MIT Researchers Just Discovered an AI Mimicking the Brain on Its Own

Eric James Beyer

Interesting Engineering

Originally posted 18 DEC 21

Here is an excerpt:

In the wake of these successes, Martin began to wonder whether or not the same principle could be applied to higher-level cognitive functions like language processing. 

“I said, let’s just look at neural networks that are successful and see if they’re anything like the brain. My bet was that it would work, at least to some extent.”

To find out, Martin and colleagues compared data from 43 artificial neural network language models against fMRI and ECoG neural recordings taken while subjects listened to or read words as part of a text. The AI models the group surveyed covered all the major classes of available neural network approaches for language-based tasks. Some of them were more basic embedding models like GloVe, which clusters semantically similar words together in groups. Others, like the models known as GPT and BERT, were far more complex. These models are trained to predict the next word in a sequence or predict a missing word within a certain context, respectively. 

“The setup itself becomes quite simple,” Martin explains. “You just show the same stimuli to the models that you show to the subjects [...]. At the end of the day, you’re left with two matrices, and you test if those matrices are similar.”

And the results? 

“I think there are three-and-a-half major findings here,” Schrimpf says with a laugh. “I say ‘and a half’ because the last one we still don’t fully understand.”

Machine learning that mirrors the brain

The finding that sticks out to Martin most immediately is that some of the models predict neural data extremely well. In other words, regardless of how good a model was at performing a task, some of them appear to resemble the brain’s cognitive mechanics for language processing. Intriguingly, the team at MIT identified the GPT model variants as the most brain-like out of the group they looked at.

The info is here.

Social threat indirectly increases moral condemnation via thwarting fundamental social needs

Henderson, R.K., Schnall, S.

Sci Rep 11, 21709 (2021).

Abstract

Individuals who experience threats to their social needs may attempt to avert further harm by condemning wrongdoers more severely. Three pre-registered studies tested whether threatened social esteem is associated with increased moral condemnation. In Study 1 (N = 381) participants played a game in which they were socially included or excluded and then evaluated the actions of moral wrongdoers. We observed an indirect effect: Exclusion increased social needs-threat, which in turn increased moral condemnation. Study 2 (N = 428) was a direct replication, and also showed this indirect effect. Both studies demonstrated the effect across five moral foundations, and was most pronounced for harm violations. Study 3 (N = 102) examined dispositional concerns about social needs threat, namely social anxiety, and showed a positive correlation between this trait and moral judgments. Overall, results suggest threatened social standing is linked to moral condemnation, presumably because moral wrongdoers pose a further threat when one’s ability to cope is already compromised.

From the General Discussion

These findings indicating that social threat is associated with harsher moral judgments suggest that various threats to survival can influence assessments of moral wrongdoing. Indeed, it has been proposed that the reason social exclusion reliably results in negative emotions is because social disconnectedness has been detrimental throughout human societies. As we found in Studies 1 and 2 and consistent with prior research, even brief exclusion via a simulated computer game can thwart fundamental social needs. Taken together, these experimental and correlational findings suggest that an elevated sense of danger appears to fortify moral judgment, because when safety is compromised, wrongdoers represent yet another source of potential danger. As a consequence, vulnerable individuals may be motivated to condemn moral violations more harshly. Interestingly, the null finding for loneliness suggests that amplified moral condemnation is not associated with having no social connections in the first place, but rather, with the existence or prospect of social threat. Relatedly, prior research has shown that greater cortisol release is associated with social anxiety but not with loneliness indicating that the body’s stress response does not react to loneliness in the same way as it does to social threat.

The research is here.

The effect of gender and parenting daughters on judgments of morally controversial companies

Niszczota P, Białek M (2021)

PLoS ONE 16(12): e0260503.

