targets, that is, when operating in an ecology of predominantly negative
information, they generated more negative traits in the shared compared to
the unshared condition. Conversely, they provided fewer positive traits in the
shared compared to the unshared condition. This pattern of results provided
distinct evidence for the “common good” implication of the EvIE’s assumed
frequency property. Looking for similarities between targets amplifies, and
looking for differences between targets attenuates, the underlying base-rate;
and this base-rate is, in most cases, marked by a high frequency of positive
information, leading to a “Common Good” phenomenon.
Thus, based on the assumption that positive information is more frequent,
we predicted and found a novel phenomenon in person perception – the
common good effect. The attributes people have in common are usually good
attributes, and negative attributes are rather unique. In addition, searching
for similarities leads to the discovery of the common good, while searching
for differences subjectively attenuates the prevalence of positive information.
Intergroup biases: a cognitive-ecological explanation
Having shown implications of positive information’s higher similarity
(strong halo effects from positive traits; friends are more alike than enemies)
and positive information’s higher frequency (the common good phenomenon), our final example provides a genuinely new explanation for intergroup biases (Alves, Koch, & Unkelbach, 2018), by combining basic
cognitive processes with our assumptions about the EvIE.
One of the most prominent effects in social psychology is that people tend to
devalue minorities (e.g., refugees, immigrants) and out-groups (e.g., rival sport
teams, other states). There is a wealth of models and theories to explain these
biases (e.g., Tajfel and Turner’s Social Identity Theory, 1979; or Brewer’s
theory of optimal distinctiveness, 1991). However, taking the assumed EvIE
properties offers a novel explanation.
For this explanation, we only need the assumption that out-groups and
minorities are “novel” groups in comparison to ingroups and majorities. This
is highly plausible, as people usually come in contact first with their ingroups
(e.g., family, fellow citizens) and majorities (e.g., Whites, Christians); they
learn about outgroups and minorities later and these groups are then novel
in comparison to the former.
On the cognitive side, novel groups are defined in relation to existing groups
(i.e., ingroups, majorities) by the attributes that make them unique, rather than
by the attributes they have in common with existing groups (Hodges, 2005;
Sherman et al., 2009; Tversky & Gati, 1978). On the ecological side, as the
presented evidence suggests, positive attributes are less diverse or more similar
than negative information, and positive information is more frequent than
negative information. Consequently, unique attributes that differentiate
a novel group from already-known groups are likely to be negative.
Thus, the argument is as follows: Minorities and outgroups are most likely
novel groups to social perceivers, compared to majorities and ingroups.
Novel groups are defined by their unique attributes (i.e., the cognitive part)
and unique attributes are most likely negative (i.e., the ecological part),
leading to an association between outgroups and minorities with negative
attributes, which in turn may cause negative stereotypes and prejudice.
To test this explanation, we invited participants to take the role of space
explorers. On a novel planet, they would encounter members of two alien
tribes. We used the neutral aliens provided by Gupta et al. (2004) as stimuli.
Participants would encounter one member of the first tribe and receive
information about one of the alien’s trait; that is, they saw a picture of the
alien and the alien’s respective trait (e.g., helpful, intelligent, anxious, or
aggressive). After participants had encountered six members of the alien
tribe, we instructed participants to imagine that they would now continue
their travels and encounter another alien tribe. Then, they would learn about
the traits of six members of the second tribe. In the real world, people should
probabilistically learn first about members of their ingroup before learning
about members of outgroups. Similarly, they are more likely to meet majority
group members before meeting minority group members. Thus, the first
tribe is functionally similar to a majority or ingroup, and the second tribe is
functionally similar to minorities or out-groups. After these learning phases,
participants chose which group they preferred.
The central manipulation across three experiments was the trait pool from
which we assigned the two tribes’ traits. After learning, we asked participants
which tribe they prefer; that is, we elicited a binary preference choice
between the first and the second tribe as the central dependent variable.
Experiment 1 manipulated whether the positive or whether the negative
attributes were shared or unshared among the two groups. That is, in one
condition, the groups’ positive attributes were identical, while their negative
attributes differed, and this was reversed in the other condition. Table 4‘s left
section presents the resulting preference frequencies. As predicted from our
cognitive-ecological explanation, participants preferred the first group when
the positive attributes were shared and negative attributes were unique, but
preferred the second group when positive attributes were unique and negative attributes were shared. In other words, although the distribution of
positive and negative traits was identical, there was a bias against the novel
group in a standard ecology (i.e., where negative information is unique),
which reversed as a function of the trait ecology.
Experiment 2 then manipulated the similarity of evaluative information in
the ecology. We created two attribute ecologies. In the standard ecology,
positive attributes were less diverse compared to negative attributes. In the
reversed ecology, negative attributes were less diverse. We manipulated
diversity by the number of unique traits in a given ecology. In the standard
ecology condition, we randomly sampled each alien tribe’s three positive
traits from a set of four traits, while we sampled the three negative traits from
a set of 16 traits (i.e., there were more ways to be negative). In the reversed
ecology condition, we sampled the alien tribes’ three negative traits from a set
of four traits, and their positive traits from a set of 16 traits (i.e., there were
more ways to be positive). Consequently, in the standard ecology, the
positive traits were likely to be shared and the first tribe should be preferred.
In the reversed ecology, the negative traits were likely to be shared and
the second tribe should be preferred. As Table 4‘s middle panel shows, the
preference frequencies replicated Experiment 1. Participants preferred the
first group in the standard ecology (i.e., when negative attributes were likely
unique), but in the reversed ecology they preferred the second group (i.e.,
when positive attributes were likely unique).
Experiment 3 then manipulated the EvIE’s second property, the frequency
of evaluative information. In the standard ecology, both groups possessed
more positive than negative attributes, while in the reversed ecology, negative attributes were more frequent. Specifically, in the standard ecology, both
tribes displayed four positive traits and one negative trait. Both positive and
negative traits were randomly sampled from a set of six positive and six
negative traits. In the reversed ecology, both tribes displayed four positive
and one negative trait. Consequently, in the standard ecology (positive
frequent), unique attributes were likely to be negative, while in the reversed
ecology, unique attributes were likely to be negative.
Table 4‘s right section shows the respective preference frequencies.
Replicating Experiments 1 and 2, participants preferred the first group in
the standard ecology, but they preferred the second group in the reversed
ecology. One apparent feature of Table 4 is that the standard ecologies (i.e.,
when negative information is unique) yield stronger differences between the
tribes, while the preference differential is less strong when positive information is unique. This is actually in line with our overall assumptions about the
EvIE. We did not control for the connotative similarity of the positive and
negative traits, but research on the similarity of personality traits
(Bruckmüller & Abele, 2013; Gräf & Unkelbach, 2016; Leising et al., 2012)
shows that positive traits are more similar to each other compared to
negative traits. By implication, the positive unique traits were, less “unique”
compared to the negative unique traits. This differential valence asymmetry
explains at least part of the differential impact of the ordering.
Thus, across three experiments, participants associated a novel group
with its unique attributes, which differentiate the group from previously
encountered groups. Depending on the ecology’s properties, unique attributes were more likely to be positive or negative, and participants’ preferences followed accordingly. As the general structural properties of the EvIE
make unique attributes more likely negative, p(negative|unique) > p(positive|unique), an evaluative disadvantage for novel groups, and thereby for
minorities and outgroups, follows. In other words, people do not need
a real conflict (Sherif, Harvey, White, Hood, & Sherif, 1961), motivated
reasoning (Kunda, 1990), or a hostile personality structure to show differential preferences for minorities and outgroups (Altemeyer, 1998). Rather,
all they need is a cognitive system that tries to differentiate different groups
in an ecology that is marked by high similarity and a high frequency of
positive information
Summary of the implications
We have provided two examples of how our EvIE model refines our knowledge
about classic and important social psychological phenomena. First, halo effects;
we have delineated and shown that halo effects appear predominantly for
positive traits, but are largely absent for negative traits, despite the typically
assumed stronger impact of negative information (Baumeister et al., 2001; Ito,
Larsen, Smith, & Cacioppo, 1998). Second, intergroup biases; we have provided a cognitive-ecological explanation for intergroup biases that do not rely
on motivated reasoning (Kunda, 1990; Tajfel & Turner, 1979), but builds solely
on cognitive processes that interact with the EvIE’s properties.
We have also provided two examples that illustrate the discovery of genuinely new phenomena. First, people’s friends are all alike. Based on the proposed
similarity property, we have shown that people perceive others they know and
like as more similar to one another, just because there is not much room for
variety on the positive side. Second, the common good phenomenon; based on
the proposed frequency property, we have shown that what people have in
common are usually positive attributes, just because negative attributes are
infrequent, and their joint occurrence is therefore unlikely.
Saturday, November 30, 2019
Is Orgasmic Meditation a Form of Sex?
Siegel, Vivian, Caryn Roth, Elisabeth Bolaza, and Benjamin Emmert-Aronson. 2019. “Is Orgasmic Meditation a Form of Sex?” SocArXiv. November 27. doi:10.31235/osf.io/89fvt
Abstract: Orgasmic Meditation(OM) is a structured, partnered meditative practice in which one person, who can be any gender, strokes the clitoris of their partner for 15 minutes. As such, it resembles a sexual activity. OM is taught as a practice that is distinct from sex, and we wondered whether people who engage in OM actually maintain that distinction themselves. We conducted an online convenience sample survey including qualitative open-ended text questions and quantitative Likert-style questions that was distributed to email listservs for practitioners of OM. The 30-item questionnaire included questions designed to differentiate the potentially related concepts of OM, seated meditation, fondling, and sex, as bases for comparison. The quantitative results of this mixed method study show that OM practitioners view the practice as significantly more similar to meditation than to sex or fondling. These results were consistent, regardless of whether the question was asked in the positive or negative and whether OM was being compared to one behavior individually or to multiple behaviors at the same time. The distinction between OM and sex/fondling rapidly becomes more pronounced as practitioners complete more OMs. This suggests that the novelty of genital touching in meditation may diminish over time, as practitioners get used to the more alternative point of focus. The results of this study have implications for the practice and how it is approached and regulated.
Abstract: Orgasmic Meditation(OM) is a structured, partnered meditative practice in which one person, who can be any gender, strokes the clitoris of their partner for 15 minutes. As such, it resembles a sexual activity. OM is taught as a practice that is distinct from sex, and we wondered whether people who engage in OM actually maintain that distinction themselves. We conducted an online convenience sample survey including qualitative open-ended text questions and quantitative Likert-style questions that was distributed to email listservs for practitioners of OM. The 30-item questionnaire included questions designed to differentiate the potentially related concepts of OM, seated meditation, fondling, and sex, as bases for comparison. The quantitative results of this mixed method study show that OM practitioners view the practice as significantly more similar to meditation than to sex or fondling. These results were consistent, regardless of whether the question was asked in the positive or negative and whether OM was being compared to one behavior individually or to multiple behaviors at the same time. The distinction between OM and sex/fondling rapidly becomes more pronounced as practitioners complete more OMs. This suggests that the novelty of genital touching in meditation may diminish over time, as practitioners get used to the more alternative point of focus. The results of this study have implications for the practice and how it is approached and regulated.
Discussion
This is the first study of its kind on the topic of Orgasmic Meditation and how practitioners
perceive this practice. There is little research on Orgasmic Meditation in general, and this study
helps place it in the larger context of meditation and sexuality, two fields with much ongoing
research. The quantitative results of this mixed method study show that OM practitioners view
the practice as significantly more similar to meditation than to sex or fondling. These results
were consistent, regardless of whether the question was asked in the positive or negative (i.e.
disagreeing with the question OM is sex, agreeing with the question OM is not sex) – and
whether OM was being compared to one behavior individually or to multiple behaviors at the
same time.
The results of this study also show that the distinction between OM and sex/fondling rapidly
becomes more pronounced as practitioners complete more OMs. This suggests that the novelty
of genital touching in meditation may diminish over time, as practitioners get used to the more
alternative point of focus. If OM is viewed differently by different groups of practitioners, there
may be programmatic and policy implications in the management of OM instruction. For
example, if new OM practitioners are more likely to conflate OM and sex, there is a heightened
likelihood of unintended outcomes related to sexual stigma, trauma, or perceived sexual
harassment at that stage. Such sexuality-related side-effects should be addressed in the
instruction and in communications with participants, and additional supports may be necessary
to help new practitioners navigate these complexities until they are clear on the practice and
how to internalize their experiences.
In addition, of the gender and sexual orientation combinations with large enough sample sizes
to study, bisexual women viewed the practice as most different from sex/fondling. This was
surprising because of the scientific research showing that lesbians were more likely to classify
manual-genital contact as sex. The finding suggests that the context highly impacts how an act
of a sexual nature is perceived. The same physical behavior that in the bedroom may be
considered sex, is considered meditation in the container of a practice setting. The fact that this
subgroup was more adamant that the practice is not sex highlights how much the intention
behind the act makes a difference. That is, the practice is not differentiated from sex because of
the actual physical action, but because of the intention behind it.
Future Research
Given the fact that OM involves genital stroking, it is therefore interesting to ask the question
why the response to the survey is so clear. One possibility is that there are certain aspects of
the practice itself that clearly divides it from sex and fondling. For example, the stroker in OM is
fully clothed and wears gloves. There is no eye gazing or kissing. Practitioners generally do the
practice away from their beds, usually on the floor with meditation and other cushions that do
not resemble bed pillows. Practitioners are also taught that if the practice stirs desires for sex,
that they complete the practice and put away the practice supplies before deciding whether to
have sex. If practitioners follow this guidance, then OM will effectively be separated from sex or
from activities that might lead to sex.
It is also important to note that OM is not the only practice that confounds the traditional
conceptualizations of sex and meditation. Tantra, for example, seeks to use sexual energy to
reach a meditative state (Nagaraj 2013). It would be interesting to know how practitioners of
tantra would respond to a similar survey.
Individuals with dark traits have the ability to empathize, but have a low disposition to do so
Individuals with dark traits have the ability but not the disposition to empathize. Petri J. Kajonius, Therese Björkman. Personality and Individual Differences, November 30 2019, 109716. https://doi.org/10.1016/j.paid.2019.109716
Highlights
• We tested if the Dark Triad was best described by ability- or trait-empathy.
