The Unlikely Introduction of Algorithmic Prediction in Environmental Policy: Learning from the ToxCast Puzzle

• In Event: Precision and Uncertainty in a World of Data - II

Abstract

This paper looks at the development of an infrastructure for in vitro and in silico testing of chemical effects. The ToxCast programme was initiated by the US Environmental Protection Agency in 2005, consisting of the testing of several thousands of chemical substances through seven hundred assays, to produce multiple data points about the biological impacts of chemicals. The resulting, massive set of data points is supposed to help identity pathways of toxicity of chemicals in the body, as well as possible negative interactions between chemicals. An intended by-product of the program is the iformulation of rapid assays enabling to predict whether a new, as yet untested chemical, may activate one of these pathways or not - to decide what to do about the said chemical in regulatory terms. In many regards, the ToxCast program is a surprising innovation in the world of risk assessment: It is based on new concepts of risk (toxicity pathway); it overturns the dominance of conventional animal testing as the main standard of proof in the world of regulatory science; it involve the leverage of massive funds for the Office of Research and Development of the EPA, under financial pressure at the time, and this in anticipation of any concrete result; it involves an unlikely success in terms of organizing, coordinating a wide set of expertises over a long period of time, across several research and regulatory bodies. In all of these respects, and more, ToxCast embodies an unlikely innovation in the supposedly conservative world of regulatory knowledge. The paper mobilizes a set of interviews with initiators, leaders and current members of the program, to elucidate the conditions that determine this innovation in regulatory knowledge. It will test the « economy of technoscientific promisses » perspective, verifying what are the organizational and material infrastructure that produced and sustained the credibility of promissory predictive algorithms. It will, among other things, explore the hypothesis that what makes the design of ToxCast credible, was the importing, in environmental risk assessment, of bioinformatic technologies, expertises and modes of organization that were already being practiced, with good results, in the world of private pharmaceuticals development. It will thus explore the admittedly paradoxical hypothesis that the introduction of promissory, algorithmic predictive system in policy reflects past experience and learning capacities afforded by the public-private interactions that define the bioeconomy.

Author

• David Demortain, INRAE, LISIS