Read on Twitter
Extremely proud to share our work "A brain network supporting social influences in human decision-making", now out in @ScienceAdvances! w/ @glascherlab
Ever wondered how direct learning & social learning interact in the same environment? A thread.
(1/15)
https://advances.sciencemag.org/content/6/34/eabb4159
Ever wondered how direct learning & social learning interact in the same environment? A thread.
(1/15)https://advances.sciencemag.org/content/6/34/eabb4159
We make decisions based on our own learning experience, but also learning from others, like at restaurants. But how do we learn from others to better inform our own learning? If they work in parallel, is social learning processed differently from direct learning? (2/15)
We used a classic probabilistic reversal learning task, but in a social context - we invited 5 people each time, and after each participant made their choice, they could observe what the others had chosen; next, everyone could make adjustments, followed by outcomes. (3/15)
Not surprisingly, we saw that the more opposite choices the more people switched, but they increased their confidence (bet) when seeing agreeing choices. On top of previous literature, it is interesting to see how group coherence playes a role here. (4/15)
But what are the possible computations? We asked ourselves, what could be the recipes for the model? Importantly, besides the behavior, there is instantaneous social information (before seeing outcome) and potentially social learning (after seeing outcome). (5/15)
So we put these elements altogether, built a comprehensive model. Option values were updated via both direct learning (counterfactual RL) & social learning (discounted cumulative reward history) to guide future choices. We used @mcmc_stan to estimate models hierarchically. (6/15)
Within each group, we scanned one person's brain with fMRI. This allowed us to measure when and where the brain carries out value computations from direct learning & social learning, and to characterize their potentially distinct neural signatures. (7/15)
We observed dissociable value signals: value from direct learning is only associated with vmPFC, yet value from social learning is only related to ACC. This is nicely compatible with a fab recent review on social learning by @EmotionLabKI etal. (8/15)
https://www.nature.com/articles/s41583-020-0276-4
https://www.nature.com/articles/s41583-020-0276-4
We also found the well-known reward prediction error (RPE) in VS/NAcc. We looked into RPE's subcomponents to justify that VS is encoding RPE, rather than outcome valence. We recently articulated this approach in a tutorial paper. (9/15)
https://academic.oup.com/scan/article/15/6/695/5864690
https://academic.oup.com/scan/article/15/6/695/5864690
Now that we have these brain regions, how do they interact? Could it be possible that the brain's reward-related regions covary with social-related regions? For this, we used simple connectivity analyses. (10/15)
PPI showed the putamen as the target region. But what computational signal does it carry out? We used the same approach for RPE, and found an interesting social prediction error representation. We actually encourage using this approach to justify any error-like signals. (11/15)
These findings, together with another connectivity analysis, indicate an integrated brain network supporting social influence in human decision-making. (12/15)
In case you are interested in reproducing some of the figures in the paper, I have made nearly everything available on @github #openscience. This is so far my most complete repo accompanying a paper. Pls do check it out! (13/15)
https://github.com/lei-zhang/SIT
https://github.com/lei-zhang/SIT
And if you have no time to read the paper (20 pages!), I recently gave an online talk on the paper, and it's on @YouTube. Also, @NatalieParletta wrote a lovely #scicomm piece on our paper in @CosmosMagazine. (14/15)
https://cosmosmagazine.com/health/body-and-mind/sometimes-we-need-to-learn-from-others/
https://cosmosmagazine.com/health/body-and-mind/sometimes-we-need-to-learn-from-others/
The first pilot of this study started in 2014 and the fMRI scanning started that winter. So almost 6 years! We are so grateful to everyone who helped at @SysNeuroHamburg. Memorable time indeed! Thank the reviewers, too!
Hope you will like the paper & enjoy reading!
Fin. (15/15)
Hope you will like the paper & enjoy reading!
Fin. (15/15)
