It’s all too easy to claim that children don’t need any innate priors or assumptions or representations, only to sneak them back in when it comes to get serious

It’s all too easy to claim that children don’t need any innate priors or assumptions or representations, only to sneak them back in when it comes to get serious and implement a model. That was the trick in a lot of the neural-network models of language that were popular in the 80s and 90s – when the rubber met the road, they always built in innate structure without calling attention to it. From the outside, seems like there are still similarities wrt ML and current AI.