Milky Way (reward learning)

While you were spending your time in the outer space milk market discovering which cows produce more milk, we were able to study how you learn from feedback and make decisions.

What are we studying?

We learn something new every day, such as trying a new recipe or taking a walk along a new route. Along the way, we learn which of our actions lead to positive outcomes (e.g., discovering a new park that you really enjoy) and which lead to negative results (e.g., the new recipe not being tasty). Based on these experiences, we repeat the things we find rewarding and avoid the things that lead negative outcomes. However, life is constantly changing and sometimes actions that led to positive outcomes in the past may not be as rewarding anymore (e.g., you grew bored of your favourite TV series), which will make you re-evaluate your choices and pursue those things that make you happier.

In the Milky Way task, you had to track which cow was producing more milk and choose the one that was most promising. To be good at this game, you had to be alert and know when the outcomes changed and when your favourite cow is no longer producing enough milk.

Neuroscience has shown that the brain processes that are responsible for this learning can be very well captured using mathematical models from artificial intelligence. These mathematical models are similar to the ones used to train robots and they help us to understand how you learn about what is good and what is not, and link it to the brain processes. For exemple, it has been shown that imprecision in our learning process (e.g. how “noisy” we are when extracting information from the environment) can significantly influence our decision-making and push us to explore new options.

Why are we interested in that?

Some people struggle more than others to learn new information. In this study, we would like to explore whether we can identify specific subgroups, such as those suffering from mental health problems, who struggle with learning and making choices that benefit them most.

Further reading

Hauser TU, Will GJ, Dubois M & Dolan RJ (2019). Developmental Computational Psychiatry. J Child Psychology & Psychiatry 60(4): 412-426

Charles Findling, Vasilisa Skvortsova, Rémi Dromnelle, Stefano Palminteri & Valentin Wyart (2019). Computational noise in reward-guided learning drives behavioral variability in volatile environments. Nature Neuroscience 22,  2066–2077