Science has no certainties, but that is normal

Confidence in scientific knowledge follows a peculiar dynamic: it slowly consolidates, but crumbles every time a prediction is disregarded or an explanation appears contradictory. Today, the very concept of scientific proof seems somewhat weakened, suspended between the overproduction of data and the collective inability to interpret them critically. "We are going through a transitional phase in which the question is whether artificial intelligence is a mere tool or something that redefines the very boundaries of knowledge," explains Adam Kucharski, professor at the London School of Hygiene & Tropical Medicine and author of texts such as 'Proof: the uncertain science of certainty' (Profile Books) and 'The rules of contagion' (Marsilio).

Especially after the major health crises of recent years, from Ebola to Covid-19, uncertainty is no longer perceived as a natural condition of science, but rather as its inherent limitation. And this misunderstanding makes it difficult to accept the probabilistic nature of much scientific knowledge. "Exposure to uncertainty and randomness is wider than ever before, and the ability to explain why certain things happen seems to have diminished," Kucharski notes. "Uncertainty, far from being a flaw, is the engine of science. But when it comes to public decisions, especially in emergency situations, it turns into a political problem'.

In English, this is referred to asweak evidence, the condition in which one has only partial - and not conclusive - clues when making urgent choices: in contexts such as a pandemic or a judicial process, waiting for conclusive evidence may mean not acting in time. It is often not possible to wait for solid evidence, so decisions have to be made with the information available, balancing the risks and accepting the margin of error as an unavoidable component of decision-making. "Even when there is good certainty about the problem and the tools, there can still be disagreement about the policies to be adopted. This is why it is useful to approach uncertainty by treating it as a multi-layered process, aware that arriving at a single, granitic truth is often utopian,' Kucharski continues. This complexity becomes even more evident in long-term processes, such as climate change: despite the very broad scientific consensus on the seriousness of the phenomenon, policies remain fragmented and proceed in no particular order, not because of a lack of data but because of diverging interests, values and political lines.

The concept of uncertainty can extend beyond science, taking on a cultural dimension. "Throughout history, what was considered obvious has changed radically," he clarifies. "Europe, for example, long rejected negative numbers because they were incompatible with Greek geometry, while in Asia they were accepted because they were linked to the practice of debt." This relativism is exacerbated today by the increasing use of artificial intelligence: algorithms capable of predicting the evolution of a protein, or of writing texts in natural language, challenge the classical idea that every decision must be supported by an understandable explanation. "There are situations where AI works best if we don't worry too much about how it achieves its results," observes Kucharski. "There is an analogy with what happens with anaesthesia: we know that a combination of drugs induces unconsciousness, but we don't know exactly the biochemical mechanism."

The problem then becomes epistemological: can one trust evidence generated by an algorithm if one is unable to verify it? Today, AI does not merely provide support, but generates content, makes operational decisions, anticipates scenarios. It often proceeds in a way that is incomprehensible to our intellect, without making the path leading to the result transparent. "In some cases, generative algorithms become complacent, i.e. they seek the answer we want to hear, not necessarily the most correct one," Kucharski points out. "The risk is twofold: on the one hand technical, because reliability is difficult to assess; on the other hand conceptual, because we get used to taking solutions for granted and give up governing the decision-making process." AI may represent a turning point in the history of knowledge: not just a tool, but a new cultural subject.