AlphaGo is a hybrid system that uses deep learning for pattern recognition and Monte Carlo tree search, a symbolic operation, to search possibilities.
Resolved true The record shows it happened or was so.
Factual: Statements about the present or the past, checked against the record: did they hold up?
What would count as true or false
This is a factual claim judged as of 2019-11-14 against the published description of AlphaGo's architecture. It is true if AlphaGo (as described in Silver et al. 2016 and related DeepMind publications) combines deep neural networks (policy and value networks) with Monte Carlo tree search to explore possible moves. Whether MCTS counts as a 'symbolic' operation is Marcus's characterization. It should be accepted as true if MCTS works by explicit search over a tree of discrete game states, which is a classical (non-neural) search algorithm. Mostly true: the architecture is as described but some minor detail is overstated. Mixed: AlphaGo used MCTS, but it is widely and substantively argued that MCTS is not symbolic in the sense intended, so the 'hybrid' framing is contested. Mostly false: the description fits only some versions of the system, or misstates the role of the components. False: AlphaGo did not use tree search, or used no deep learning.
Set before the research began.
Evidence (4)
-
E1 Mastering the game of Go with deep neural networks and tree search - Nature Nature (Silver et al.) · 27 January 2016
“We also introduce a new search algorithm that combines Monte Carlo simulation with value and policy networks.”
-
E2 Mastering the game of Go with deep neural networks and tree search - Nature Nature (Silver et al.) · 27 January 2016
“Here we introduce a new approach to computer Go that uses ‘value networks’ to evaluate board positions and ‘policy networks’ to select moves.”
-
E3 AlphaGo Google DeepMind
“We created AlphaGo, an AI system that combines deep neural networks with advanced search algorithms.”
-
E4 Mastering the game of Go without human knowledge UCL Discovery (repository record for Silver et al., Nature 2017) · 19 October 2017
“The tree search in AlphaGo evaluated positions and selected moves using deep neural networks.”
Reasoning
The 2016 Nature paper by Silver et al. describes policy and value networks used to select moves and evaluate positions (E2). It also describes a search algorithm that combines Monte Carlo simulation with those networks (E1). The 2017 follow-up paper says AlphaGo's tree search used deep neural networks to evaluate positions and select moves (E4), and DeepMind's overview describes AlphaGo as combining deep neural networks with search algorithms (E3). The criteria say to accept Marcus's "symbolic" label if MCTS is an explicit search over a tree of game states; the cited sources describe AlphaGo as using a tree search that evaluates positions and selects moves, combining Monte Carlo simulation with value and policy networks [E1][E4], so the claim matches the published architecture.
Cite this claim
Gary Marcus, “AlphaGo is a hybrid system that uses deep learning for pattern recognition and Monte Carlo tree search, a symbolic operation, to search possibilities.” (said 14 November 2019). Verdict: resolved true. Reliability API, claim c_f9d0dbb18e39, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_f9d0dbb18e39
This claim is in dataset release v1.1.2, with the same verdict.
Reply to this claim
More factual claims by Gary Marcus
-
Mostly true
“Reviews by Joanna Stern and Marques Brownle of one early prototype were damning;”
Reviews by Joanna Stern and Marques Brownlee of one early humanoid robot prototype were damning.
-
Mostly true
“Many would agree we have reached a point of diminishing returns for scaling; faith in scaling as a route to AGI has dissipated.”
-
Mostly true
“Despite all the hype, agents didn’t turn out to be reliable.”
-
Mixed
“but the systems don’t have the knowledge graph to do that.”
-
Resolved true
“Hasn’t worked so far. That’s for sure.”
Bolting extra components onto an LLM and scaling it up has so far not produced human-level AGI.
Verdicts are made automatically; how verdicts are made.