Abstract

Earlier findings suggest that men with daughters make judgments and decisions somewhat in line with those made by women. In this paper, we attempt to extend those findings, by testing how gender and parenting daughters affect judgments of the appropriateness of investing in and working for morally controversial companies (“sin stocks”). To do so, in Study 1 (N = 634) we investigate whether women judge the prospect of investing in sin stocks more harshly than men do, and test the hypothesis that men with daughters judge such investments less favorably than other men. In Study 2 (N = 782), we investigate the willingness to work in morally controversial companies at a significant wage premium. Results show that—for men—parenting daughters yields harsher evaluations of sin stocks, but no evidence that it lowers the propensity to work in such companies. This contrasts to the effect of gender: women reliably judge both investment and employment in morally controversial companies more harshly than men do. We suggest that an aversion towards morally controversial companies might be a partial determinant of the gender gap in wages.

From the Discussion section

There are several insights from our work. Firstly, we investigate laypeople instead of people of high social status, such as CEOs, members of congress, or judges. This would be consequential if parental investment in sons and daughters might depend on the social status of the parent. Studying laypeople makes our findings more relevant to the general population, and to more common decisions (e.g., concerning what mutual funds to invest in). Secondly, our models are aimed at directly testing whether the effect of parenting daughters is different across men and women. This would be expected from the female socialization hypothesis: parenting daughters might make the preferences of men more similar to those exhibited by women, as it would help them adopt alternative perspectives on issues in which the opinions of men and women might differ. Yet, they would not cause a shift in the preferences of women, as they have the same gender as their daughters. Our findings show that parenting daughters leads to harsher evaluations of morally controversial investments, but only in men. In fact, women parenting a daughter judge morally controversial investments more favorably than women without daughters, a somewhat unexpected finding.

Our results showed a boundary condition of the daughter effect. In our case, a full conceptual replication of the findings of Cronqvist and Yu would translate into a more negative view of morally controversial companies as investment propositions, and a lower willingness to be employed in such companies (at a significant premium). We observed the daughter effect in the former, but not in the latter decision. This is noteworthy, considering that the gender effect was of similar strength in Study 1 (that concerned investment) and Study 2 (that concerned employment). In short, gender differences are robust to the factors that affect the daughter effect, but these are yet to be discovered. We need to point out that we are not the first to show no clear support for the daughter effect; however, see for a methodological comment on that particular finding). Moreover, in one study, Dahl and colleagues showed that the birth of a child (even daughters, if the first-born child was not female) makes male CEOs less generous to employees.

The research is here.

What Dilemma? Moral Evaluation Shapes Factual Belief

B. Lui, & P. Ditto

Social Psychological and Personality Science. 2013;4(3):316-323. doi:10.1177/1948550612456045

Abstract

Moral dilemmas—like the “trolley problem” or real-world examples like capital punishment—result from a conflict between consequentialist and deontological intuitions (i.e., whether ends justify means). The authors contend that people often resolve such moral conflict by aligning factual beliefs about consequences of acts with evaluations of the act’s inherent morality (i.e., morality independent of its consequences). In both artificial (Study 1) and real-world (Study 2) dilemmas, the more an act was deemed inherently immoral, the more it was seen as unlikely to produce beneficial consequences and likely to involve harmful costs. Coherence between moral evaluations and factual beliefs increased with greater moral conviction, self-proclaimed topical knowledge, and political conservatism (Study 2). Reading essays about the inherent morality or immorality of capital punishment (Study 3) changed beliefs about its costs and benefits, even though no information about consequences was supplied. Implications for moral reasoning and political conflict are discussed.

From the General Discussion

While individuals can and do appeal to principle in some cases to support their moral positions, we argue that this is a difficult stance psychologically because it conflicts with well-rehearsed economic intuitions urging that the most rational course of action is the one that produces the most favorable cost–benefit ratio. Our research suggests that people resolve such dilemmas by bringing cost–benefit beliefs into line with moral evaluations, such that the right course of action morally becomes the right course of action practically as well.Study 3 provides experimental confirmation of a pattern implied by both our own and others’ correlational research(e.g., Kahan, 2010): People shape their descriptive understand-ing of the world to fit their prescriptive understanding of it.

A pdf of the research is here.

A Little Good Goes an Unexpectedly Long Way: Underestimating the Positive Impact of Kindness on Recipients

Kumar, A., & Epley, N. 