• Dark Triad had no relationship with ability-empathy.
• Dark Triad had a strong negative relationship with trait-empathy.
• Cognitive ability explained ability-empathy.
• Dark personalities seem cognizant, but not inclined to empathize.
Abstract: Empathy is fundamental to social cognition and societal values. Empathy is theorized as having both the ability as well as the disposition to imagine the content of other people's minds. We tested whether the notorious low empathy in dark personalities (Machiavellianism, psychopathy, and narcissism; the Dark Triad) is best characterized by a lack of capacity (ability) or lack of disposition (trait). Data was collected for 278 international participants through an anonymous online survey shared on the online platform LinkedIn, consisting of trait-based Dark Triad personality (SD3) and empathy (IRI), and cognitive ability (ICAR16) and ability-based empathy (MET). Dark personality traits had no relationship with ability-based empathy, but strongly so with trait-based empathy (β = -0.47). Instead, cognitive ability explained ability-based empathy (β = 0.31). The finding is that dark personalities in a community sample is normally cognizant to empathize but has a low disposition to do so. This finding may help shed further light on how personality is interlinked with ability.
Keywords: PersonalityDark triadEmpathyCognitive ability
From the introduction: Empathy is a core feature of human beings in social interaction (Myyrya, Juujärvi, & Pesso, 2010). No matter where in the world we live, individuals in any given community are expected to be able to cognizant and sensitive to other people's minds. Individuals who violate such values are often looked down upon in society (Persson & Kajonius, 2016). In personality psychology, there has been an increase of interest in so-called dark personality traits (Moshagen, Hilbig, & Zettler, 2018), which are characterized by violating social values (Kajonius, Persson, & Jonason, 2015). The idea behind the most used Dark Triad personality model (DT; Paulhus & Williams, 2002) is to capture the multidimensionality of complex traits leading up to this, and that these can be described by subclinical Machiavellianism (tendency to manipulate), psychopathy (callousness), and narcissism (grandiosity) (Jonason & Kroll, 2015). It is still unclear whether individuals scoring high on these dark personality traits are mostly lacking the capacity (ability) or mostly lacking the disposition (trait) to feel what others feel (see Keysers & Gazzola, 2014). The purpose of the present study is to explore empathy and to test the idea that it is not inability but more a lack of disposition that drives dark personalities’ low empathy.
4. Discussion
The present study aimed at investigating whether persons scoring high on the Dark Triad (Machiavellianism, psychopathy, and narcissism) mostly relate to the lack of ability or the lack of disposition to empathize. The results showed that it was more a lack of empathic disposition than inability that characterizes dark personalities. First, the Dark Triad had a very strong relationship with dispositional trait-based empathy. Second, the Dark Triad had a weak (almost non-existent) relationship with ability-based empathy. Third, cognitive ability explained most of ability-based empathy.
The first hypothesis of the present study were largely in line with previous research, which mainly has shown a consistent negative relationship between dark personality and trait-based empathy (BaronCohen, Wheelwright, Hill, Raste, & Plumb, 2001; Jonason, Lyons, Bethell, & Ross, 2013; Pajevic et al., 2018). The relationship between higher Dark Triad (SD3) and lower trait-based empathy (IRI) was large (Fig. 1). This effect size is trending towards high convergence, even more so if controlled for reliability and subscale variance – According to updated guidelines in psychology research, correlations between r = 0.00–0.09 should be interpreted as trivial to non-existent, r = 0.10–0.19 weak, r = 0.20–0.29 medium, and above r > 0.30 strong (Gignac & Szodorai, 2016). Interestingly, lack of empathy as measured by IRI entails not only lack empathic concern, but also no imagination for others’ minds (FT), not being cognizant in perspective taking (PT), and no distress for others’ welfare (PD). This result confirms the general notion that the Dark Triad and dispositions towards empathy are related negatively (Wai & Tiliopoulos, 2012).
Conceivably, a somewhat novel finding is the non-significant relationship between the Dark Triad and ability-based empathy in the present study, confirming the second hypothesis. Among the sparse studies on the subject, Wai and Tiliopoulos (2012) similarly found no connection. The measurement of MET indicates that dark personalities are more or less normally distributed in relation to the ability of reading emotions in faces. In our community sample, the popular notion of an intelligent, cunning psychopath or narcissist being a master-mind in capacity of reading people was not found in evidence. Similarly, the opposite notion of an impulsive thug incapable of interpreting people's faces cannot be supported. Interestingly, there seems to be almost no relationship between ability-based empathy (MET) and the subscales of trait-based empathy (IRI), as seen in Table 1. The present study seems to support the notion that ability and traits are very different empathy constructs.
Moreover, cognitive ability clearly governed ability-based empathy (Fig. 1). Individuals’ cognitive ability (ICAR) overlapped with the ability to read people's emotions through facial expressions (MET). This may not be all too surprising since general intelligence (aka the Gfactor) is known to permeate most psychological domains related to mental performance (Nisbett et al., 2012). Perhaps somewhat unexpected, a small positive relationship between the Dark Triad and cognitive ability was also found in the present study. If anything, this should according to literature be close to zero or even negative. Apart from having been a spurious result (n.b. this was only marginally significant, p = .04), one explanation is that being smart and slightly antagonistic may very well have been one of the characteristics of someone choosing to partake in a study on dark personalities, slightly increasing this correlation. Based on the tested model, it seems clear that the higher the cognitive ability, the higher the ability to read other's emotions, but also that this is likely unrelated to dark personalities.
Highlights
• We tested if the Dark Triad was best described by ability- or trait-empathy.
• Dark Triad had no relationship with ability-empathy.
• Dark Triad had a strong negative relationship with trait-empathy.
• Cognitive ability explained ability-empathy.
• Dark personalities seem cognizant, but not inclined to empathize.
Abstract: Empathy is fundamental to social cognition and societal values. Empathy is theorized as having both the ability as well as the disposition to imagine the content of other people's minds. We tested whether the notorious low empathy in dark personalities (Machiavellianism, psychopathy, and narcissism; the Dark Triad) is best characterized by a lack of capacity (ability) or lack of disposition (trait). Data was collected for 278 international participants through an anonymous online survey shared on the online platform LinkedIn, consisting of trait-based Dark Triad personality (SD3) and empathy (IRI), and cognitive ability (ICAR16) and ability-based empathy (MET). Dark personality traits had no relationship with ability-based empathy, but strongly so with trait-based empathy (β = -0.47). Instead, cognitive ability explained ability-based empathy (β = 0.31). The finding is that dark personalities in a community sample is normally cognizant to empathize but has a low disposition to do so. This finding may help shed further light on how personality is interlinked with ability.
Keywords: PersonalityDark triadEmpathyCognitive ability
From the introduction: Empathy is a core feature of human beings in social interaction (Myyrya, Juujärvi, & Pesso, 2010). No matter where in the world we live, individuals in any given community are expected to be able to cognizant and sensitive to other people's minds. Individuals who violate such values are often looked down upon in society (Persson & Kajonius, 2016). In personality psychology, there has been an increase of interest in so-called dark personality traits (Moshagen, Hilbig, & Zettler, 2018), which are characterized by violating social values (Kajonius, Persson, & Jonason, 2015). The idea behind the most used Dark Triad personality model (DT; Paulhus & Williams, 2002) is to capture the multidimensionality of complex traits leading up to this, and that these can be described by subclinical Machiavellianism (tendency to manipulate), psychopathy (callousness), and narcissism (grandiosity) (Jonason & Kroll, 2015). It is still unclear whether individuals scoring high on these dark personality traits are mostly lacking the capacity (ability) or mostly lacking the disposition (trait) to feel what others feel (see Keysers & Gazzola, 2014). The purpose of the present study is to explore empathy and to test the idea that it is not inability but more a lack of disposition that drives dark personalities’ low empathy.
4. Discussion
The present study aimed at investigating whether persons scoring high on the Dark Triad (Machiavellianism, psychopathy, and narcissism) mostly relate to the lack of ability or the lack of disposition to empathize. The results showed that it was more a lack of empathic disposition than inability that characterizes dark personalities. First, the Dark Triad had a very strong relationship with dispositional trait-based empathy. Second, the Dark Triad had a weak (almost non-existent) relationship with ability-based empathy. Third, cognitive ability explained most of ability-based empathy.
The first hypothesis of the present study were largely in line with previous research, which mainly has shown a consistent negative relationship between dark personality and trait-based empathy (BaronCohen, Wheelwright, Hill, Raste, & Plumb, 2001; Jonason, Lyons, Bethell, & Ross, 2013; Pajevic et al., 2018). The relationship between higher Dark Triad (SD3) and lower trait-based empathy (IRI) was large (Fig. 1). This effect size is trending towards high convergence, even more so if controlled for reliability and subscale variance – According to updated guidelines in psychology research, correlations between r = 0.00–0.09 should be interpreted as trivial to non-existent, r = 0.10–0.19 weak, r = 0.20–0.29 medium, and above r > 0.30 strong (Gignac & Szodorai, 2016). Interestingly, lack of empathy as measured by IRI entails not only lack empathic concern, but also no imagination for others’ minds (FT), not being cognizant in perspective taking (PT), and no distress for others’ welfare (PD). This result confirms the general notion that the Dark Triad and dispositions towards empathy are related negatively (Wai & Tiliopoulos, 2012).
Conceivably, a somewhat novel finding is the non-significant relationship between the Dark Triad and ability-based empathy in the present study, confirming the second hypothesis. Among the sparse studies on the subject, Wai and Tiliopoulos (2012) similarly found no connection. The measurement of MET indicates that dark personalities are more or less normally distributed in relation to the ability of reading emotions in faces. In our community sample, the popular notion of an intelligent, cunning psychopath or narcissist being a master-mind in capacity of reading people was not found in evidence. Similarly, the opposite notion of an impulsive thug incapable of interpreting people's faces cannot be supported. Interestingly, there seems to be almost no relationship between ability-based empathy (MET) and the subscales of trait-based empathy (IRI), as seen in Table 1. The present study seems to support the notion that ability and traits are very different empathy constructs.
Moreover, cognitive ability clearly governed ability-based empathy (Fig. 1). Individuals’ cognitive ability (ICAR) overlapped with the ability to read people's emotions through facial expressions (MET). This may not be all too surprising since general intelligence (aka the Gfactor) is known to permeate most psychological domains related to mental performance (Nisbett et al., 2012). Perhaps somewhat unexpected, a small positive relationship between the Dark Triad and cognitive ability was also found in the present study. If anything, this should according to literature be close to zero or even negative. Apart from having been a spurious result (n.b. this was only marginally significant, p = .04), one explanation is that being smart and slightly antagonistic may very well have been one of the characteristics of someone choosing to partake in a study on dark personalities, slightly increasing this correlation. Based on the tested model, it seems clear that the higher the cognitive ability, the higher the ability to read other's emotions, but also that this is likely unrelated to dark personalities.
Friday, November 29, 2019
Hard Problems in Cryptocurrency: Cryptographic (expected to be solvable with purely mathematical techniques), consensus theory (improvements to proof of work and proof of stake), and economic
Hard Problems in Cryptocurrency: Five Years Later. Vitalik Buterin. No 22 2019. https://vitalik.ca/general/2019/11/22/progress.html
[Check original post for lots of links]
Special thanks to Justin Drake and Jinglan Wang for feedback
In 2014, I made a post and a presentation with a list of hard problems in math, computer science and economics that I thought were important for the cryptocurrency space (as I then called it) to be able to reach maturity. In the last five years, much has changed. But exactly how much progress on what we thought then was important has been achieved? Where have we succeeded, where have we failed, and where have we changed our minds about what is important? In this post, I'll go through the 16 problems from 2014 one by one, and see just where we are today on each one. At the end, I’ll include my new picks for hard problems of 2019.
The problems are broken down into three categories: (i) cryptographic, and hence expected to be solvable with purely mathematical techniques if they are to be solvable at all, (ii) consensus theory, largely improvements to proof of work and proof of stake, and (iii) economic, and hence having to do
with creating structures involving incentives given to different participants, and often involving the application layer more than the protocol layer. We see significant progress in all categories, though some more than others.
Cryptographic problems
1 Blockchain Scalability
One of the largest problems facing the cryptocurrency space today is the issue of scalability ... The main concern with [oversized blockchains] is trust: if there are only a few entities capable of running full nodes, then those entities can conspire and agree to give themselves a large number of additional bitcoins, and there would be no way for other users to see for themselves that a block is invalid without processing an entire block themselves.
Problem: create a blockchain design that maintains Bitcoin-like security guarantees, but where the maximum size of the most powerful node that needs to exist for the network to keep functioning is substantially sublinear in the number of transactions.
Status: Great theoretical progress, pending more real-world evaluation.
Scalability is one technical problem that we have had a huge amount of progress on theoretically. Five years ago, almost no one was thinking about sharding; now, sharding designs are commonplace. Aside from ethereum 2.0, we have OmniLedger, LazyLedger, Zilliqa and research papers seemingly coming out every month. In my own view, further progress at this point is incremental. Fundamentally, we already have a number of techniques that allow groups of validators to securely come to consensus on much more data than an individual validator can process, as well as techniques allow clients to indirectly verify the full validity and availability of blocks even under 51% attack conditions.
These are probably the most important technologies:
. Random sampling, allowing a small randomly selected committee to statistically stand in for the full validator set: https://github.com/ethereum/wiki/wiki/Sharding-FAQ#how-can-we-solve-the-single-shard-takeover-attack-in-an-uncoordinated-majority-model
. Fraud proofs, allowing individual nodes that learn of an error to broadcast its presence to everyone else: https://bitcoin.stackexchange.com/questions/49647/what-is-a-fraud-proof
. Proofs of custody, allowing validators to probabilistically prove that they individually downloaded and verified some piece of data: https://ethresear.ch/t/1-bit-aggregation-friendly-custody-bonds/2236
. Data availability proofs, allowing clients to detect when the bodies of blocks that they have headers for are unavailable: https://arxiv.org/abs/1809.09044. See also the newer coded Merkle trees proposal.
