Scientific Theories and Artificial Intelligence

Artificial Intelligence presents an important paradigm shift for science. Science is tradition10 ally founded on theories and models, most often formalized with mathematical formulas 11 handcrafted by theoretical scientists and refined through experiments. Machine learning, 12 an important branch of modern Artificial Intelligence, focuses on learning from data. This 13 leads to a fundamentally different approach to model-building: we step back and focus on 14 the design of algorithms capable of building models from data, but the models themselves 15 are not designed by humans. This is even more true with deep learning, which requires 16 little engineering by hand and is responsible for many of Artificial Intelligence’s spectacular 17 successes [27]. In contrast to logic systems, knowledge from a deep learning model is dif18 ficult to understand, reuse, and may involve up to a billion parameters [8]. On the other 19 hand, probabilistic machine learning techniques such as deep learning offer an opportunity 20 to tackle large complex problems that are out of the reach of traditional theory-making. It 21 is possible that the more intuition-like [27] reasoning performed by deep learning systems 22 is mostly incompatible with the logic formalism of mathematics. Yet recent studies have 23 shown that deep learning can be useful to logic systems and vice versa. Success at unifying 24 different paradigms of Artificial Intelligence from logic to probability theory offers unique 25 opportunities to combine data-driven approaches with traditional theories. These advance26 ments are susceptible to impact significantly biological sciences, where dimensionality is high 27 and limit the investigation of traditional theories. 28 Science would greatly benefit from a unification of Artificial Intelligence with traditional 29 mathematical theories. Modern research at the intersection of logic, probability theory, and 30 fuzziness yielded rich representations increasingly capable of formalizing scientific knowledge. 31 Such formal corpus could both include hand-crafted theories from Einstein’s e = mc to the 32 Breeder’s equation [35], but also harness modern A.I. algorithms for testing and learning. 33 Comprehensive synthesis is difficult in fields like biology, which have not been reduced to 34 a small set of formulas. For example, while we have a good idea of the underlying forces 35 driving evolution, we struggle to build effective predictive models of molecular evolution 36