Peter D. Warp
Much past AI research in scientific reasoning has fallen within the paradigm of historical study: we research some historical instance of scientific reasoning, and create a computer program (and associated theory) that automates that reasoning. This historical paradigm of studying scientific reasoning is of little or no further use to AI; the field must focus on building programs that make truly novel scientific discoveries. The first reason is that the historical para~iigm is not convincing to those who doubt that computers can in fact make scientific discoveries. It is high time to prove that computers are capable of making novel scientific discoveries by exhibiting many examples of such discoveries. The second reason is that scientific-discovery researchers are in danger of being beaten to the gold by computational scientists: while we leisurely work on historical problems, they are already racing forward to make real discoveries.