There are also other smaller developments like Cross-shard communication via receipts as well as "constant-factor" enhancements such as BLS signature aggregation.
That said, fully sharded blockchains have still not been seen in live operation (the partially sharded Zilliqa has recently started running). On the theoretical side, there are mainly disputes about details remaining, along with challenges having to do with stability of sharded networking, developer experience and mitigating risks of centralization; fundamental technical possibility no longer seems in doubt. But the challenges that do remain are challenges that cannot be solved by just thinking about them; only developing the system and seeing ethereum 2.0 or some similar chain running live will suffice.
2 Timestamping
Problem: create a distributed incentive-compatible system, whether it is an overlay on top of a blockchain or its own blockchain, which maintains the current time to high accuracy. All legitimate users have clocks in a normal distribution around some "real" time with standard deviation 20 seconds ... no two nodes are more than 20 seconds apart The solution is allowed to rely on an existing concept of "N nodes"; this would in practice be enforced with proof-of-stake or non-sybil tokens (see #9). The system should continuously provide a time which is within 120s (or less if possible) of the internal clock of >99% of honestly participating nodes. External systems may end up relying on this system; hence, it should remain secure against attackers controlling < 25% of nodes regardless of incentives.
Status: Some progress.
Ethereum has actually survived just fine with a 13-second block time and no particularly advanced timestamping technology; it uses a simple technique where a client does not accept a block whose stated timestamp is earlier than the client's local time. That said, this has not been tested under serious attacks. The recent network-adjusted timestamps proposal tries to improve on the status quo by allowing the client to determine the consensus on the time in the case where the client does not locally know the current time to high accuracy; this has not yet been tested. But in general, timestamping is not currently at the foreground of perceived research challenges; perhaps this will change once more proof of stake chains (including Ethereum 2.0 but also others) come online as real live systems and we see what the issues are.
3 Arbitrary Proof of Computation
Problem: create programs POC_PROVE(P,I) -> (O,Q) and POC_VERIFY(P,O,Q) -> { 0, 1 } such that POC_PROVE runs program P on input I and returns the program output O and a proof-of-computation Q and POC_VERIFY takes P, O and Q and outputs whether or not Q and O were legitimately produced by the POC_PROVE algorithm using P.
Status: Great theoretical and practical progress.
This is basically saying, build a SNARK (or STARK, or SHARK, or...). And we've done it! SNARKs are now increasingly well understood, and are even already being used in multiple blockchains today (including tornado.cash on Ethereum). And SNARKs are extremely useful, both as a privacy technology (see Zcash and tornado.cash) and as a scalability technology (see ZK Rollup, STARKDEX and STARKing erasure coded data roots).
There are still challenges with efficiency; making arithmetization-friendly hash functions (see here and here for bounties for breaking proposed candidates) is a big one, and efficiently proving random memory accesses is another. Furthermore, there's the unsolved question of whether the O(n * log(n)) blowup in prover time is a fundamental limitation or if there is some way to make a succinct proof with only linear overhead as in bulletproofs (which unfortunately take linear time to verify). There are also ever-present risks that the existing schemes have bugs. In general, the problems are in the details rather than the fundamentals.
4 Code Obfuscation
The holy grail is to create an obfuscator O, such that given any program P the obfuscator can produce a second program O(P) = Q such that P and Q return the same output if given the same input and, importantly, Q reveals no information whatsoever about the internals of P. One can hide inside of Q a password, a secret encryption key, or one can simply use Q to hide the proprietary workings of the algorithm itself.
Status: Slow progress.
In plain English, the problem is saying that we want to come up with a way to "encrypt" a program so that the encrypted program would still give the same outputs for the same inputs, but the "internals" of the program would be hidden. An example use case for obfuscation is a program containing a private key where the program only allows the private key to sign certain messages.
A solution to code obfuscation would be very useful to blockchain protocols. The use cases are subtle, because one must deal with the possibility that an on-chain obfuscated program will be copied and run in an environment different from the chain itself, but there are many possibilities. One that personally interests me is the ability to remove the centralized operator from collusion-resistance gadgets by replacing the operator with an obfuscated program that contains some proof of work, making it very expensive to run more than once with different inputs as part of an attempt to determine individual participants' actions.
Unfortunately this continues to be a hard problem. There is continuing ongoing work in attacking the problem, one side making constructions (eg. this) that try to reduce the number of assumptions on mathematical objects that we do not know practically exist (eg. general cryptographic multilinear maps) and another side trying to make practical implementations of the desired mathematical objects. However, all of these paths are still quite far from creating something viable and known to be secure. See https://eprint.iacr.org/2019/463.pdf for a more general overview to the problem.
5 Hash-Based Cryptography
Problem: create a signature algorithm relying on no security assumption but the random oracle property of hashes that maintains 160 bits of security against classical computers (ie. 80 vs. quantum due to Grover's algorithm) with optimal size and other properties.
Status: Some progress.
There have been two strands of progress on this since 2014. SPHINCS, a "stateless" (meaning, using it multiple times does not require remembering information like a nonce) signature scheme, was released soon after this "hard problems" list was published, and provides a purely hash-based signature scheme of size around 41 kB. Additionally, STARKs have been developed, and one can create signatures of similar size based on them. The fact that not just signatures, but also general-purpose zero knowledge proofs, are possible with just hashes was definitely something I did not expect five years ago; I am very happy that this is the case. That said, size continues to be an issue, and ongoing progress (eg. see the very recent DEEP FRI) is continuing to reduce the size of proofs, though it looks like further progress will be incremental.
The main not-yet-solved problem with hash-based cryptography is aggregate signatures, similar to what BLS aggregation makes possible. It's known that we can just make a STARK over many Lamport signatures, but this is inefficient; a more efficient scheme would be welcome. (In case you're wondering if hash-based public key encryption is possible, the answer is, no, you can't do anything with more than a quadratic attack cost)
Consensus theory problems
6 ASIC-Resistant Proof of Work
One approach at solving the problem is creating a proof-of-work algorithm based on a type of computation that is very difficult to specialize ... For a more in-depth discussion on ASIC-resistant hardware, see https://blog.ethereum.org/2014/06/19/mining/.
Status: Solved as far as we can.
About six months after the "hard problems" list was posted, Ethereum settled on its ASIC-resistant proof of work algorithm: Ethash. Ethash is known as a memory-hard algorithm. The theory is that random-access memory in regular computers is well-optimized already and hence difficult to improve on for specialized applications. Ethash aims to achieve ASIC resistance by making memory access the dominant part of running the PoW computation. Ethash was not the first memory-hard algorithm, but it did add one innovation: it uses pseudorandom lookups over a two-level DAG, allowing for two ways of evaluating the function. First, one could compute it quickly if one has the entire (~2 GB) DAG; this is the memory-hard "fast path". Second, one can compute it much more slowly (still fast enough to check a single provided solution quickly) if one only has the top level of the DAG; this is used for block verification.
Ethash has proven remarkably successful at ASIC resistance; after three years and billions of dollars of block rewards, ASICs do exist but are at best 2-5 times more power and cost-efficient than GPUs. ProgPoW has been proposed as an alternative, but there is a growing consensus that ASIC-resistant algorithms will inevitably have a limited lifespan, and that ASIC resistance has downsides because it makes 51% attacks cheaper (eg. see the 51% attack on Ethereum Classic).
I believe that PoW algorithms that provide a medium level of ASIC resistance can be created, but such resistance is limited-term and both ASIC and non-ASIC PoW have disadvantages; in the long term the better choice for blockchain consensus is proof of stake.
7 Useful Proof of Work
[M]aking the proof of work function something which is simultaneously useful; a common candidate is something like Folding@home, an existing program where users can download software onto their computers to simulate protein folding and provide researchers with a large supply of data to help them cure diseases.
Status: Probably not feasible, with one exception.
The challenge with useful proof of work is that a proof of work algorithm requires many properties:
. Hard to compute
. Easy to verify
. Does not depend on large amounts of external data
. Can be efficiently computed in small "bite-sized" chunks
Unfortunately, there are not many computations that are useful that preserve all of these properties, and most computations that do have all of those properties and are "useful" are only "useful" for far too short a time to build a cryptocurrency around them.
However, there is one possible exception: zero-knowledge-proof generation. Zero knowledge proofs of aspects of blockchain validity (eg. data availability roots for a simple example) are difficult to compute, and easy to verify. Furthermore, they are durably difficult to compute; if proofs of "highly structured" computation become too easy, one can simply switch to verifying a blockchain's entire state transition, which becomes extremely expensive due to the need to model the virtual machine and random memory accesses.
Zero-knowledge proofs of blockchain validity provide great value to users of the blockchain, as they can substitute the need to verify the chain directly; Coda is doing this already, albeit with a simplified blockchain design that is heavily optimized for provability. Such proofs can significantly assist in improving the blockchain's safety and scalability. That said, the total amount of computation that realistically needs to be done is still much less than the amount that's currently done by proof of work miners, so this would at best be an add-on for proof of stake blockchains, not a full-on consensus algorithm.
8 Proof of Stake
Another approach to solving the mining centralization problem is to abolish mining entirely, and move to some other mechanism for counting the weight of each node in the consensus. The most popular alternative under discussion to date is "proof of stake" - that is to say, instead of treating the consensus model as "one unit of CPU power, one vote" it becomes "one currency unit, one vote".
Status: Great theoretical progress, pending more real-world evaluation.
Near the end of 2014, it became clear to the proof of stake community that some form of "weak subjectivity" is unavoidable. To maintain economic security, nodes need to obtain a recent checkpoint extra-protocol when they sync for the first time, and again if they go offline for more than a few months. This was a difficult pill to swallow; many PoW advocates still cling to PoW precisely because in a PoW chain the "head" of the chain can be discovered with the only data coming from a trusted source being the blockchain client software itself. PoS advocates, however, were willing to swallow the pill, seeing the added trust requirements as not being large. From there the path to proof of stake through long-duration security deposits became clear.
Most interesting consensus algorithms today are fundamentally similar to PBFT, but replace the fixed set of validators with a dynamic list that anyone can join by sending tokens into a system-level smart contract with time-locked withdrawals (eg. a withdrawal might in some cases take up to 4 months to complete). In many cases (including ethereum 2.0), these algorithms achieve "economic finality" by penalizing validators that are caught performing actions that violate the protocol in certain ways (see here for a philosophical view on what proof of stake accomplishes).
As of today, we have (among many other algorithms):
Casper FFG: https://arxiv.org/abs/1710.09437
Tendermint: https://tendermint.com/docs/spec/consensus/consensus.html
HotStuff: https://arxiv.org/abs/1803.05069
Casper CBC: https://vitalik.ca/general/2018/12/05/cbc_casper.html
There continues to be ongoing refinement (eg. here and here) . Eth2 phase 0, the chain that will implement FFG, is currently under implementation and enormous progress has been made. Additionally, Tendermint has been running, in the form of the Cosmos chain for several months. Remaining arguments about proof of stake, in my view, have to do with optimizing the economic incentives, and further formalizing the strategy for responding to 51% attacks. Additionally, the Casper CBC spec could still use concrete efficiency improvements.
9 Proof of Storage
A third approach to the problem is to use a scarce computational resource other than computational power or currency. In this regard, the two main alternatives that have been proposed are storage and bandwidth. There is no way in principle to provide an after-the-fact cryptographic proof that bandwidth was given or used, so proof of bandwidth should most accurately be considered a subset of social proof, discussed in later problems, but proof of storage is something that certainly can be done computationally. An advantage of proof-of-storage is that it is completely ASIC-resistant; the kind of storage that we have in hard drives is already close to optimal.
Status: A lot of theoretical progress, though still a lot to go, as well as more real-world evaluation.
There are a number of blockchains planning to use proof of storage protocols, including Chia and Filecoin. That said, these algorithms have not been tested in the wild. My own main concern is centralization: will these algorithms actually be dominated by smaller users using spare storage capacity, or will they be dominated by large mining farms?
Economics
10 Stable-value cryptoassets
One of the main problems with Bitcoin is the issue of price volatility ... Problem: construct a cryptographic asset with a stable price.
Status: Some progress.
MakerDAO is now live, and has been holding stable for nearly two years. It has survived a 93% drop in the value of its underlying collateral asset (ETH), and there is now more than $100 million in DAI issued. It has become a mainstay of the Ethereum ecosystem, and many Ethereum projects have or are integrating with it. Other synthetic token projects, such as UMA, are rapidly gaining steam as well.
However, while the MakerDAO system has survived tough economic conditions in 2019, the conditions were by no means the toughest that could happen. In the past, Bitcoin has fallen by 75% over the course of two days; the same may happen to ether or any other collateral asset some day. Attacks on the underlying blockchain are an even larger untested risk, especially if compounded by price decreases at the same time. Another major challenge, and arguably the larger one, is that the stability of MakerDAO-like systems is dependent on some underlying oracle scheme. Different attempts at oracle systems do exist (see #16), but the jury is still out on how well they can hold up under large amounts of economic stress. So far, the collateral controlled by MakerDAO has been lower than the value of the MKR token; if this relationship reverses MKR holders may have a collective incentive to try to "loot" the MakerDAO system. There are ways to try to protect against such attacks, but they have not been tested in real life.
11 Decentralized Public Goods Incentivization
One of the challenges in economic systems in general is the problem of "public goods". For example, suppose that there is a scientific research project which will cost $1 million to complete, and it is known that if it is completed the resulting research will save one million people $5 each. In total, the social benefit is clear ... [but] from the point of view of each individual person contributing does not make sense ... So far, most problems to public goods have involved centralization
Additional Assumptions And Requirements: A fully trustworthy oracle exists for determining whether or not a certain public good task has been completed (in reality this is false, but this is the domain of another problem)
Status: Some progress.
The problem of funding public goods is generally understood to be split into two problems: the funding problem (where to get funding for public goods from) and the preference aggregation problem (how to determine what is a genuine public good, rather than some single individual's pet project, in the first place). This problem focuses specifically on the former, assuming the latter is solved (see the "decentralized contribution metrics" section below for work on that problem).
