Chemoinformatics and (artificial) intelligence
Conférence organisée par le laboratoire SABNP.
Par D. HORVATH
Laboratoire de Chemoinformatique
Université de Strasbourg
Le 1er juillet à 14h
Bâtiment Maupertuis - salle 02 W34
3 rue du Père Jarlan - Evry
Non-expert users of chemoinformatics typically fall in one of two categories : believers - who trust chemoinformatics predictions as if they were the Delphi Oracle - and non-believers, who think that all this is just "artificial intelligence" (an euphemism for "crass stupidity").
Few have a pragmatic understanding of what chemoinformatics can and can not - albeit this should be rather obvious, as chemoinformatics is no more, nor less empirical than chemistry itself, and therefore bound to succeed or fail in exactly the same way in which the chemist’s own intuitions and predictions succeed and fail. There is a clear analogy between in silico algorithms and human reasoning (obviously, since - so far - humans are the ones "transferring" their way of reasoning to machines).
When a machine (or a scientist) fails, it’s because it/he/she did not get a proper training. Unfortunately, in many (if not most) cases the entire body of available data from which a given problem could be learned is utterly insuficient for "proper education" of machines (and humans) alike. But Columbus did not learn about America in primary school geography class, either. Forget marketing slogans : chemoinformatics is not "rational drug design" - it’s, much more modestly, "less random drug design". But that’s a good start.