Skip to main content

Cognitive and Social Foundations of Rationality

Final Report Summary - RATIONALITY (Cognitive and Social Foundations of Rationality)

People are rational, social beings; yet we are biological machines of astonishing complexity. A fundamental scientific question is how rational thought and social interaction can emerges from what is ultimately a mechanistic system. This project "Cognitive and social foundations of rationality" aims to contribute to this question by exploring the interface between natural scientific and social scientific explanation of human behaviour. From the point of view of the natural sciences, the brain is a biological computing machine, subject to specific information processing principles and limitations. From the point of view and social sciences, behaviour, including communication, is understood in terms reasoned thought and action. Understanding the complementarity and conflict between these different styles explanation is likely to be of crucial significance in providing a scientific understanding of the richness of human thought and behaviour. The project is organized into four distinct but interconnected themes: inference, learning, decision and coordination.

1. Inference. This theme explored the hypothesis that the computing machinery of the mind and the model as carrying out Bayesian probabilistic reasoning. A particular theme has been the question of how the mechanistic constraints of the brain, and in particular its highly parallel processing architecture, are compatible with approximating Bayesian inference using sampling methods. If we view the brain as a Bayesian sampler, well-known departures from standard rational norms can be viewed as an inevitable by-product of the sampling process.

2. Learning. This theme aims to exploit two key ideas. The first concerns the ‘social’ nature of many aspects of learning. Specifically, when we learn linguistic patterns or cultural norms, we must learn to conform with patterns which have previously been generated by other learners: we have, in short, to learn to walk in each other's footsteps. This observation radically simplifies the learning problem, to the extent that generations of learners share a common set of cognitive biases. Roughly, the best guesses of a new learner will tend to be the right guesses, because the standard of correctness is set by the guesses of previous generations of learners. The second key idea is that learning problems that have sometimes seemed insuperable on purely “logical” grounds may be usefully addressed by applying recent advances in computational learning theory and machine learning.

3. Decision. Learning and inference help us build models of the physical, social and linguistic environment. How, then, does the brain use such models to inform decisions concerning how to act? The main focus of this theme is that decisions are made in a piecemeal, local way---sampling and applying relatively small amounts of relevant information.

4. Coordination. Human social behaviour involves astonishingly rich coordination between participants: each participant can successfully make their contribution only in the light of the expected contribution of the other—as if we form “momentary teams” to coordinate our behaviour. I suggest that such “teams” can be constructed through the mechanism of “virtual bargaining,” a new mechanism for social reasoning, captured in a mathematical theory,

Finally, many of the results of the project are synthesized in a single-authored book, The Mind is Flat, which is scheduled to be published by Penguin Books in the spring of 2018. This book attempts to bring together and synthesise the results of the project, into a vision of rationality and mind which is aimed not only at specialists, but at a broad intellectual readership and even, to a degree, the general public.