In general, there haven't been large new breakthroughs here. There's two major categories of solutions. First, we can try to elicit individual contributions, giving people social rewards for doing so. My own proposal for charity through marginal price discrimination is one example of this; another is the anti-malaria donation badges on Peepeth. Second, we can collect funds from applications that have network effects. Within blockchain land there are several options for doing this:
. Issuing coins
. Taking a portion of transaction fees at protocol level (eg. through EIP 1559)
. Taking a portion of transaction fees from some layer-2 application (eg. Uniswap, or some scaling solution, or even state rent in an execution environment in ethereum 2.0)
. Taking a portion of other kinds of fees (eg. ENS registration)
Outside of blockchain land, this is just the age-old question of how to collect taxes if you're a government, and charge fees if you're a business or other organization.
12 Reputation systems
Problem: design a formalized reputation system, including a score rep(A,B) -> V where V is the reputation of B from the point of view of A, a mechanism for determining the probability that one party can be trusted by another, and a mechanism for updating the reputation given a record of a particular open or finalized interaction.
Status: Slow progress.
There hasn't really been much work on reputation systems since 2014. Perhaps the best is the use of token curated registries to create curated lists of trustable entities/objects; the Kleros ERC20 TCR (yes, that's a token-curated registry of legitimate ERC20 tokens) is one example, and there is even an alternative interface to Uniswap (http://uniswap.ninja) that uses it as the backend to get the list of tokens and ticker symbols and logos from. Reputation systems of the subjective variety have not really been tried, perhaps because there is just not enough information about the "social graph" of people's connections to each other that has already been published to chain in some form. If such information starts to exist for other reasons, then subjective reputation systems may become more popular.
13 Proof of excellence
One interesting, and largely unexplored, solution to the problem of [token] distribution specifically (there are reasons why it cannot be so easily used for mining) is using tasks that are socially useful but require original human-driven creative effort and talent. For example, one can come up with a "proof of proof" currency that rewards players for coming up with mathematical proofs of certain theorems
Status: No progress, problem is largely forgotten.
The main alternative approach to token distribution that has instead become popular is airdrops; typically, tokens are distributed at launch either proportionately to existing holdings of some other token, or based on some other metric (eg. as in the Handshake airdrop). Verifying human creativity directly has not really been attempted, and with recent progress on AI the problem of creating a task that only humans can do but computers can verify may well be too difficult.
15 [sic]. Anti-Sybil systems
A problem that is somewhat related to the issue of a reputation system is the challenge of creating a "unique identity system" - a system for generating tokens that prove that an identity is not part of a Sybil attack ... However, we would like to have a system that has nicer and more egalitarian features than "one-dollar-one-vote"; arguably, one-person-one-vote would be ideal.
Status: Some progress.
There have been quite a few attempts at solving the unique-human problem. Attempts that come to mind include (incomplete list!):
. HumanityDAO: https://www.humanitydao.org/
. Pseudonym parties: https://bford.info/pub/net/sybil.pdf
. POAP ("proof of attendance protocol"): https://www.poap.xyz/
. BrightID: https://www.brightid.org/
With the growing interest in techniques like quadratic voting and quadratic funding, the need for some kind of human-based anti-sybil system continues to grow. Hopefully, ongoing development of these techniques and new ones can come to meet it.
14 [sic]. Decentralized contribution metrics
Incentivizing the production of public goods is, unfortunately, not the only problem that centralization solves. The other problem is determining, first, which public goods are worth producing in the first place and, second, determining to what extent a particular effort actually accomplished the production of the public good. This challenge deals with the latter issue.
Status: Some progress, some change in focus.
More recent work on determining value of public-good contributions does not try to separate determining tasks and determining quality of completion; the reason is that in practice the two are difficult to separate. Work done by specific teams tends to be non-fungible and subjective enough that the most reasonable approach is to look at relevance of task and quality of performance as a single package, and use the same technique to evaluate both.
Fortunately, there has been great progress on this, particularly with the discovery of quadratic funding. Quadratic funding is a mechanism where individuals can make donations to projects, and then based on the number of people who donated and how much they donated, a formula is used to calculate how much they would have donated if they were perfectly coordinated with each other (ie. took each other's interests into account and did not fall prey to the tragedy of the commons). The difference between amount would-have-donated and amount actually donated for any given project is given to that project as a subsidy from some central pool (see #11 for where the central pool funding could come from). Note that this mechanism focuses on satisfying the values of some community, not on satisfying some given goal regardless of whether or not anyone cares about it. Because of the complexity of values problem, this approach is likely to be much more robust to unknown unknowns.
Quadratic funding has even been tried in real life with considerable success in the recent gitcoin quadratic funding round. There has also been some incremental progress on improving quadratic funding and similar mechanisms; particularly, pairwise-bounded quadratic funding to mitigate collusion. There has also been work on specification and implementation of bribe-resistant voting technology, preventing users from proving to third parties who they voted for; this prevents many kinds of collusion and bribe attacks.
16 Decentralized success metrics
Problem: come up with and implement a decentralized method for measuring numerical real-world variables ... the system should be able to measure anything that humans can currently reach a rough consensus on (eg. price of an asset, temperature, global CO2 concentration)
Status: Some progress.
This is now generally just called "the oracle problem". The largest known instance of a decentralized oracle running is Augur, which has processed outcomes for millions of dollars of bets. Token curated registries such as the Kleros TCR for tokens are another example. However, these systems still have not seen a real-world test of the forking mechanism (search for "subjectivocracy" here) either due to a highly controversial question or due to an attempted 51% attack. There is also research on the oracle problem happening outside of the blockchain space in the form of the "peer prediction" literature; see here for a very recent advancement in the space.
Another looming challenge is that people want to rely on these systems to guide transfers of quantities of assets larger than the economic value of the system's native token. In these conditions, token holders in theory have the incentive to collude to give wrong answers to steal the funds. In such a case, the system would fork and the original system token would likely become valueless, but the original system token holders would still get away with the returns from whatever asset transfer they misdirected. Stablecoins (see #10) are a particularly egregious case of this. One approach to solving this would be a system that assumes that altruistically honest data providers do exist, and creating a mechanism to identify them, and only allowing them to churn slowly so that if malicious ones start getting voted in the users of systems that rely on the oracle can first complete an orderly exit. In any case, more development of oracle tech is very much an important problem.
New problems
If I were to write the hard problems list again in 2019, some would be a continuation of the above problems, but there would be significant changes in emphasis, as well as significant new problems. Here are a few picks:
. Cryptographic obfuscation: same as #4 above
. Ongoing work on post-quantum cryptography: both hash-based as well as based on post-quantum-secure "structured" mathematical objects, eg. elliptic curve isogenies, lattices...
. Anti-collusion infrastructure: ongoing work and refinement of https://ethresear.ch/t/minimal-anti-collusion-infrastructure/5413, including adding privacy against the operator, adding multi-party computation in a maximally practical way, etc.
. Oracles: same as #16 above, but removing the emphasis on "success metrics" and focusing on the general "get real-world data" problem
. Unique-human identities (or, more realistically, semi-unique-human identities): same as what was written as #15 above, but with an emphasis on a less "absolute" solution: it should be much harder to get two identities than one, but making it impossible to get multiple identities is both impossible and potentially harmful even if we do succeed
. Homomorphic encryption and multi-party computation: ongoing improvements are still required for practicality
. Decentralized governance mechanisms: DAOs are cool, but current DAOs are still very primitive; we can do better
Fully formalizing responses to PoS 51% attacks: ongoing work and refinement of https://ethresear.ch/t/responding-to-51-attacks-in-casper-ffg/6363
. More sources of public goods funding: the ideal is to charge for congestible resources inside of systems that have network effects (eg. transaction fees), but doing so in decentralized systems requires public legitimacy; hence this is a social problem along with the technical one of finding possible sources
. Reputation systems: same as #12 above
In general, base-layer problems are slowly but surely decreasing, but application-layer problems are only just getting started.
[Check original post for lots of links]
Special thanks to Justin Drake and Jinglan Wang for feedback
In 2014, I made a post and a presentation with a list of hard problems in math, computer science and economics that I thought were important for the cryptocurrency space (as I then called it) to be able to reach maturity. In the last five years, much has changed. But exactly how much progress on what we thought then was important has been achieved? Where have we succeeded, where have we failed, and where have we changed our minds about what is important? In this post, I'll go through the 16 problems from 2014 one by one, and see just where we are today on each one. At the end, I’ll include my new picks for hard problems of 2019.
The problems are broken down into three categories: (i) cryptographic, and hence expected to be solvable with purely mathematical techniques if they are to be solvable at all, (ii) consensus theory, largely improvements to proof of work and proof of stake, and (iii) economic, and hence having to do
with creating structures involving incentives given to different participants, and often involving the application layer more than the protocol layer. We see significant progress in all categories, though some more than others.
Cryptographic problems
1 Blockchain Scalability
One of the largest problems facing the cryptocurrency space today is the issue of scalability ... The main concern with [oversized blockchains] is trust: if there are only a few entities capable of running full nodes, then those entities can conspire and agree to give themselves a large number of additional bitcoins, and there would be no way for other users to see for themselves that a block is invalid without processing an entire block themselves.
Problem: create a blockchain design that maintains Bitcoin-like security guarantees, but where the maximum size of the most powerful node that needs to exist for the network to keep functioning is substantially sublinear in the number of transactions.
Status: Great theoretical progress, pending more real-world evaluation.
Scalability is one technical problem that we have had a huge amount of progress on theoretically. Five years ago, almost no one was thinking about sharding; now, sharding designs are commonplace. Aside from ethereum 2.0, we have OmniLedger, LazyLedger, Zilliqa and research papers seemingly coming out every month. In my own view, further progress at this point is incremental. Fundamentally, we already have a number of techniques that allow groups of validators to securely come to consensus on much more data than an individual validator can process, as well as techniques allow clients to indirectly verify the full validity and availability of blocks even under 51% attack conditions.
These are probably the most important technologies:
. Random sampling, allowing a small randomly selected committee to statistically stand in for the full validator set: https://github.com/ethereum/wiki/wiki/Sharding-FAQ#how-can-we-solve-the-single-shard-takeover-attack-in-an-uncoordinated-majority-model
. Fraud proofs, allowing individual nodes that learn of an error to broadcast its presence to everyone else: https://bitcoin.stackexchange.com/questions/49647/what-is-a-fraud-proof
. Proofs of custody, allowing validators to probabilistically prove that they individually downloaded and verified some piece of data: https://ethresear.ch/t/1-bit-aggregation-friendly-custody-bonds/2236
. Data availability proofs, allowing clients to detect when the bodies of blocks that they have headers for are unavailable: https://arxiv.org/abs/1809.09044. See also the newer coded Merkle trees proposal.
There are also other smaller developments like Cross-shard communication via receipts as well as "constant-factor" enhancements such as BLS signature aggregation.
That said, fully sharded blockchains have still not been seen in live operation (the partially sharded Zilliqa has recently started running). On the theoretical side, there are mainly disputes about details remaining, along with challenges having to do with stability of sharded networking, developer experience and mitigating risks of centralization; fundamental technical possibility no longer seems in doubt. But the challenges that do remain are challenges that cannot be solved by just thinking about them; only developing the system and seeing ethereum 2.0 or some similar chain running live will suffice.
2 Timestamping
Problem: create a distributed incentive-compatible system, whether it is an overlay on top of a blockchain or its own blockchain, which maintains the current time to high accuracy. All legitimate users have clocks in a normal distribution around some "real" time with standard deviation 20 seconds ... no two nodes are more than 20 seconds apart The solution is allowed to rely on an existing concept of "N nodes"; this would in practice be enforced with proof-of-stake or non-sybil tokens (see #9). The system should continuously provide a time which is within 120s (or less if possible) of the internal clock of >99% of honestly participating nodes. External systems may end up relying on this system; hence, it should remain secure against attackers controlling < 25% of nodes regardless of incentives.
Status: Some progress.
Ethereum has actually survived just fine with a 13-second block time and no particularly advanced timestamping technology; it uses a simple technique where a client does not accept a block whose stated timestamp is earlier than the client's local time. That said, this has not been tested under serious attacks. The recent network-adjusted timestamps proposal tries to improve on the status quo by allowing the client to determine the consensus on the time in the case where the client does not locally know the current time to high accuracy; this has not yet been tested. But in general, timestamping is not currently at the foreground of perceived research challenges; perhaps this will change once more proof of stake chains (including Ethereum 2.0 but also others) come online as real live systems and we see what the issues are.
3 Arbitrary Proof of Computation
Problem: create programs POC_PROVE(P,I) -> (O,Q) and POC_VERIFY(P,O,Q) -> { 0, 1 } such that POC_PROVE runs program P on input I and returns the program output O and a proof-of-computation Q and POC_VERIFY takes P, O and Q and outputs whether or not Q and O were legitimately produced by the POC_PROVE algorithm using P.
Status: Great theoretical and practical progress.
This is basically saying, build a SNARK (or STARK, or SHARK, or...). And we've done it! SNARKs are now increasingly well understood, and are even already being used in multiple blockchains today (including tornado.cash on Ethereum). And SNARKs are extremely useful, both as a privacy technology (see Zcash and tornado.cash) and as a scalability technology (see ZK Rollup, STARKDEX and STARKing erasure coded data roots).
There are still challenges with efficiency; making arithmetization-friendly hash functions (see here and here for bounties for breaking proposed candidates) is a big one, and efficiently proving random memory accesses is another. Furthermore, there's the unsolved question of whether the O(n * log(n)) blowup in prover time is a fundamental limitation or if there is some way to make a succinct proof with only linear overhead as in bulletproofs (which unfortunately take linear time to verify). There are also ever-present risks that the existing schemes have bugs. In general, the problems are in the details rather than the fundamentals.
4 Code Obfuscation
The holy grail is to create an obfuscator O, such that given any program P the obfuscator can produce a second program O(P) = Q such that P and Q return the same output if given the same input and, importantly, Q reveals no information whatsoever about the internals of P. One can hide inside of Q a password, a secret encryption key, or one can simply use Q to hide the proprietary workings of the algorithm itself.
Status: Slow progress.
