Gary Marcus

Watson. The only commonality between Watson and AlphaGo is the AI moniker, so it feels like an odd comparison

This is incorrect in several ways though. First, Watson seems to have been an everything and the kitchen sink approach. They used logistic regression, rule based, lexical databases, information retrieval, grammar based parsers, SVM based relation extractors etc.

Part of the etc. is their heavy use of simulations to fit parameters for their strategy modules. I'll have to recheck but I recall the use of bayesian methods, reinforcement learning, Neural networks and monte carlo search in the game strategies paper: https://ieeexplore.ieee.org/document/6177733 …

I'm sure they did lots of things to make it work. It's still the most symbolic and least DRL system coming out of any modern research lab. It's quite a bit closer to the Gary's cognitive hybrid systems approach than anything out of DeepMind.

hype aside, alphago is a hybrid, with a monte carlo tree sim backbone that traversed trees out of symbol CS 101, alongside the DRL.

That's an odd argument to make since you also claim DeepMind relies too much on DRL.

Both can be & are true. Really telling to me is that DRL on its own worked for Atari games but not Go— and that DM’s spin on Go really downplayed the hybrid aspect that was essential to its success. (It was also apparently necessary to build in the rules for Go, unlike Atari.)

So we build hybrid systems but are also too reliant on DRL? How can both of those be true?