(2021, November 15). 

https://doi.org/10.31234/osf.io/3yast

Abstract

Performing random acts of kindness increases happiness in both givers and receivers, but we find that givers systematically undervalue their positive impact on recipients. In both field and laboratory settings (Experiments 1a-2b), those performing a random act of kindness predicted how positive recipients would feel and recipients reported how they actually felt. From giving away a cup of hot chocolate in a park to giving away a gift in the lab, those performing a random act of kindness consistently underestimated how positive their recipients would feel, thinking their act was of less value than recipients perceived it to be. Givers’ miscalibrated expectations are driven partly by an egocentric bias in evaluations of the act itself (Experiment 3). Whereas recipients’ positive reactions are enhanced by the warmth conveyed in a kind act, givers’ expectations are relatively insensitive to the warmth conveyed in their action. Underestimating the positive impact of a random act of kindness also leads givers to underestimate the behavioral consequences their prosociality will produce in recipients through indirect reciprocity (Experiment 4). We suggest that givers’ miscalibrated expectations matter because they can create a barrier to engaging in prosocial actions more often in everyday life (Experiments 5a-5b), to the detriment of people’s own wellbeing, to others’ wellbeing, and to civil society.

General Discussion

Prosocial actions, such as performing random acts of kindness, tend to improve wellbeing for both those who perform prosocial acts as well as for those who receive them.  Indeed, those who performed a random act of kindness in our experiments reported feeling significantly more positive than they normally do, and two of the experiments confirmed that performers felt better than participants who were not given the opportunity to perform a random act of kindness. Another found that performers of acts of kindness felt more positive after being kind than they reported feeling at the beginning of the experiment.Being more prosocial did not come at a cost to people’s own wellbeing; it enhanced it.  Daily life, however, affords many opportunities for engaging in these sorts of prosocial activities that people do not seem to take.  We believe our research suggests one possible reason why: that those performing random acts of kindness undervalue the positive impact they are having on recipients.  People’s choices are often guided by either an implicit or explicit calculation of expected value (Becker, 1993).  Underestimating how positive a recipient would feel after even a small act of kindness could lead people to engage in prosocial actions less often than might be optimal for both their own and others’ wellbeing.

Across many different actions, with many different participants, and in many different contexts, performers systematically perceived their random act of kindness to be a more minor action than recipients perceived it to be, and also systematically underestimated how positive recipients would feel afterwards.Performers were not confused, of course, that recipients would feel good about their experience.  In all cases performers expected recipients to feel more positive than they normally do.  Nevertheless, performers were still systematically miscalibrated as recipients felt even better than expected.

The research is here.

Beyond Populism: The Psychology of Status-Seeking and Extreme Political Discontent

Petersen, M., Osmundsen, M., & Bor, A. 

(2020, July 8).

https://doi.org/10.31234/osf.io/puqzs

Abstract

Modern democracies are currently experiencing destabilizing events including the emergence of demagogic leaders, the onset of street riots, circulation of misinformation and extremely hostile political engagements on social media. Some of the forms of discontent are commonly argued to be related to populism. In this chapter, however, we argue that the evolved psychology of status-seeking lies at the core of this syndrome of extreme political discontent. Thus, social status constitutes one of the key adaptive resources for any human, as it induces deference from others in conflicts of interest. Prior research has identified two routes to status: Privilege acquired through service and dominance acquired through coercion. We argue that extreme political discontent involves behaviors aimed at dominance through engagement in either individual aggression or in mobilization processes that facilitate coalitional aggression. Consistent with this, we empirically demonstrate that measures of status-seeking via dominance correlate with indices of a large number of extreme forms of political discontent and do so more strongly than a measure of populism. Finally, we argue that the reason why dominance strategies become activated in the context of modern democratic politics is that increased inequality activates heightened needs for status and, under such conditions, dominance for some groups constitutes a more attainable route to status than prestige.

Towards depolarized societies 

Understanding the psychological and structural roots of extreme discontent is key if we are to move  towards more peaceful societies.  An exclusive focus on populism might lead to the expectation that the  roots of discontent are value-based. For example, the rise of right-wing populism may suggest that  frustrations are rooted in a decreasing respect for authorities and traditional forms of life. If that was indeed the case, a depolarized society might be reached only if non-populists were willing to compromise on important political values and to a larger extent embrace tradition and authority.

