Groups of interacting individuals are often found to have an
advantage over individuals in contexts of complex problem
solving. We suggest that social interaction allows group
members to share diverse introspections, perspectives and
strategies, promoting the formation of more abstract
problem representations, which – in turn – apply more
flexibly to new problem contexts. In a reinforcement
learning task inspired by the Wisconsin Card Sorting Test
(WCST), participants categorized aliens as friendly or
dangerous based on an underlying rule specifying feature
combinations. After a number of correctly categorized
trials, the rule would change (without explicit notification).
Participants could solve the task by learning every new
rule, but could also discover an underlying abstract rule,
which would facilitate faster recovery from local rule
changes. We compared pairs of participants individually
trained on different rules (diversity pairs), with pairs trained
on the same rule (non-diversity pairs), and individuals. We
found that diversity pairs outperformed non- diverse pairs
and individuals. Our findings suggest that diversity in prior
experience benefits groups, likely due to processes of
abstraction and cognitive flexibility.