A Reservoir Model of Explicit Human Intelligence

TL;DR

This paper proposes a reservoir network model inspired by human brain architecture, emphasizing the roles of offline world modeling and language in enabling rapid, flexible, and robust explicit human intelligence.

Contribution

It introduces a novel reservoir-based network architecture that incorporates offline modeling and language, offering a new perspective on the neural basis of explicit human intelligence.

Findings

The proposed model can learn rapidly and flexibly.

It suggests explicit intelligence relies on reservoir attractor networks.

The model integrates with implicit deep networks for comprehensive cognition.

Abstract

A fundamental feature of human intelligence is that we accumulate and transfer knowledge as a society and across generations. We describe here a network architecture for the human brain that may support this feature and suggest that two key innovations were the ability to consider an offline model of the world, and the use of language to record and communicate knowledge within this model. We propose that these two innovations, together with pre-existing mechanisms for associative learning, allowed us to develop a conceptually simple associative network that operates like a reservoir of attractors and can learn in a rapid, flexible, and robust manner. We hypothesize that explicit human intelligence is based primarily on this type of network, which works in conjunction with older and likely more complex deep networks that perform sensory, motor, and other implicit forms of processing.