In contrast, the present arguments and results suggest that the true roots of the most extreme forms of discontent are less based on a conflict of abstract political values and more on a lack of social status and recognition. If so, the path towards depolarization lies in more inclusion and more equality, for example, based on an affirmation of the classical liberal doctrine of the importance of  open,  non-dominant  exchange  of  arguments  (Popper,  1945).  Unfortunately,  this  is  not something  that can be fixed quickly, as would be  the case  if discontent was rooted  in transient factors such as the behavior of social media algorithms.  Rather, depolarization requires difficult structural changes that alleviates the onset of dominance motivations.

The research is here.

Hidden wisdom or pseudo-profound bullshit? The effect of speaker admirability

Kara-Yakoubian, et al.

(2021, October 28).

https://doi.org/10.31234/osf.io/tpnkw

Abstract

How do people reason in response to ambiguous messages shared by admirable individuals? Using behavioral markers and self-report questionnaires, in two experiments (N = 571) we examined the influence of speakers’ admirability on meaning-seeking and wise reasoning in response to pseudo-profound bullshit. In both studies, statements that sounded superficially impressive but lacked intent to communicate meaning generated meaning-seeking, but only when delivered by high admirability speakers (e.g., the Dalai Lama) as compared to low admirability speakers (e.g., Kim Kardashian). The effect of speakers’ admirability on meaning-seeking was unique to pseudo-profound bullshit statements and was absent for mundane (Study 1) and motivational (Study 2) statements. In Study 2, participants also engaged in wiser reasoning for pseudo-profound bullshit (vs. motivational) statements and did more so when speakers were high in admirability. These effects occurred independently of the amount of time spent on statements or the complexity of participants’ reflections. It appears that pseudo-profound bullshit can promote epistemic reflection and certain aspects of wisdom, when associated with an admirable speaker.

From the General Discussion

Pseudo-profound language represents a type of misinformation (Čavojová et al., 2019b; Littrell et al., 2021; Pennycook & Rand, 2019a) where ambiguity reigns. Our findings suggest that source admirability could play an important role in the cognitive processing of ambiguous misinformation, including fake news (Pennycook & Rand, 2020) and euphemistic language (Walker et al., 2021). For instance, in the case of fake news, people may be more inclined to engage in epistemic reflection if the source of an article is highly admirable. However, we also observed that statements from high (vs. low) admirability sources were judged as more profound and were better liked. Extended to misinformation, a combination of greater perceived profundity, liking, and acquired meaning could potentially facilitate the sharing of ambiguous fake news content throughout social networks. Increased reflective thinking (as measured by the CRT) has also been linked to greater discernment on social media, with individuals who score higher on the CRT being less likely to believe fake news stories and share this type of content (Mosleh et al., 2021; Pennycook & Rand, 2019a). Perhaps, people might engage in more epistemic reflection if the source of an article is highly admirable, which may in turn predict a decrease in the sharing behaviour of fake news. Similarly, people may be more inclined to engage in epistemic reflection for euphemistic language, such as the term “enhanced interrogation” used in replacement of “torture,” and conclude that this type of language means something other than what it refers to, if used by a more admirable (compared to a less admirable) individual.

A copy of the research is here.

Are some cultures more mind-minded in their moral judgements than others?

Barrett HC, Saxe RR. (2021)

Phil. Trans. R. Soc. B 376: 20200288.

https://doi.org/10.1098/rstb.2020.0288

Abstract

Cross-cultural research on moral reasoning has brought to the fore the question of whether moral judgements always turn on inferences about the mental states of others. Formal legal systems for assigning blame and punishment typically make fine-grained distinctions about mental states, as illustrated by the concept of mens rea, and experimental studies in the USA and elsewhere suggest everyday moral judgements also make use of such distinctions. On the other hand, anthropologists have suggested that some societies have a morality that is disregarding of mental states, and have marshalled ethnographic and experimental evidence in support of this claim. Here, we argue against the claim that some societies are simply less ‘mind-minded’ than others about morality. In place of this cultural main effects hypothesis about the role of mindreading in morality, we propose a contextual variability view in which the role of mental states in moral judgement depends on the context and the reasons for judgement. On this view, which mental states are or are not relevant for a judgement is context-specific, and what appear to be cultural main effects are better explained by culture-by-context interactions.

