Biology as a scientific and research domain seems to have undergone major breakthroughs and paradigm shifts every time external disciplines have intersected with it. Typically, after a given time of somewhat uneasy coexistence, the community embraces the new conceptual frame and its associated technologies as a lens through which biological phenomena can be (re)interpreted. The happy encounter between biology and chemistry gave birth to biochemistry, enzymology, and metabolism. Much later, the interest of post-war physicists for live systems brought about the onset of molecular biology, which reached its biggest milestones in the elucidation of the DNA double helix and the deciphering of the genetic code. During the many decades dominated by molecular biology and molecular genetics, the emphasis has been on mechanistic understanding of biological phenomena, enabled by rigorous hypothesis-driven approaches imported from physics and formal mathematical logic. These departed from mere trial-and-error approaches that prevailed in previous stages and have enabled all-rational understanding of many key biological processes on the same principles that govern the rest of the material world. This is, after all, the ultimate mission of science as a human endeavour: rational understanding of reality with universal laws and principles.

There are at least 3 intertwined qualities of life that can make full comprehension of living systems ultimately unreachable. First, they grow, mutate, and evolve. This means that one live object we inspect at a given moment in space and time will not be identical to the same a moment later, let alone if we perturb it to sample or measure specific properties. Second, parameters associated to specific biological devices are context-dependent. One promoter in a location of the genome will show different input–output transfer functions when placed on another site or when cells are grown under different nutritional or environmental conditions. Since the number of such possible conditions is virtually infinite, so is the variability of the parameters at stake. Finally, biological matter is often soft, i.e., not hard in a physical sense. Instead, biology is about flexible materials, plastic shapes, glues, etc. Such components also change their form, which adds another degree of difficulty to making predictions with an accuracy remotely comparable to other branches of science and technology. This does not mean that Biology escapes the laws of physics, but that its ever-growing complexity makes its complete comprehension with standard approaches unreachable—even theoretically. This is the point where machine learning and artificial intelligence (ML/AI) come to the rescue.

ML is about adopting learning algorithms whose objective is to obtain a result (e.g., patterns, rules, correlations) dependent on the input variables (data) but without assuming any prefixed interplay among them. Through the use of computational and statistical methods, such algorithms are trained to make classifications and scores with an explicit degree of reliability. On this basis, it cannot come as a surprise that the large volume of data generated by omics technologies become an excellent input for training ML platforms. In turn, they can deliver dependable predictions on specific, yet complex, biological questions that are not yet amenable to mechanistic understanding.

Mechanistic understanding from first principles anticipates solar eclipses and the successful test of the first prototype of an atomic bomb with 100% accuracy. This is hardly attainable in Biology if the phenomenon at stake has many components. On the other hand, ML-based Alpha Fold enables prediction of complex protein structures with an average accuracy of >80%. True, it is not 100% but it is good enough for moving ahead in most cases. If we accept that we cannot follow all variables of a system—in particular if they change through time and space—resorting to patterns and probabilities might enable us to nevertheless solve practical problems without knowing the underlying basis.

This raises a deeper question about how we study the composition and evolution of biological systems. As physicists gave up on describing the exact position of particles beyond a certain subatomic scale, biologists may also have to give up mechanistic understanding of live systems beyond a given level of complexity in favour of a probabilistic, experience-based approach. In specific scales, mechanistic comprehension—and thus complete predictability—could be inherently impossible. Thus, efforts in that direction are likely to be ultimately futile. Yet, we can still handle description and even understanding of biological systems in terms of chances, scores, and so on.