Modeling Scientific Discourse

We aim to build intelligent systems which can reason autonomously about the risk of carcinogenicity of chemicals, drawing on whatever theoretical or experimental evidence is available. In earlier work [14], reviewing the literature on methods of carcinogen risk assessment, we catalogued the different types of evidence adduced to support these claims, which may be in the form of: experimental results on tissue cultures, animals or human epidemiological studies; analytical comparisons with known carcinogens; or explication of biomedical causal pathways. Because the research which underpins conclusions in this domain is usually at the leadingedge of the scientific disciplines concerned, evidence from these different sources may be inconsistent or conflicting. Consequently, carcinogen risk assessment usually involves the comparison and resolution of multiple evidences and arguments for and against a particular scientific claim [23, 27]. To represent this domain in an intelligent system, therefore, we first require a philosophical model of scientific enquiry. Which philosophy of science is appropriate for such representation, and why? Next, having adopted such a conceptual model, we will need to formalize it. How can this be achieved? In particular, how may we represent the scientific uncertainty characteristic of knowledge in the carcinogen domain? These questions are the focus of this paper, which outlines our current thinking and approach.