Marie Devaine,Jean Daunizeau. Learning about and from others’ prudence, impatience or laziness: The computational bases of attitude alignment. PLOS Computational Biology,2017

by Amir-Hossein Tehrani Safa (*)

Interacting with others lets us learn about their subjective preferences, feelings, attitudes and desires. In Psychology, the ability to interpret others’ explicit behavior and infer their hidden mental states or attitudes is called Theory of Mind.

In their paper, Devaine and Daunizeau make a bold claim: that learning about others’ attitudes and desires automatically and implicitly (i.e. without us wanting to or realizing) triggers the alignment of our attitude towards them. As we learn about others, they suggest, the very process of learning makes us become more similar to them. For example, when people tell us about their favorite pop songs, examining what songs they introduce to us shapes our preferences and makes us more likely to like the same song. Social learning, their math implies, comes with the unintended consequence of changing the learner.  

They test this prediction by measuring how participants learn about another person’s highly subjective personality traits of prudence (how to deal with RISK?), patience (how to deal with DELAY?) and laziness (how to deal with EFFORT?). The experiment consisted of three phases (see Figure 1A). 

First, people chose between (low-cost/small reward) and (high cost/big reward) options (see Figure 1B). In various trials, the cost could be risk, delay or effort telling us about the participant’s attitude in prudence, patience and laziness, respectively.

Next, people tried to guess what another specific person would do in the same test, thus learning the other’s prudence, patience and laziness.

Finally, step 1 was repeated to see how the participant’s attitude may have changed. The difference between extracted attitudes in Decision phases 1 and 2 would indicate whether learning about others changed the participant’s own preferences.

Their model made two key predictions which were confirmed by the data: 

(1) False consensus bias: each participant’s prior belief about others should be close to their own attitude. In other words, in middle part of Figure 1A, people start guessing about others by reporting their own preferences but then slowly – and through feedback – converge to what the other person’s attitude really is. 

(2) Social influence: after learning the other’s preference, participants’ would align their attitudes with what they have learned from the other. People’s cost-benefit arbitrage should drift towards those of the one others’.

Cool things about this paper: 

• The key intuition comes from a very relatable (Aristotelian) everyday life observation that “knowing what other people like changes our attitude about what is good”.
• They embedded well known models from classical decision theory and behavioral economics into a Bayesian model of theory of mind to build a unified computational model that explains attitude learning and predicts alignment. 
• Although the paper is full of complicated and not-so-easy-to-understand mathematical formulae, which one has to say, is inevitable for a complicated computational paper, the structure of the paper is clear and the reader could achieve a holistic view of the model without having to fully understand the details. 

However, if you want to understand the details, you should make an effort and learn Variational Bayesian Analysis. Going in depth without learning the math is not possible. You can start by downloading the VBA freeware and running the simulations and follow the M-files. It will help you understand the authors’ mind.

(*) Amir Hossein Tehrani-Safa is doing his PhD at Institute for Cognitive and Brain Sciences, Shahid Beheshti University, Tehran IRAN