Like humans, an #AI was better at #logic when the tasks were about 'realistic" (rather than more abstract or "arbitrary") stuff.

Ishita Dasgupta shared these results from their team at #DeepMind.

A glimpse of a new human-vs-AI #comparativePsychology?
• This context effect emerged from a monolithic system. What does this say about #DualProcess theories/debates?
• How might this result in "new levels of analysis"?
similar "behavior" ‹ similar "representations" ‹ similar "learning"

#rationality