Learning One Abstract Bit at a Time Through Self-Invented Experiments Encoded as Neural Networks

TL;DR

This paper explores how artificial scientists can autonomously generate and evaluate experiments and hypotheses, leading to effective exploration and knowledge expansion through self-invented and thought experiments.

Contribution

It introduces a novel framework where neural networks generate and assess experiments and hypotheses, combining reinforcement learning and neural network weights for autonomous scientific inquiry.

Findings

Self-invented experiments promote effective exploration.

Neural network-generated thought experiments can become boring over time.

Bias towards simple, surprising experiments enhances discovery.

Abstract

There are two important things in science: (A) Finding answers to given questions, and (B) Coming up with good questions. Our artificial scientists not only learn to answer given questions, but also continually invent new questions, by proposing hypotheses to be verified or falsified through potentially complex and time-consuming experiments, including thought experiments akin to those of mathematicians. While an artificial scientist expands its knowledge, it remains biased towards the simplest, least costly experiments that still have surprising outcomes, until they become boring. We present an empirical analysis of the automatic generation of interesting experiments. In the first setting, we investigate self-invented experiments in a reinforcement-providing environment and show that they lead to effective exploration. In the second setting, pure thought experiments are implemented as the weights of recurrent neural networks generated by a neural experiment generator. Initially interesting thought experiments may become boring over time.