In plain English, the problem is saying that we want to come up with a way to "encrypt" a program so that the encrypted program would still give the same outputs for the same inputs, but the "internals" of the program would be hidden. An example use case for obfuscation is a program containing a private key where the program only allows the private key to sign certain messages.
A solution to code obfuscation would be very useful to blockchain protocols. The use cases are subtle, because one must deal with the possibility that an on-chain obfuscated program will be copied and run in an environment different from the chain itself, but there are many possibilities. One that personally interests me is the ability to remove the centralized operator from collusion-resistance gadgets by replacing the operator with an obfuscated program that contains some proof of work, making it very expensive to run more than once with different inputs as part of an attempt to determine individual participants' actions.
Unfortunately this continues to be a hard problem. There is continuing ongoing work in attacking the problem, one side making constructions (eg. this) that try to reduce the number of assumptions on mathematical objects that we do not know practically exist (eg. general cryptographic multilinear maps) and another side trying to make practical implementations of the desired mathematical objects. However, all of these paths are still quite far from creating something viable and known to be secure. See https://eprint.iacr.org/2019/463.pdf for a more general overview to the problem.
5 Hash-Based Cryptography
Problem: create a signature algorithm relying on no security assumption but the random oracle property of hashes that maintains 160 bits of security against classical computers (ie. 80 vs. quantum due to Grover's algorithm) with optimal size and other properties.
Status: Some progress.
There have been two strands of progress on this since 2014. SPHINCS, a "stateless" (meaning, using it multiple times does not require remembering information like a nonce) signature scheme, was released soon after this "hard problems" list was published, and provides a purely hash-based signature scheme of size around 41 kB. Additionally, STARKs have been developed, and one can create signatures of similar size based on them. The fact that not just signatures, but also general-purpose zero knowledge proofs, are possible with just hashes was definitely something I did not expect five years ago; I am very happy that this is the case. That said, size continues to be an issue, and ongoing progress (eg. see the very recent DEEP FRI) is continuing to reduce the size of proofs, though it looks like further progress will be incremental.
The main not-yet-solved problem with hash-based cryptography is aggregate signatures, similar to what BLS aggregation makes possible. It's known that we can just make a STARK over many Lamport signatures, but this is inefficient; a more efficient scheme would be welcome. (In case you're wondering if hash-based public key encryption is possible, the answer is, no, you can't do anything with more than a quadratic attack cost)
Consensus theory problems
6 ASIC-Resistant Proof of Work
One approach at solving the problem is creating a proof-of-work algorithm based on a type of computation that is very difficult to specialize ... For a more in-depth discussion on ASIC-resistant hardware, see https://blog.ethereum.org/2014/06/19/mining/.
Status: Solved as far as we can.
About six months after the "hard problems" list was posted, Ethereum settled on its ASIC-resistant proof of work algorithm: Ethash. Ethash is known as a memory-hard algorithm. The theory is that random-access memory in regular computers is well-optimized already and hence difficult to improve on for specialized applications. Ethash aims to achieve ASIC resistance by making memory access the dominant part of running the PoW computation. Ethash was not the first memory-hard algorithm, but it did add one innovation: it uses pseudorandom lookups over a two-level DAG, allowing for two ways of evaluating the function. First, one could compute it quickly if one has the entire (~2 GB) DAG; this is the memory-hard "fast path". Second, one can compute it much more slowly (still fast enough to check a single provided solution quickly) if one only has the top level of the DAG; this is used for block verification.
Ethash has proven remarkably successful at ASIC resistance; after three years and billions of dollars of block rewards, ASICs do exist but are at best 2-5 times more power and cost-efficient than GPUs. ProgPoW has been proposed as an alternative, but there is a growing consensus that ASIC-resistant algorithms will inevitably have a limited lifespan, and that ASIC resistance has downsides because it makes 51% attacks cheaper (eg. see the 51% attack on Ethereum Classic).
I believe that PoW algorithms that provide a medium level of ASIC resistance can be created, but such resistance is limited-term and both ASIC and non-ASIC PoW have disadvantages; in the long term the better choice for blockchain consensus is proof of stake.
7 Useful Proof of Work
[M]aking the proof of work function something which is simultaneously useful; a common candidate is something like Folding@home, an existing program where users can download software onto their computers to simulate protein folding and provide researchers with a large supply of data to help them cure diseases.
Status: Probably not feasible, with one exception.
The challenge with useful proof of work is that a proof of work algorithm requires many properties:
. Hard to compute
. Easy to verify
. Does not depend on large amounts of external data
. Can be efficiently computed in small "bite-sized" chunks
Unfortunately, there are not many computations that are useful that preserve all of these properties, and most computations that do have all of those properties and are "useful" are only "useful" for far too short a time to build a cryptocurrency around them.
However, there is one possible exception: zero-knowledge-proof generation. Zero knowledge proofs of aspects of blockchain validity (eg. data availability roots for a simple example) are difficult to compute, and easy to verify. Furthermore, they are durably difficult to compute; if proofs of "highly structured" computation become too easy, one can simply switch to verifying a blockchain's entire state transition, which becomes extremely expensive due to the need to model the virtual machine and random memory accesses.
Zero-knowledge proofs of blockchain validity provide great value to users of the blockchain, as they can substitute the need to verify the chain directly; Coda is doing this already, albeit with a simplified blockchain design that is heavily optimized for provability. Such proofs can significantly assist in improving the blockchain's safety and scalability. That said, the total amount of computation that realistically needs to be done is still much less than the amount that's currently done by proof of work miners, so this would at best be an add-on for proof of stake blockchains, not a full-on consensus algorithm.
8 Proof of Stake
Another approach to solving the mining centralization problem is to abolish mining entirely, and move to some other mechanism for counting the weight of each node in the consensus. The most popular alternative under discussion to date is "proof of stake" - that is to say, instead of treating the consensus model as "one unit of CPU power, one vote" it becomes "one currency unit, one vote".
Status: Great theoretical progress, pending more real-world evaluation.
Near the end of 2014, it became clear to the proof of stake community that some form of "weak subjectivity" is unavoidable. To maintain economic security, nodes need to obtain a recent checkpoint extra-protocol when they sync for the first time, and again if they go offline for more than a few months. This was a difficult pill to swallow; many PoW advocates still cling to PoW precisely because in a PoW chain the "head" of the chain can be discovered with the only data coming from a trusted source being the blockchain client software itself. PoS advocates, however, were willing to swallow the pill, seeing the added trust requirements as not being large. From there the path to proof of stake through long-duration security deposits became clear.
Most interesting consensus algorithms today are fundamentally similar to PBFT, but replace the fixed set of validators with a dynamic list that anyone can join by sending tokens into a system-level smart contract with time-locked withdrawals (eg. a withdrawal might in some cases take up to 4 months to complete). In many cases (including ethereum 2.0), these algorithms achieve "economic finality" by penalizing validators that are caught performing actions that violate the protocol in certain ways (see here for a philosophical view on what proof of stake accomplishes).
As of today, we have (among many other algorithms):
Casper FFG: https://arxiv.org/abs/1710.09437
Tendermint: https://tendermint.com/docs/spec/consensus/consensus.html
HotStuff: https://arxiv.org/abs/1803.05069
Casper CBC: https://vitalik.ca/general/2018/12/05/cbc_casper.html
There continues to be ongoing refinement (eg. here and here) . Eth2 phase 0, the chain that will implement FFG, is currently under implementation and enormous progress has been made. Additionally, Tendermint has been running, in the form of the Cosmos chain for several months. Remaining arguments about proof of stake, in my view, have to do with optimizing the economic incentives, and further formalizing the strategy for responding to 51% attacks. Additionally, the Casper CBC spec could still use concrete efficiency improvements.
9 Proof of Storage
A third approach to the problem is to use a scarce computational resource other than computational power or currency. In this regard, the two main alternatives that have been proposed are storage and bandwidth. There is no way in principle to provide an after-the-fact cryptographic proof that bandwidth was given or used, so proof of bandwidth should most accurately be considered a subset of social proof, discussed in later problems, but proof of storage is something that certainly can be done computationally. An advantage of proof-of-storage is that it is completely ASIC-resistant; the kind of storage that we have in hard drives is already close to optimal.
Status: A lot of theoretical progress, though still a lot to go, as well as more real-world evaluation.
There are a number of blockchains planning to use proof of storage protocols, including Chia and Filecoin. That said, these algorithms have not been tested in the wild. My own main concern is centralization: will these algorithms actually be dominated by smaller users using spare storage capacity, or will they be dominated by large mining farms?
Economics
10 Stable-value cryptoassets
One of the main problems with Bitcoin is the issue of price volatility ... Problem: construct a cryptographic asset with a stable price.
Status: Some progress.
MakerDAO is now live, and has been holding stable for nearly two years. It has survived a 93% drop in the value of its underlying collateral asset (ETH), and there is now more than $100 million in DAI issued. It has become a mainstay of the Ethereum ecosystem, and many Ethereum projects have or are integrating with it. Other synthetic token projects, such as UMA, are rapidly gaining steam as well.
However, while the MakerDAO system has survived tough economic conditions in 2019, the conditions were by no means the toughest that could happen. In the past, Bitcoin has fallen by 75% over the course of two days; the same may happen to ether or any other collateral asset some day. Attacks on the underlying blockchain are an even larger untested risk, especially if compounded by price decreases at the same time. Another major challenge, and arguably the larger one, is that the stability of MakerDAO-like systems is dependent on some underlying oracle scheme. Different attempts at oracle systems do exist (see #16), but the jury is still out on how well they can hold up under large amounts of economic stress. So far, the collateral controlled by MakerDAO has been lower than the value of the MKR token; if this relationship reverses MKR holders may have a collective incentive to try to "loot" the MakerDAO system. There are ways to try to protect against such attacks, but they have not been tested in real life.
11 Decentralized Public Goods Incentivization
One of the challenges in economic systems in general is the problem of "public goods". For example, suppose that there is a scientific research project which will cost $1 million to complete, and it is known that if it is completed the resulting research will save one million people $5 each. In total, the social benefit is clear ... [but] from the point of view of each individual person contributing does not make sense ... So far, most problems to public goods have involved centralization
Additional Assumptions And Requirements: A fully trustworthy oracle exists for determining whether or not a certain public good task has been completed (in reality this is false, but this is the domain of another problem)
Status: Some progress.
The problem of funding public goods is generally understood to be split into two problems: the funding problem (where to get funding for public goods from) and the preference aggregation problem (how to determine what is a genuine public good, rather than some single individual's pet project, in the first place). This problem focuses specifically on the former, assuming the latter is solved (see the "decentralized contribution metrics" section below for work on that problem).
In general, there haven't been large new breakthroughs here. There's two major categories of solutions. First, we can try to elicit individual contributions, giving people social rewards for doing so. My own proposal for charity through marginal price discrimination is one example of this; another is the anti-malaria donation badges on Peepeth. Second, we can collect funds from applications that have network effects. Within blockchain land there are several options for doing this:
. Issuing coins
. Taking a portion of transaction fees at protocol level (eg. through EIP 1559)
. Taking a portion of transaction fees from some layer-2 application (eg. Uniswap, or some scaling solution, or even state rent in an execution environment in ethereum 2.0)
. Taking a portion of other kinds of fees (eg. ENS registration)
Outside of blockchain land, this is just the age-old question of how to collect taxes if you're a government, and charge fees if you're a business or other organization.
12 Reputation systems
Problem: design a formalized reputation system, including a score rep(A,B) -> V where V is the reputation of B from the point of view of A, a mechanism for determining the probability that one party can be trusted by another, and a mechanism for updating the reputation given a record of a particular open or finalized interaction.
Status: Slow progress.
There hasn't really been much work on reputation systems since 2014. Perhaps the best is the use of token curated registries to create curated lists of trustable entities/objects; the Kleros ERC20 TCR (yes, that's a token-curated registry of legitimate ERC20 tokens) is one example, and there is even an alternative interface to Uniswap (http://uniswap.ninja) that uses it as the backend to get the list of tokens and ticker symbols and logos from. Reputation systems of the subjective variety have not really been tried, perhaps because there is just not enough information about the "social graph" of people's connections to each other that has already been published to chain in some form. If such information starts to exist for other reasons, then subjective reputation systems may become more popular.
13 Proof of excellence
One interesting, and largely unexplored, solution to the problem of [token] distribution specifically (there are reasons why it cannot be so easily used for mining) is using tasks that are socially useful but require original human-driven creative effort and talent. For example, one can come up with a "proof of proof" currency that rewards players for coming up with mathematical proofs of certain theorems
Status: No progress, problem is largely forgotten.
The main alternative approach to token distribution that has instead become popular is airdrops; typically, tokens are distributed at launch either proportionately to existing holdings of some other token, or based on some other metric (eg. as in the Handshake airdrop). Verifying human creativity directly has not really been attempted, and with recent progress on AI the problem of creating a task that only humans can do but computers can verify may well be too difficult.
15 [sic]. Anti-Sybil systems
A problem that is somewhat related to the issue of a reputation system is the challenge of creating a "unique identity system" - a system for generating tokens that prove that an identity is not part of a Sybil attack ... However, we would like to have a system that has nicer and more egalitarian features than "one-dollar-one-vote"; arguably, one-person-one-vote would be ideal.
Status: Some progress.
There have been quite a few attempts at solving the unique-human problem. Attempts that come to mind include (incomplete list!):
. HumanityDAO: https://www.humanitydao.org/
. Pseudonym parties: https://bford.info/pub/net/sybil.pdf
. POAP ("proof of attendance protocol"): https://www.poap.xyz/
. BrightID: https://www.brightid.org/
With the growing interest in techniques like quadratic voting and quadratic funding, the need for some kind of human-based anti-sybil system continues to grow. Hopefully, ongoing development of these techniques and new ones can come to meet it.
14 [sic]. Decentralized contribution metrics
Incentivizing the production of public goods is, unfortunately, not the only problem that centralization solves. The other problem is determining, first, which public goods are worth producing in the first place and, second, determining to what extent a particular effort actually accomplished the production of the public good. This challenge deals with the latter issue.
Status: Some progress, some change in focus.