From the Summing Up section

Our critique of CME theories, we think, is likely to apply to many domains, not just moral judgement. Dimensions of cultural difference such as the ‘collectivist/individualist’ dimension may capture some small main effects of cultural difference, but we suspect that collectivism/individualism is a parameter that can be flipped contextually within societies to a much greater degree than it varies as a main effect across societies. We may be collectivists within families, for example, but individualists at work. Similarly, we suggest that everywhere there are contexts in which one’s mental states may be deemed morally irrelevant and others where they are not. Such judgements vary not just across contexts, but across individuals and time.

What we argue against, then, is thinking of mindreading as a resource that is scarce in some places and plentiful in others. Instead, we should think about it as a resource that is available everywhere, and whose use in moral judgement depends on a multiplicity of factors, including social norms but also, importantly, the reasons for which people are making judgements.

A pdf is here.

Sequential decision-making impacts moral judgment: How iterative dilemmas can expand our perspective on sacrificial harm

D.H. Bostyn and A.Roets

Journal of Experimental Social Psychology

Volume 98, January 2022, 104244

Abstract

When are sacrificial harms morally appropriate? Traditionally, research within moral psychology has investigated this issue by asking participants to render moral judgments on batteries of single-shot, sacrificial dilemmas. Each of these dilemmas has its own set of targets and describes a situation independent from those described in the other dilemmas. Every decision that participants are asked to make thus takes place within its own, separate moral universe. As a result, people's moral judgments can only be influenced by what happens within that specific dilemma situation. This research methodology ignores that moral judgments are interdependent and that people might try to balance multiple moral concerns across multiple decisions. In the present series of studies we present participants with iterative versions of sacrificial dilemmas that involve the same set of targets across multiple iterations. Using this novel approach, and across five preregistered studies (total n = 1890), we provide clear evidence that a) responding to dilemmas in a sequential, iterative manner impacts the type of moral judgments that participants favor and b) that participants' moral judgments are not only motivated by the desire to refrain from harming others (usually labelled as deontological judgment), or a desire to minimize harms (utilitarian judgment), but also by a desire to spread out harm across all possible targets.

Highlights

• Research on sacrificial harm usually asks participants to judge single-shot dilemmas.

• We investigate sacrificial moral dilemma judgment in an iterative context.

• Sequential decision making impacts moral preferences.

• Many participants express a non-utilitarian concern for the overall spread of harm.

Moral deliberation in iterative contexts

The iterative lens we have adopted prompts some intriguing questions about the nature of moral deliberation in the context of sacrificial harm. Existing theoretical models on sacrificial harm can be described as ‘competition models’ (for instance, Conway & Gawronski, 2013; Gawronski et al., 2017; Greene et al., 2001, 2004; Hennig & Hütter, 2020). These models argue that opposing psychological processes compete to deliver a specific moral judgment and that the process that wins out, will determine the nature of that moral judgment. As such, these models presume that the goal of moral deliberation is about deciding whether to refrain from harm or minimize harm in a mutually exclusive manner. Even if participants are tempted by both options, eventually, their judgment settles wholly on one or the other. This is sensible in the context of non-iterative dilemmas in which outcomes hinge on a single decision but is it equally sensible in iterative contexts?

Consider the results of Study 4. In this study, we asked (a subset of) participants how many shocks they would divert out of a total six shocks. Interestingly, 32% of these participants decided to divert a single shock out of the six (See Fig. 6), thus shocking the individual once, and the group five times. How should such a decision be interpreted? These participants did not fully refrain from harming others, nor did they fully minimize harm, nor did they spread harm in the most balanced of ways.  Responses like this seem to straddle different moral concerns. While future research will need to corroborate these findings, we suggest that responses like this, i.e. responses that seem to straddle multiple moral concerns, cannot be explained by competition models but necessitate theoretical models that explicitly take into account that participants might strive to strike a (idiosyncratic) pluralistic balance between multiple moral concerns. 

The research is here.