Artificial Intelligence and Scientific Creativity
Simon Colton · 2002
There has been much recent success for AI systems undertaking creative tasks in scientific domains such as astronomy, biology, medicine, chemistry, physics and mathematics. In many scientific domains, we can build on the wealth of philosophical and computational studies into creative aspects of human intelligence, and use the abstract nature of the data to derive specialist algorithms for discovery. To achieve high level scientific creativity, the computational techniques employed are often domain specific. However, there are aspects of scientific creativity that can be identified and applied across domains. The process of developing the core notions of machine discovery in science is underway, as emphasised by, amongst others, the 1995 AAAI Spring Symposium dedicated to scientific discovery, the machine discovery workshops at ICDC’98 and ECAI’98, the 1997 issue of Artificial Intelligence dedicated to scientific discovery (vol. 91, issue 2), and the forthcoming machine discovery issue of the International Journal of Human-Computer Studies. This process was continued at the 1999 AISB Symposium on AI and Scientific Creativity, which took place in Edinburgh, Scotland, in April. Papers presented at the symposium addressed the theoretical aspects of and computational possibilities for machine creativity. They also reported on systems implemented to achieve automated discovery in science. The intention of the symposium was that that the papers proposing models of scientific creativity would help researchers concerned with implementing discovery programs, and the papers discussing the successes and techniques employed in working systems will help researchers extract general frameworks for scientific machine discovery. This note is a survey of current research on creativity in science, and in particular the automation of discovery tasks in science. While it provides an opportunity to detail the papers presented at the symposium, we endeavour to extract commonalities between the research discussed and to present the work in a wider context. To write creative programs which perform discovery tasks in science, it is important to understand the frameworks for creativity in science. These include philosophical frameworks, as discussed in section 2, which can draw on case histories and psychological studies, and computational frameworks available for machine discovery programs, which are discussed in section 3. Also, it is vital to learn from implementations of discovery programs, as discussed in section 4. Understanding the aims, the techniques employed and results achieved for a particular discovery program will enable new implementations to build on, extend and improve the work of others and will lead to more successful creative programs in science. One of the goals of the machine discovery community is to increase the quality and quantity of creative programs which act as assistants to scientists. It is hoped that the AISB Symposium and this survey will help us to take a step towards this goal.