More recent work on determining value of public-good contributions does not try to separate determining tasks and determining quality of completion; the reason is that in practice the two are difficult to separate. Work done by specific teams tends to be non-fungible and subjective enough that the most reasonable approach is to look at relevance of task and quality of performance as a single package, and use the same technique to evaluate both.
Fortunately, there has been great progress on this, particularly with the discovery of quadratic funding. Quadratic funding is a mechanism where individuals can make donations to projects, and then based on the number of people who donated and how much they donated, a formula is used to calculate how much they would have donated if they were perfectly coordinated with each other (ie. took each other's interests into account and did not fall prey to the tragedy of the commons). The difference between amount would-have-donated and amount actually donated for any given project is given to that project as a subsidy from some central pool (see #11 for where the central pool funding could come from). Note that this mechanism focuses on satisfying the values of some community, not on satisfying some given goal regardless of whether or not anyone cares about it. Because of the complexity of values problem, this approach is likely to be much more robust to unknown unknowns.
Quadratic funding has even been tried in real life with considerable success in the recent gitcoin quadratic funding round. There has also been some incremental progress on improving quadratic funding and similar mechanisms; particularly, pairwise-bounded quadratic funding to mitigate collusion. There has also been work on specification and implementation of bribe-resistant voting technology, preventing users from proving to third parties who they voted for; this prevents many kinds of collusion and bribe attacks.
16 Decentralized success metrics
Problem: come up with and implement a decentralized method for measuring numerical real-world variables ... the system should be able to measure anything that humans can currently reach a rough consensus on (eg. price of an asset, temperature, global CO2 concentration)
Status: Some progress.
This is now generally just called "the oracle problem". The largest known instance of a decentralized oracle running is Augur, which has processed outcomes for millions of dollars of bets. Token curated registries such as the Kleros TCR for tokens are another example. However, these systems still have not seen a real-world test of the forking mechanism (search for "subjectivocracy" here) either due to a highly controversial question or due to an attempted 51% attack. There is also research on the oracle problem happening outside of the blockchain space in the form of the "peer prediction" literature; see here for a very recent advancement in the space.
Another looming challenge is that people want to rely on these systems to guide transfers of quantities of assets larger than the economic value of the system's native token. In these conditions, token holders in theory have the incentive to collude to give wrong answers to steal the funds. In such a case, the system would fork and the original system token would likely become valueless, but the original system token holders would still get away with the returns from whatever asset transfer they misdirected. Stablecoins (see #10) are a particularly egregious case of this. One approach to solving this would be a system that assumes that altruistically honest data providers do exist, and creating a mechanism to identify them, and only allowing them to churn slowly so that if malicious ones start getting voted in the users of systems that rely on the oracle can first complete an orderly exit. In any case, more development of oracle tech is very much an important problem.
New problems
If I were to write the hard problems list again in 2019, some would be a continuation of the above problems, but there would be significant changes in emphasis, as well as significant new problems. Here are a few picks:
. Cryptographic obfuscation: same as #4 above
. Ongoing work on post-quantum cryptography: both hash-based as well as based on post-quantum-secure "structured" mathematical objects, eg. elliptic curve isogenies, lattices...
. Anti-collusion infrastructure: ongoing work and refinement of https://ethresear.ch/t/minimal-anti-collusion-infrastructure/5413, including adding privacy against the operator, adding multi-party computation in a maximally practical way, etc.
. Oracles: same as #16 above, but removing the emphasis on "success metrics" and focusing on the general "get real-world data" problem
. Unique-human identities (or, more realistically, semi-unique-human identities): same as what was written as #15 above, but with an emphasis on a less "absolute" solution: it should be much harder to get two identities than one, but making it impossible to get multiple identities is both impossible and potentially harmful even if we do succeed
. Homomorphic encryption and multi-party computation: ongoing improvements are still required for practicality
. Decentralized governance mechanisms: DAOs are cool, but current DAOs are still very primitive; we can do better
Fully formalizing responses to PoS 51% attacks: ongoing work and refinement of https://ethresear.ch/t/responding-to-51-attacks-in-casper-ffg/6363
. More sources of public goods funding: the ideal is to charge for congestible resources inside of systems that have network effects (eg. transaction fees), but doing so in decentralized systems requires public legitimacy; hence this is a social problem along with the technical one of finding possible sources
. Reputation systems: same as #12 above
In general, base-layer problems are slowly but surely decreasing, but application-layer problems are only just getting started.
Neuroticism (negative), extraversion, agreeableness, and to a lesser extent conscientiousness predicted wellbeing; the hypothesis that self-enhancement is beneficial for wellbeing is doubtful
An integrated model of social psychological and personality psychological perspectives on personality and wellbeing. Ulrich Schimmack, Hyunji Kim. Journal of Research in Personality, Volume 84, February 2020, 103888. https://doi.org/10.1016/j.jrp.2019.103888
Highlights
• Largest sample size for multi-method studies of self-enhancement.
• No support for benefits of positive illusions on wellbeing.
• Multi-method evidence that personality influences well-being.
Abstract: This article uses multi-rater data from 458 triads (students, mother, father, total N = 1374) to examine the relationship of personality ratings with wellbeing ratings, using a multi-method approach to separate accurate perceptions (shared across raters) from biased perceptions of the self (rater-specific variance). The social-psychological perspective predicts effects of halo bias in self-ratings on wellbeing, whereas the personality-psychological perspective predicts effects of personality traits on wellbeing. Results are more consistent with the personality perspective in that neuroticism (negative), extraversion, agreeableness, and to a lesser extent conscientiousness predicted wellbeing, whereas positive illusions about the self were only weakly and not significantly related to wellbeing. These results cast doubt on the hypothesis that self-enhancement is beneficial for wellbeing.
Highlights
• Largest sample size for multi-method studies of self-enhancement.
• No support for benefits of positive illusions on wellbeing.
• Multi-method evidence that personality influences well-being.
Abstract: This article uses multi-rater data from 458 triads (students, mother, father, total N = 1374) to examine the relationship of personality ratings with wellbeing ratings, using a multi-method approach to separate accurate perceptions (shared across raters) from biased perceptions of the self (rater-specific variance). The social-psychological perspective predicts effects of halo bias in self-ratings on wellbeing, whereas the personality-psychological perspective predicts effects of personality traits on wellbeing. Results are more consistent with the personality perspective in that neuroticism (negative), extraversion, agreeableness, and to a lesser extent conscientiousness predicted wellbeing, whereas positive illusions about the self were only weakly and not significantly related to wellbeing. These results cast doubt on the hypothesis that self-enhancement is beneficial for wellbeing.
4. Discussion
The main contribution of this article was to examine wellbeing from an integrated personality and social psychological perspective. While personality psychologists focused on the contribution of actual traits, social psychologists focused on biases in self-perceptions of traits. Multi-method measurement models were used to separate valid trait variance from illusory perceptions of personality in self-ratings and ratings of other family members. The results show that actual personality traits are more important for wellbeing than positive biases in self-perceptions. In fact, the most important finding was that positive illusions about the self were unrelated to wellbeing impressions that are shared across informants. This finding challenges Taylor and Brown (1988) influential and highly controversial claim that positive illusions not only foster higher wellbeing, but are a sign of optimal and normal functioning. Subsequently, we discuss the implications of our findings for the future of wellbeing science and for individuals’ pursuit of wellbeing.
4.1. Positive illusions and public wellbeing
The social psychological perspective on wellbeing is grounded in the basic assumption that human information processing is riddled with errors. Taylor and Brown (1988) quote Fiske and Taylor (1984) book about social cognitions to support this assumption. “Instead of a naïve scientist entering the environment in search of the truth, we find the rather unflattering picture of a charlatan trying to make the data come out in a manner most advantageous to his or her already-held theories” (p. 88). Thirty years later, it has become apparent that human information processing is more accurate than Fiske and Taylor (1984) assumed (Funder, 1995, Jussim, 1991, McCrae and Costa, 1991, Schimmack and Oishi, 2005). Thus, Taylor and Brown (1988) model of wellbeing is based on outdated evidence and needs to be revised.
The vast majority of studies have relied on self-ratings of wellbeing to measure benefits of wellbeing. This is problematic because self-ratings of wellbeing can be inflated by the very same processes that inflate self-ratings of personality (Humberg et al., 2019). There have been only a handful of studies with valid illusion measures and informant ratings of wellbeing and these studies have found similar weak results (Dufner et al., 2019).
The lack of evidence for benefits of positive illusions is not for a lack of trying. Taylor, Lerner, Sherman, Sage, and McDowell (2003) claimed that effects of positive illusions are not limited to self-ratings. “We conducted a study with multiple measures of self-enhancement along with multiple measures and judges of mental health, comprehensively assessing their relationship. The results indicated that self-enhancement is positively associated with multiple indicators of mental health” (p. 165). Contrary to this claim, Table 5 shows correlations of various self-enhancement measures with peer-rated mental health ranging from r = −0.13 to 0.09. None of these correlations were significant, in part due to the low statistical power of the study (N = 55). Thus, even Taylor and colleagues never provided positive evidence that positive illusions increase wellbeing in ways that can be measured with a method other than self-reports. The social cognitive model of wellbeing also faces other problems. One problem is causality. Even if there were a small correlation between positive illusions about the self and wellbeing, it is not clear that it is causal. It is equally plausible that happiness distorts self-perceptions. Thirty years of research have failed to address this problem (cf. Humberg et al., 2019). Another problem is that third variables produce a spurious correlation between illusions about the self and wellbeing. For example, relationship researchers have shown that illusions about a partner predict relationship satisfaction (see Weidmann, Ledermann, & Grob, 2016, for a review), and Kim et al. (2012) showed that individuals with positive illusions about the self also tend to have positive illusions about others. Thus, it is possible that positive illusions about others, not the self, are beneficial for social relationships and wellbeing. Future research needs to include measures of positive illusions about the self and others to examine this question. Given these problems, we question broad conclusions about the benefits of positive illusions for wellbeing (Dufner et al., 2019; Humberg et al., 2019).
4.2. Positive illusions and private wellbeing
The present study replicated the finding that positive illusions predict unique variance in self-ratings of wellbeing. That is, individuals who claim to be more extraverted and more agreeable than others perceive them also claim to be happier than others perceive them to be (Dufner et al., 2019, Humberg et al., 2019; Taylor et al., 2003). As noted in the introduction, there are two possible explanation for this finding. One explanation is that positive illusions enhance wellbeing in a way that is not observable to others. The challenge for this model is to explain how positive illusions foster private wellbeing and to provide empirical evidence for this model. To explain why informants are unable to see the happiness of individuals with positive illusions, we have to assume that the illusion-based happiness is not visible to others. This requires a careful examination of the variance in self-ratings of wellbeing that is not shared with informants (Schneider & Schimmack, 2010).
The private-wellbeing illusion model also faces an interesting contradiction in assumptions about the validity of personality and wellbeing judgments. To allow for effects of positive illusions on private wellbeing, the model assumes that people have illusions about their personality, while their self-ratings of wellbeing are highly accurate and trustworthy. In contrast, social psychologists have argued that wellbeing judgments are highly sensitive to context effects and provide little valid information about individuals’ wellbeing (Schwarz & Strack, 1999). In contrast, personality psychologists have pointed to self-informant agreement in wellbeing judgments as evidence for the validity of self-ratings of wellbeing. If informant ratings validate self-ratings, then we would expect predictors of wellbeing also to be related to self-ratings of wellbeing and to informant ratings of wellbeing. Our main contribution is to show that this is not the case for positive illusions, or at least, that the effect size is small. No single study can resolve deep philosophical questions, but our study suggests that hundreds of studies that relied on self-ratings of wellbeing to demonstrate the benefits of positive illusions may have produced illusory evidence of these benefits.
4.3. Positive illusions as halo bias
Evidence for halo biases in personality ratings is nearly 100 years old (Thorndike, 1920). Ironically, some of the strongest evidence for the pervasiveness of halo biases stems from social psychology (Nisbett & Wilson, 1977). Given the evidence that halo biases in ratings are pervasive, halo bias provides a simple and parsimonious explanation for the finding that positive illusions are only related to the unique variance in self-ratings and not to informant ratings of wellbeing. One explanation for halo bias is that many trait concepts have a denotative and a connotative (evaluative) meaning (Osgood, Suci, & Tannenbaum, 1957). While denotative meaning and valid information produce agreement between raters, ratings are also biased by the connotative meaning of words and liking of a target. For example, lazy has a denotative meaning of not putting a lot of effort into tasks and a negative connotation. Ratings of laziness will be enhanced by dislike and attenuated by liking of an individual independent of the objective effort targets exert (Leising, Erbs, & Fritz, 2010). It seems plausible that halo bias also influences ratings of desirable attributes like happiness and having a good life. Thus, halo bias offers a plausible explanation for our results that is also consistent with heuristic and bias models in social psychology.
4.4. Personality and wellbeing
The present study provided new evidence on the relationship between personality and wellbeing from a multi-rater perspective. Results confirmed that neuroticism is the strongest predictor of wellbeing and that the influence on wellbeing is mediated by hedonic balance. This finding is consistent with the hypothesis that neuroticism is a broad disposition to experience more unpleasant mood states (Costa and McCrae, 1980, Schimmack, Radhakrishnan, Oishi et al., 2002, Watson and Tellegen, 1985). As experiencing unpleasant mood is undesirable it lowers wellbeing independent of actual life-circumstances. Twin studies suggest that individual differences in neuroticism are partially heritable and that the genetic variance in neuroticism accounts for a considerable portion of the shared variance between neuroticism and wellbeing (Nes et al., 2013).
