Title: Computational Support for Scientific Discovery
- Speaker: Dr Oliver Ray (University of Bristol)
- Host: Ross King
- 19th March 2014 at 14:00 in Kilburn L.T. 1.4
This talk introduces a logic-based framework for formalising and automating the core scientific tasks of hypothesis generation and experiment selection. The proposed approach goes beyond recent work in the automation of scientific inference by non-monotonically revising incorrect hypotheses to better fit the observed data (and not merely extending partially incomplete hypotheses) and by meta-logically inferring experiments to expose observable differences among competing hypotheses (and not merely selecting naively generated experiments). The presentation will focus on the key motivations and intended impacts of the work through a longstanding collaboration with Ross King, who developed the "robot scientists" Adam and Eve. In particular, it will outline plans (as part of a brand new EPSRC project) to show how such robots can be made to operate in a truly "closed loop" fashion by giving them the ability to "learn from" their own mistakes (by correcting inconsistencies in their knowledge) and to "reason about" their own capabilities (by exploiting changes in their hardware). I will suggest how both features could significantly enhance the cost-effectiveness of robot scientists and I will briefly discuss how they can be mechanised logically. The potential benefits will be illustrated using a functional genomics domain on which we intend to validate our approach.