An interesting idea; although I think somewhat unlikely. This is one of those instances where I would assume the brain is doing something carefully calibrated.

An interesting idea; although I think somewhat unlikely. This is one of those instances where I would assume the brain is doing something carefully calibrated. But it'd be easiest to just not record as many experiences for the first few years rather than do a complete flush. My guess has always been there is some optimal approach to learning that works by developing a really basic schema (what does a person look like, which ones are mine, roughly how does this body thing work, etc) then flushing all the training data and starting again. I vaguely expect the machine learning people develop some sort of similar process where they get a lot of value lightly conditioning a model before sinking compute into doing a full train. Basically, my wild guess is there is a lot to learn in a baby's first few experiences but the risk of mis-encoding the lessons is so high that the brain uses the data to bootstrap but then throws it out as too unreliable and starts again once it orients.