In comparison, the other personality traits explain relatively small amounts of variance in wellbeing. While, the effects of extraversion and agreeableness were also mediated by hedonic balance, the results for conscientiousness suggested a unique influence on life evaluations. Future research needs to go beyond demonstrating effects of the Big Five and wellbeing and start to investigate the causal processes that link personality to wellbeing. McCrae and Costa (1991) proposed that agreeableness is beneficial for more harmonious social relationships, while conscientiousness is beneficial for work, but there have been few attempts to test these predictions. One way to test potential mediators are integrated top-down bottom-up models with domain satisfaction as mediators (Brief et al., 1993, Schimmack, Diener, Oishi, 2002). It is important to use multi-method measurement models to separate top-down effects from halo bias (Schneider & Schimmack, 2010). It is also important to examine the relationship of personality and wellbeing with a more detailed assessment of personality traits. While the Big Five have the advantage of covering a broad range of personality traits with a few, largely orthogonal dimensions, the disadvantage is that they cannot represent all of the variation in personality. Some studies showed that the depression facet of neuroticism and the cheerfulness facet of extraversion explain additional variance in wellbeing (Allik et al., 2018, Schimmack et al., 2004). More research with narrow personality traits is needed to specify the precise personality traits that are related to wellbeing.
Pseudo-profound bullshit titles makes the art grow profounder
Bullshit makes the art grow profounder. Martin Harry Turpin et al. Judgment and Decision Making, Vol. 14, No. 6, November 2019, pp. 658-670. http://journal.sjdm.org/19/190712/jdm190712.html
Abstract: Across four studies participants (N = 818) rated the profoundness of abstract art images accompanied with varying categories of titles, including: pseudo-profound bullshit titles (e.g., The Deaf Echo), mundane titles (e.g., Canvas 8), and no titles. Randomly generated pseudo-profound bullshit titles increased the perceived profoundness of computer-generated abstract art, compared to when no titles were present (Study 1). Mundane titles did not enhance the perception of profoundness, indicating that pseudo-profound bullshit titles specifically (as opposed to titles in general) enhance the perceived profoundness of abstract art (Study 2). Furthermore, these effects generalize to artist-created abstract art (Study 3). Finally, we report a large correlation between profoundness ratings for pseudo-profound bullshit and “International Art English” statements (Study 4), a mode and style of communication commonly employed by artists to discuss their work. This correlation suggests that these two independently developed communicative modes share underlying cognitive mechanisms in their interpretations. We discuss the potential for these results to be integrated into a larger, new theoretical framework of bullshit as a low-cost strategy for gaining advantages in prestige awarding domains.
Keywords: pseudo-profound bullshit, impression management, abstract art, meaning, social navigation
The complex relation between receptivity to pseudo-profound bullshit and political ideology. Nilsson, Artur; ERLANDSSON, Arvid and Västfjäll, Daniel (2018) In Personality and Social Psychology Bulletin, Jan 2019. https://www.bipartisanalliance.com/2019/01/bullshit-receptivity-robustly.html
Check also the first author's MA Thesis... Bullshit Makes the Art Grow Profounder: Evidence for False Meaning Transfer Across Domains. Martin Harry Turpin. MA Thesis, Waterloo Univ., Ontario. https://www.bipartisanalliance.com/2018/10/pairing-abstract-art-pieces-with.html
And Bullshit-sensitivity predicts prosocial behavior. Arvid Erlandsson et al. PLOS, https://www.bipartisanalliance.com/2018/08/bullshit-receptivity-perceived.html
Abstract: Across four studies participants (N = 818) rated the profoundness of abstract art images accompanied with varying categories of titles, including: pseudo-profound bullshit titles (e.g., The Deaf Echo), mundane titles (e.g., Canvas 8), and no titles. Randomly generated pseudo-profound bullshit titles increased the perceived profoundness of computer-generated abstract art, compared to when no titles were present (Study 1). Mundane titles did not enhance the perception of profoundness, indicating that pseudo-profound bullshit titles specifically (as opposed to titles in general) enhance the perceived profoundness of abstract art (Study 2). Furthermore, these effects generalize to artist-created abstract art (Study 3). Finally, we report a large correlation between profoundness ratings for pseudo-profound bullshit and “International Art English” statements (Study 4), a mode and style of communication commonly employed by artists to discuss their work. This correlation suggests that these two independently developed communicative modes share underlying cognitive mechanisms in their interpretations. We discuss the potential for these results to be integrated into a larger, new theoretical framework of bullshit as a low-cost strategy for gaining advantages in prestige awarding domains.
Keywords: pseudo-profound bullshit, impression management, abstract art, meaning, social navigation
The complex relation between receptivity to pseudo-profound bullshit and political ideology. Nilsson, Artur; ERLANDSSON, Arvid and Västfjäll, Daniel (2018) In Personality and Social Psychology Bulletin, Jan 2019. https://www.bipartisanalliance.com/2019/01/bullshit-receptivity-robustly.html
Check also the first author's MA Thesis... Bullshit Makes the Art Grow Profounder: Evidence for False Meaning Transfer Across Domains. Martin Harry Turpin. MA Thesis, Waterloo Univ., Ontario. https://www.bipartisanalliance.com/2018/10/pairing-abstract-art-pieces-with.html
And Bullshit-sensitivity predicts prosocial behavior. Arvid Erlandsson et al. PLOS, https://www.bipartisanalliance.com/2018/08/bullshit-receptivity-perceived.html
Non-believers: Reflection increases belief in God through self-questioning
Reflection increases belief in God through self-questioning among non-believers. Onurcan Yilmaz, Ozan Isler. Judgment and Decision Making, Vol. 14, No. 6, November 2019, pp. 649-657. http://journal.sjdm.org/19/190605/jdm190605.html
The dual-process model of the mind predicts that religious belief will be stronger for intuitive decisions, whereas reflective thinking will lead to religious disbelief (i.e., the intuitive religious belief hypothesis). While early research found intuition to promote and reflection to weaken belief in God, more recent attempts found no evidence for the intuitive religious belief hypothesis. Many of the previous studies are underpowered to detect small effects, and it is not clear whether the cognitive process manipulations used in these failed attempts worked as intended. We investigated the influence of intuitive and reflective thought on belief in God in two large-scale preregistered experiments (N = 1,602), using well-established cognitive manipulations (i.e., time-pressure with incentives for compliance) and alternative elicitation methods (between and within-subject designs). Against our initial hypothesis based on the literature, the experiments provide first suggestive then confirmatory evidence for the reflective religious belief hypothesis. Exploratory examination of the data suggests that reflection increases doubts about beliefs held regarding God’s existence. Reflective doubt exists primarily among non-believers, resulting in an overall increase in belief in God when deciding reflectively.
Keywords: reflection, intuition, analytic cognitive style, belief, belief in God or gods
Why does reflection increase belief in God in the current research? Our exploratory analysis strongly suggests that reflection, rather than directly increasing belief in God, increases doubt about one’s initial and intuitively held belief regarding God’s existence. It is likely that reflection increased religious belief in our overall sample because religious self-questioning is stronger among non-believers than among believers. On the other hand, we show that endorsement of agnosticism, deism, and polytheism is associated with both increase and decrease in belief in God, which may drive reflective doubt. Future research should try to experimentally distinguish this reflective religious doubt hypothesis implicated by our exploratory analysis from the reflective religious belief hypothesis. Nevertheless, we expect the effect of reflection on religious belief to be small because the belief in God question, as regularly used in the literature, will tend to probe stable opinions. Having answered the same question numerous times over the course of one’s life, participants are likely to know, as a defining characteristic of their personal identity, whether and to what extent they believe in God.
We also hypothesized but found no strong evidence that Pascal’s Wager may motivate a religious belief. Accordingly, reflected evaluation of the possibility of God’s existence could highlight the potentially infinite benefits of belief and costs of disbelief, hence questioning religious disbelief through a rational utility calculus. Although plausible, the tendency in our sample to agree with Pascal’s Wager did not clearly explain the reflected change in religious belief. However, our test was limited by the fact that religious believers (i.e., those with already high levels of belief) agreed with the Wager more than non-believers as well as by the fact that there were fewer atheists and agnostics in our sample.
An alternative explanation of the positive effect of reflection on religious belief may be that reflection makes people less extreme in their beliefs in general (i.e., religious and non-religious) but that openness to such self-criticism may be stronger among non-believers since they also tend to be reflective thinkers (Pennycook et al., 2016). Comparing religious and secular belief change among non-believers can therefore provide an explanation for our main finding. Likewise, Pascal’s Wager can be tested using improved methods, for example, by studying the effect of Pascal’s argument as an experimental manipulation. Finally, the two-stage procedure used in Experiment 2 was more insightful to studying religious belief change than the standard between-subject design of Experiment 1. The two-stage technique can be used in future studies of cooperation and morality in order to dissociate dual cognitive processes.
We also suggest that these experimental manipulations might have more influence on less stable beliefs or on those who are less confident about the existence of God. A similar distinction has been made in the field of political psychology (Talhelm, 2018; Talhelm et al., 2015; Yilmaz & Saribay, 2016, 2017). Activating reflective thinking did not have an impact on political opinions when they were measured by standard scale items based on identity labels (e.g., liberal or conservative), but it led to a significant change in less stable contextualized opinions (e.g., forming opinions about a newspaper article; Yilmaz & Saribay, 2017). A similar distinction can be made in the field of cognitive science of religion. For example, while belief in God, reflecting relatively stable opinions, may be more resistant to cognitive process manipulations, the relative reliance on natural vs. supernatural explanations for an uncertain event (e.g., the disappearance of airplanes in the Bermuda Triangle) may be more open to the influence of intuitive and reflective thinking. This possibility should be examined in future research.
A surprising contrast emerges from our data: the positive causal effect of reflection on belief in God vs. the negative correlation between individual tendency for reflected thinking and religious belief. While it is not clear why experimental and correlational tests lead to different conclusions, one may conjecture that the two approaches capture separate psychological mechanisms occurring across distinct time-frames. In particular, correlational measures may reflect self-selection of intuitively inclined people to religious belief (a long-term process of identity formation), while promoting reflection may isolate the possibly short-term effects of questioning one’s own and already established beliefs. While correlational findings are prevalent in the literature, there is a need for more experimental research on this topic. In particular, the generalizability of our results across cultures (e.g., using multi-lab experiments) is an open question.
In sum, recent failures to support the intuitive religious belief hypothesis suggested that the early evidence supporting the hypothesis is not easily reproducible. Using stronger manipulations and two large-scale experiments, we found that the effect of reflection and intuition on belief in God is in fact the opposite of intuitive belief hypothesis. Our results suggest that reflection on God’s existence may promote religious self-questioning, especially among non-believers.
The dual-process model of the mind predicts that religious belief will be stronger for intuitive decisions, whereas reflective thinking will lead to religious disbelief (i.e., the intuitive religious belief hypothesis). While early research found intuition to promote and reflection to weaken belief in God, more recent attempts found no evidence for the intuitive religious belief hypothesis. Many of the previous studies are underpowered to detect small effects, and it is not clear whether the cognitive process manipulations used in these failed attempts worked as intended. We investigated the influence of intuitive and reflective thought on belief in God in two large-scale preregistered experiments (N = 1,602), using well-established cognitive manipulations (i.e., time-pressure with incentives for compliance) and alternative elicitation methods (between and within-subject designs). Against our initial hypothesis based on the literature, the experiments provide first suggestive then confirmatory evidence for the reflective religious belief hypothesis. Exploratory examination of the data suggests that reflection increases doubts about beliefs held regarding God’s existence. Reflective doubt exists primarily among non-believers, resulting in an overall increase in belief in God when deciding reflectively.
Keywords: reflection, intuition, analytic cognitive style, belief, belief in God or gods
4 Discussion
In both experiments, we found that reflection increases belief in God and that the effect is stronger among non-believers. Exploratory analysis suggested that the overall increase in religious belief is likely due to the religious self-questioning (i.e., reflective doubt) of non-believers who tended to revise their responses on the scale towards the middle point (i.e., “not sure”). The results also showed that those who make greater use of their reflective capacities (as measured by CRT-2) are less likely to endorse belief in God or gods. These results provide evidence against the hypothesis that intuition fosters and that reflection dampens religious belief (Gervais & Norenzayan, 2012; Shenhav et al., 2012; Yilmaz et al., 2016) but it converges with the longstanding correlational results demonstrating that tendency for reflective thinking is negatively associated with religious belief (e.g., Bahçekapili & Yilmaz, 2017; Gervais et al., 2018; Pennycook et al., 2016; Stagnaro et al., 2018; Stagnaro, Ross, Pennycook & Rand, 2019).Why does reflection increase belief in God in the current research? Our exploratory analysis strongly suggests that reflection, rather than directly increasing belief in God, increases doubt about one’s initial and intuitively held belief regarding God’s existence. It is likely that reflection increased religious belief in our overall sample because religious self-questioning is stronger among non-believers than among believers. On the other hand, we show that endorsement of agnosticism, deism, and polytheism is associated with both increase and decrease in belief in God, which may drive reflective doubt. Future research should try to experimentally distinguish this reflective religious doubt hypothesis implicated by our exploratory analysis from the reflective religious belief hypothesis. Nevertheless, we expect the effect of reflection on religious belief to be small because the belief in God question, as regularly used in the literature, will tend to probe stable opinions. Having answered the same question numerous times over the course of one’s life, participants are likely to know, as a defining characteristic of their personal identity, whether and to what extent they believe in God.
We also hypothesized but found no strong evidence that Pascal’s Wager may motivate a religious belief. Accordingly, reflected evaluation of the possibility of God’s existence could highlight the potentially infinite benefits of belief and costs of disbelief, hence questioning religious disbelief through a rational utility calculus. Although plausible, the tendency in our sample to agree with Pascal’s Wager did not clearly explain the reflected change in religious belief. However, our test was limited by the fact that religious believers (i.e., those with already high levels of belief) agreed with the Wager more than non-believers as well as by the fact that there were fewer atheists and agnostics in our sample.
An alternative explanation of the positive effect of reflection on religious belief may be that reflection makes people less extreme in their beliefs in general (i.e., religious and non-religious) but that openness to such self-criticism may be stronger among non-believers since they also tend to be reflective thinkers (Pennycook et al., 2016). Comparing religious and secular belief change among non-believers can therefore provide an explanation for our main finding. Likewise, Pascal’s Wager can be tested using improved methods, for example, by studying the effect of Pascal’s argument as an experimental manipulation. Finally, the two-stage procedure used in Experiment 2 was more insightful to studying religious belief change than the standard between-subject design of Experiment 1. The two-stage technique can be used in future studies of cooperation and morality in order to dissociate dual cognitive processes.
