Anthropomorphizing AI is an obstacle to understanding it and what it does and will do. What AI of the DeepMind kind already does is having thousands of bots playing each other in complex optimization tasks - this is entirely genderless - only possible if you know what the goal is and can measure (degrees of) success according to existing rules. This is the crux and the power. It will be machines effectively learning from and programming each other. The constraints that evolution has put on our human behaviours are gone, equally concepts of 'agency'/will power', 'aggressivity', etc. If you go on youtube you can find very interesting analyses by Chess Champions such as Anna Rudolph explaining how non-natural, non-human that all looks (it also means the end of chess/Go champions in future, these machines have already played more combinations than ever were recorded in history for Chess and for Go...). This allows the solution of problems unthinkable to solve in real time, it requires a *very large* corpus of data to mine. Truly fascinating and v scary because the technology development is so far ahead of any societal and ethical control and that is not (yet) built into their system. Here is a starting point. The amygdala serves essential functions for cognition, putting 'valences' on cognitive objects. etc., and long-term memory consolidation, things that machines either don't need or are solving in other ways. If you are interested in the machine learning behind this that is now empowered by the ubiquity of data, here is another starting point.

Some AI research concerns machine learning in a defined problem-space, like learning to play Go, but a lot is (even in its inspiration) about seeking to realise animal capacities mechanically. Many AI researchers and their philosophical commentators say that the next hurdles for AI are creativity and subjective experience (understanding the last in terms of the folk-psychology described phenomenal characteristics of mental states). Work, for instance, has modelled in IA neural networks for human mindreading and 'theory of mind'. I'm not sure we're thinking of the same thing by 'folk psychology'. I get the idea you're imagining something like, 'everyone outside of the village has it in for me', while I only mean e.g. 'I was worried you wouldn't get home in time for supper'. Do you really want, in how we talk to each other, 'I was worried you wouldn't get home in time for supper', to be replaced with '[an exhaustive account of my brain-states before you got home e.g. memories of your arriving late; memories and projections of the food getting burnt and going cold, memories of subsequent arguments; sensory inputs; arousal and attentional states; the circuitry for conditioned responses, their sharpening and inhibition; the activation of task-focused modules, time-measurement modules, etc. etc. etc.]--all specified scientifically in close-grained neurological detail? It seems unlikely. Unlikely we will ever talk like that, and unlikely people will see the need to.

You know the mass displaces water, but the crow might just think it a tool that raises the food to the level of her beak. ( )

So those of us who care were already told that AlphaZero taught itself to play Chess by playing lots of games against itself. And this book tells us the same thing OVER and OVER again.

What AlphaZero looked for while it was playing all those games, what information it saved and how it used that information to make its timely move choices is entirely missing. (Saying that the program likes open files, open diagonals, and well posted knights is virtually meaningless. We all like those things, but at what cost?)

This book is much more like any game collection that features a single chessmaster except that AlphaZero's opponent is (nearly?) always Stockfish. Some (many?) of the games have already been published and analyzed. The analysis in Game Changer is interesting and presented with clear diagrams. It's a chess book. But if you want real information about AI and neural networks I think you need to look elsewhere. ( )