Automatic Generation of Scientific Theories

Creativity is at the heart of scientific research, both in the natural and the social sciences. It is also at the basis of innovation in business and industry. Psychological research suggests that the Darwinian mechanisms of variation and selection are the key constituents of creativity. Similarly, philosophers of science such as Popper have noted the evolutionary character of scientific knowledge. However, with the exception of preliminary work by the PI, there has been no attempt to use computers to evolve scientific theories using mechanisms based on natural evolution.

The aim of this research is to fill in this gap by proposing a novel way to generate theories. Theories will be automatically generated by computer programs using algorithms based on natural evolution. The success of the theories will be estimated by the extent to which they account for the experimental data, and the best theories will be selected for the next generation, possibly modified by mutation or crossover. The method will be applied to experiments in cognitive psychology, health psychology and behavioural economics.

Currently, the development of scientific theories is considered the exclusive province of humans. This project will genuinely be transformative as it will challenge this conception and change the way theories are viewed in the social sciences: from static abstract objects to concrete objects that can be manipulated by computer programs. The proposed method will lead to more efficient ways to exploit empirical data for the development of theoretical knowledge, with implications for applied research in the long term.

The current project is primarily a piece of basic rather than applied research, and its main beneficiaries, at least in the short and medium term, will be other academics. Our impact strategy will therefore emphasise academic channels, including a dedicated website, conference presentations and publications, and articles in peer-reviewed journals. In addition, a workshop on automatic generation of scientific theories will be organised at the University of Liverpool to disseminate our findings, promote our novel methodology and receive comments from colleagues. Finally, the implementation of our methodology as well as a sample of the models that it will develop will be made publicly available on our website. These online resources will encourage discussion, spur new developments in cognitive psychology, cognitive science, evolutionary computation and philosophy of science, and provide a forum for critical, comparative and interdisciplinary research.

Our research is also likely to have impact outside of academia, but only in the longer term, which makes it difficult to foresee the exact nature of this impact at this point in time. The better our understanding of natural and artificial phenomena, the better the applications that can be developed. As our methodology has the potential to develop better theories than those currently used and thus to make better use of empirical data, it should have impact for applications as well.

We are also aware that our inter-disciplinary research, due to its transformative and ambitious nature, will be of interest to the public at large. We will therefore ensure wider understanding of the issues involved through press releases in newspapers, public-oriented webpages, public lectures, radio and TV. Note that the applicants already have an excellent track record of publicizing their work to the general public, and their work has been described in national and international newspapers (e.g. The Times, The Financial Time, NRC-Handelsblad, La Nación, Die Zeit), magazines (e.g. Times Higher Education, Scientific American, New Scientist, Psychology Today), radio and TV (e.g. BBC, HBO, National Public Radio, Arte).