We also suggest that these experimental manipulations might have more influence on less stable beliefs or on those who are less confident about the existence of God. A similar distinction has been made in the field of political psychology (Talhelm, 2018; Talhelm et al., 2015; Yilmaz & Saribay, 2016, 2017). Activating reflective thinking did not have an impact on political opinions when they were measured by standard scale items based on identity labels (e.g., liberal or conservative), but it led to a significant change in less stable contextualized opinions (e.g., forming opinions about a newspaper article; Yilmaz & Saribay, 2017). A similar distinction can be made in the field of cognitive science of religion. For example, while belief in God, reflecting relatively stable opinions, may be more resistant to cognitive process manipulations, the relative reliance on natural vs. supernatural explanations for an uncertain event (e.g., the disappearance of airplanes in the Bermuda Triangle) may be more open to the influence of intuitive and reflective thinking. This possibility should be examined in future research.
A surprising contrast emerges from our data: the positive causal effect of reflection on belief in God vs. the negative correlation between individual tendency for reflected thinking and religious belief. While it is not clear why experimental and correlational tests lead to different conclusions, one may conjecture that the two approaches capture separate psychological mechanisms occurring across distinct time-frames. In particular, correlational measures may reflect self-selection of intuitively inclined people to religious belief (a long-term process of identity formation), while promoting reflection may isolate the possibly short-term effects of questioning one’s own and already established beliefs. While correlational findings are prevalent in the literature, there is a need for more experimental research on this topic. In particular, the generalizability of our results across cultures (e.g., using multi-lab experiments) is an open question.
In sum, recent failures to support the intuitive religious belief hypothesis suggested that the early evidence supporting the hypothesis is not easily reproducible. Using stronger manipulations and two large-scale experiments, we found that the effect of reflection and intuition on belief in God is in fact the opposite of intuitive belief hypothesis. Our results suggest that reflection on God’s existence may promote religious self-questioning, especially among non-believers.
Wronging past rights: The sunk cost bias distorts moral judgment
Wronging past rights: The sunk cost bias distorts moral judgment. Ethan A. Meyers et al. Judgment and Decision Making, Vol. 14, No. 6, November 2019, pp. 721-727. http://journal.sjdm.org/19/190909b/jdm190909b.html
When people have invested resources into an endeavor, they typically persist in it, even when it becomes obvious that it will fail. Here we show this bias extends to people’s moral decision-making. Across two preregistered experiments (N = 1592) we show that people are more willing to proceed with a futile, immoral action when costs have been sunk (Experiment 1A and 1B). Moreover, we show that sunk costs distort people’s perception of morality by increasing how acceptable they find actions that have received past investment (Experiment 2). We find these results in contexts where continuing would lead to no obvious benefit and only further harm. We also find initial evidence that the bias has a larger impact on judgment in immoral compared to non-moral contexts. Our findings illustrate a novel way that the past can affect moral judgment. Implications for rational moral judgment and models of moral cognition are discussed.
Keywords: sunk costs, morality, decision-making, judgment, open data, open materials, preregistered
Our findings illustrate a novel way that the past can impact moral judgment. Moral research conducted to-date has focused extensively on future consequences (e.g., Baez et al., 2017; Miller & Cushman, 2013). Although this makes normative sense as only the future should be relevant to decisions, it is well known that choice is affected by irrelevant factors like past investment (Kahneman, 2011; Kahneman, Slovic & Tversky, 1982; Szaszi, Palinkas, Palfi, Szollosi & Aczel, 2018; Tversky & Kahneman, 1974). As such, our findings show that as is true with other (non-moral) judgments, people’s moral judgments are affected by factors that rational agents “should” ignore when making them.
Further, our findings show that a major decision bias (i.e., the sunk cost effect) extends to moral judgment. This finding is broadly consistent with research showing that moral judgments are affected by such biases. This earlier work shows that when making moral judgments, people are sensitive to how options are framed (e.g., Shenhav & Greene, 2010) and prefer acts of omission over commission (e.g., Bostyn & Roets, 2016). For example, people make different moral judgments when the decision is presented in a gain frame than when it is presented in a loss frame, even though these two decisions are logically identical (Kern & Chugh, 2009). Likewise, people judge lying to the police about who is at fault in a car accident (a harmful commission), to be more immoral than not informing the police precisely who is at fault (a harmful omission) (Spranca, Minsk & Baron, 1991). However, unlike most of these previous demonstrations, our findings directly compare the presence of decision-making biases across moral and non-moral contexts (also see Cushman & Young, 2011).
In our first experiment, we also found that the sunk cost bias may be stronger in moral decision-making than in other situations. This is surprising. In non-moral cases proceeding with a futile course of action is wasteful. But in our moral version of the scenarios, proceeding is wasteful, harmful to others, and morally wrong. Yet, there was a greater discrepancy between willingness to act in response to sunk costs in the immoral condition. Increasing the reasons to not proceed with the action amplified the sunk cost bias. One potential explanation for this is that people are unwilling to admit their prior investments were in vain (Brockner, 1992). People succumb to the sunk cost bias in part because they feel a need to justify their past decisions as correct (Ku, 2008; also see Staw, 1976). Likewise, moral judgments seem to generate a much greater need to provide reasons to justify past decisions (Haidt, 2012). Thus, those making decisions in an immoral context might have additional pressures to justify their previous choice that stem from the nature of moral judgment itself.
Another explanation is that the initial investment was of a larger magnitude in the immoral compared to the non-moral condition. In both cases, participants incurred an economic cost, but only in one did participants incur an additional moral cost. People are more likely to succumb to the sunk cost bias when initial investments are large (Arkes & Ayton, 1999; Arkes & Blumer, 1985; Sweis et al., 2018). Perhaps sunk costs exerted a greater effect in the immoral condition because the past investments were greater (i.e., of two kinds: economic and moral, rather than just one: economic). However, as we do not know if the economic resources (e.g., pine trees and lab monkeys) were of comparable value, the discrepancy between moral conditions may entirely stem from the lab monkeys being valued higher and thus larger in investment magnitude. Thus, we are hesitant to draw any strong conclusion from this finding. The difference in sunk cost magnitude could stem from differences in financial costs between the immoral and non-moral contexts.
Our finding that moral violations led to increased willingness to act is reminiscent of the “what the hell” effect, in which people who violate their diet then give up on it and continue to overindulge (Cochran & Tesser, 1996; Polivy, Herman & Deo, 2010). We see this as similar to persisting in an immoral course of action after costs have been sunk. After engaging in a morally equivocal act, people may feel disinhibited and willing to continue the act even when its immorality becomes clear. Likewise, people may persist in an attempt to maintain the status quo (Kahneman, Knetsch & Thaler, 1991; Samuelson & Zeckhauser, 1988). These accounts, though, may not explain why sunk costs changed people’s moral perceptions. One possibility is that this resulted from cognitive dissonance between people’s actions and their moral code (Aronson, 1969; Festinger, 1957; Harmon-Jones & Mills, 1999). For example, sacrificing monkeys to develop a cure may cause dissonance between not wanting to harm but having done so. To resolve this, people might change their moral perceptions, molding their moral code to fit their behavior.
We close by considering a broader implication of this work. The extension of decision biases to moral judgment has been previously construed as supporting domain-general accounts of morality that suggest moral judgment operates similarly to ordinary judgment (Osman & Wiegmann, 2017; Greene, 2015). This is because if morality is not unique, one could reasonably expect that a factor that affects ordinary judgment would likewise affect moral judgment. Thus, if information irrelevant to the decision at hand (e.g., past investments) influences whether we continue to bulldoze land to build a highway, so too should it influence the same bulldoze decision that requires confiscating the land. This is not conclusive however, and our findings could be interpreted to support domain-specific accounts instead (e.g., Mikhail, 2011). For instance, the sunk cost bias was demonstrably larger in moral judgments. Nevertheless, an interpretation of our results as evidence for a domain-general account of morality must explain how the varying effect of past investment on judgment is a difference in degree but not kind.
When people have invested resources into an endeavor, they typically persist in it, even when it becomes obvious that it will fail. Here we show this bias extends to people’s moral decision-making. Across two preregistered experiments (N = 1592) we show that people are more willing to proceed with a futile, immoral action when costs have been sunk (Experiment 1A and 1B). Moreover, we show that sunk costs distort people’s perception of morality by increasing how acceptable they find actions that have received past investment (Experiment 2). We find these results in contexts where continuing would lead to no obvious benefit and only further harm. We also find initial evidence that the bias has a larger impact on judgment in immoral compared to non-moral contexts. Our findings illustrate a novel way that the past can affect moral judgment. Implications for rational moral judgment and models of moral cognition are discussed.
Keywords: sunk costs, morality, decision-making, judgment, open data, open materials, preregistered
4 General Discussion
We found that the sunk cost bias extends to moral judgments. When costs were sunk, participants were more willing to proceed with a futile, immoral action compared to when costs were not sunk. For example, they were more willing to sacrifice monkeys to develop a medical cure when some monkeys had already been sacrificed than when none had been. Moreover, people judged these actions as more acceptable when costs were sunk. Importantly, these effects occurred even though the benefit of the proposed immoral action was eliminated.Our findings illustrate a novel way that the past can impact moral judgment. Moral research conducted to-date has focused extensively on future consequences (e.g., Baez et al., 2017; Miller & Cushman, 2013). Although this makes normative sense as only the future should be relevant to decisions, it is well known that choice is affected by irrelevant factors like past investment (Kahneman, 2011; Kahneman, Slovic & Tversky, 1982; Szaszi, Palinkas, Palfi, Szollosi & Aczel, 2018; Tversky & Kahneman, 1974). As such, our findings show that as is true with other (non-moral) judgments, people’s moral judgments are affected by factors that rational agents “should” ignore when making them.
Further, our findings show that a major decision bias (i.e., the sunk cost effect) extends to moral judgment. This finding is broadly consistent with research showing that moral judgments are affected by such biases. This earlier work shows that when making moral judgments, people are sensitive to how options are framed (e.g., Shenhav & Greene, 2010) and prefer acts of omission over commission (e.g., Bostyn & Roets, 2016). For example, people make different moral judgments when the decision is presented in a gain frame than when it is presented in a loss frame, even though these two decisions are logically identical (Kern & Chugh, 2009). Likewise, people judge lying to the police about who is at fault in a car accident (a harmful commission), to be more immoral than not informing the police precisely who is at fault (a harmful omission) (Spranca, Minsk & Baron, 1991). However, unlike most of these previous demonstrations, our findings directly compare the presence of decision-making biases across moral and non-moral contexts (also see Cushman & Young, 2011).
In our first experiment, we also found that the sunk cost bias may be stronger in moral decision-making than in other situations. This is surprising. In non-moral cases proceeding with a futile course of action is wasteful. But in our moral version of the scenarios, proceeding is wasteful, harmful to others, and morally wrong. Yet, there was a greater discrepancy between willingness to act in response to sunk costs in the immoral condition. Increasing the reasons to not proceed with the action amplified the sunk cost bias. One potential explanation for this is that people are unwilling to admit their prior investments were in vain (Brockner, 1992). People succumb to the sunk cost bias in part because they feel a need to justify their past decisions as correct (Ku, 2008; also see Staw, 1976). Likewise, moral judgments seem to generate a much greater need to provide reasons to justify past decisions (Haidt, 2012). Thus, those making decisions in an immoral context might have additional pressures to justify their previous choice that stem from the nature of moral judgment itself.
Another explanation is that the initial investment was of a larger magnitude in the immoral compared to the non-moral condition. In both cases, participants incurred an economic cost, but only in one did participants incur an additional moral cost. People are more likely to succumb to the sunk cost bias when initial investments are large (Arkes & Ayton, 1999; Arkes & Blumer, 1985; Sweis et al., 2018). Perhaps sunk costs exerted a greater effect in the immoral condition because the past investments were greater (i.e., of two kinds: economic and moral, rather than just one: economic). However, as we do not know if the economic resources (e.g., pine trees and lab monkeys) were of comparable value, the discrepancy between moral conditions may entirely stem from the lab monkeys being valued higher and thus larger in investment magnitude. Thus, we are hesitant to draw any strong conclusion from this finding. The difference in sunk cost magnitude could stem from differences in financial costs between the immoral and non-moral contexts.
Our finding that moral violations led to increased willingness to act is reminiscent of the “what the hell” effect, in which people who violate their diet then give up on it and continue to overindulge (Cochran & Tesser, 1996; Polivy, Herman & Deo, 2010). We see this as similar to persisting in an immoral course of action after costs have been sunk. After engaging in a morally equivocal act, people may feel disinhibited and willing to continue the act even when its immorality becomes clear. Likewise, people may persist in an attempt to maintain the status quo (Kahneman, Knetsch & Thaler, 1991; Samuelson & Zeckhauser, 1988). These accounts, though, may not explain why sunk costs changed people’s moral perceptions. One possibility is that this resulted from cognitive dissonance between people’s actions and their moral code (Aronson, 1969; Festinger, 1957; Harmon-Jones & Mills, 1999). For example, sacrificing monkeys to develop a cure may cause dissonance between not wanting to harm but having done so. To resolve this, people might change their moral perceptions, molding their moral code to fit their behavior.
We close by considering a broader implication of this work. The extension of decision biases to moral judgment has been previously construed as supporting domain-general accounts of morality that suggest moral judgment operates similarly to ordinary judgment (Osman & Wiegmann, 2017; Greene, 2015). This is because if morality is not unique, one could reasonably expect that a factor that affects ordinary judgment would likewise affect moral judgment. Thus, if information irrelevant to the decision at hand (e.g., past investments) influences whether we continue to bulldoze land to build a highway, so too should it influence the same bulldoze decision that requires confiscating the land. This is not conclusive however, and our findings could be interpreted to support domain-specific accounts instead (e.g., Mikhail, 2011). For instance, the sunk cost bias was demonstrably larger in moral judgments. Nevertheless, an interpretation of our results as evidence for a domain-general account of morality must explain how the varying effect of past investment on judgment is a difference in degree but not kind.
Subscribe to:
Posts (Atom)
Posts
