Innateness, AlphaZero, and Artificial Intelligence

By Gary F. Marcus. 2018

77 citations recorded by the source corpus.

The concept of innateness is rarely discussed in the context of artificial intelligence. When it is discussed, or hinted at, it is often the context of trying to reduce the amount of innate machinery in a given system. In this paper, I consider as a test case a recent series of papers by Silver et al (Silver et al., 2017a) on AlphaGo and its successors that have been presented as an argument that a "even in the most challenging of domains: it is possible to train to superhuman level, without human examples or guidance", "starting tabula rasa." I argue that these claims are overstated, for multiple reasons. I close by arguing that artificial intelligence needs greater attention to innateness, and I point to some proposals about what that innateness might look like.

Subfields: Artificial Intelligence, Computational Theory and Mathematics, Mechanical Engineering

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computational Theory and Mathematics

In Mechanical Engineering

Citations in the corpus

Relational inductive biases, deep learning, and graph networks (2018, methodology): “…learning about entities and relations (Mitchell, 1980), which we, joining many others (Spelke et al., 1992; Spelke and Kinzler, 2007; Marcus, 2001; Tenenbaum et al., 2011; Lake et al., 2017; Lake and Baroni, 2018; Marcus, 2018b), suggest are an essential ingredient for human-like intelligence.”
Revisiting the poverty of the stimulus: hierarchical generalization without a hierarchical bias in recurrent neural networks (2018, methodology): “This version of the argument is difﬁcult to assess using RNNs because, while RNNs must possess some biases (Mitchell, 1980; Marcus, 2018), the nature of these biases—which likely arise both from the network architecture and from the learning algorithm—is currently poorly understood.”
Sociocultural memory development research drives new directions in gadgetry science (2018, background): “I see a clear parallel with much recent work in artificial intelligence (including neural networks), which shares the book’s empiricist attitude and faith in the power of domain-general learning (Marcus, 2018; see also Lake et al., 2017).”
Why AI is harder than we think (2021, background): “One example is charades, which “requires acting skills, linguistic skills, and theory of mind” [43], abilities that are far beyond anything AI can accomplish today.”
On Inductive Biases in Deep Reinforcement Learning (2019, background): “Our work relates to a broader debate (Marcus, 2018) about priors and innateness.”
Neo: A Learned Query Optimizer (2019, background): “model’s architecture to create an inductive bias [33] suitable for query optimization: the structure of the neural network itself is designed to reflect an intuitive understanding of what causes query plans to be fast or slow.”
A Group-Theoretic Approach to Abstraction: Hierarchical, Interpretable, and Task-Free Clustering (2018, background): “However, almost all require handcrafted priors—the counterpart to innate biology (Marcus, 2018; Dietterich, 2018).”
Model-free, Model-based, and General Intelligence (2018, background): “Should the moving objects emerge from a general learning scheme, from prior knowledge, or from a combination of both [Mugan and Kuipers, 2012; Thomas et al., 2017; Marcus, 2018a]?”