“With AI, it’s now easier to resolve uncertainty,” a business leader recently told me with confidence, arguing that with the mass of data now available and the almost infinite capacity to analyze it, the subject was more or less closed. This is a widespread, long-held belief… and a very false one at that, a scientistic illusion of management that refuses to die. The link between data and uncertainty is much more complex. Without a thorough understanding of this link, decision-making in uncertainty is based on flawed models, with catastrophic consequences.
“The Israelis are used to knowing exactly what the Palestinians are doing, down to the smallest detail, thanks to their sophisticated espionage capabilities.” So wrote American diplomat Martin Indyk in his analysis of the October 7, 2023 attack that took Israeli agencies completely by surprise. This was not the first time, and certainly not the last, that the accumulation of tactical data using sophisticated technologies had failed to prevent catastrophic strategic blindness, when it had not caused it in the first place.
Yes, AI can help to better manage uncertainty, but it does not eliminate it; it mainly changes the way we perceive it and how we act in response to it. Here are some basic ideas on this complex subject. First idea: AI makes it possible to replace single predictions with probabilistic estimates and sets of scenarios. It does not predict what will happen; it indicates what is more or less likely, which helps to prepare several possible responses rather than a single rigid plan. This makes uncertainty more explicit and more “manageable,” without making it disappear.
Second idea: AI is capable of identifying patterns and weak signals in very large volumes of data, which can sometimes detect changes in trends, anomalies, or emerging risks. In this sense, it can reduce short-term uncertainty on operational issues, for example by anticipating stock shortages, fraud, or performance degradation, where humans would be overwhelmed by the amount of information. The key here is continuity: it allows us to define stable models that reflect the intrinsic causes of the phenomenon, from which we can identify suspicious variations. Without continuity, it doesn’t work.
Third idea: AI can support decision-making in uncertain situations by simulating possible futures and virtually testing alternative choices. We no longer seek the optimal decision in a predictable world, but the most robust decision in several plausible scenarios. This shifts rationality toward managing variability rather than seeking illusory certainty.
Fourth idea, often overlooked: AI only works on data formalized in digital form, whereas much of the relevant information, the information that really matters, does not exist in this form. Intentions, motivations, latent conflicts, feelings, trust, fear, local culture, informal power relations, or weak political signals are rarely well captured by quantitative data. In a company, for example, tensions between teams, loss of motivation, or emerging creativity are only very imperfectly visible in digital indicators. In geopolitics, decisions may depend on psychological, symbolic, or ideological factors that are largely beyond the scope of models. This means that AI can give the impression of control when it ignores decisive, non-digitized factors in understanding what is really going to happen.
Fifth idea: even when information is digital, data is often ambiguous, incomplete, biased, or even false. It reflects imperfect measuring instruments, subjective categorization choices, strategic behavior, or human error. On social media, for example, some content is performative, exaggerated, or manipulated, which can skew the analysis of opinions. In economics, certain indicators mask very heterogeneous realities behind misleading averages. In healthcare, databases may underrepresent certain population groups. In addition, data can be deliberately distorted on a massive scale (fake news). While AI can learn very effectively from data, it also learns their distortions, which can produce predictions that are mathematically accurate but fragile, or even erroneous, in terms of meaning. As computer scientists say, “Garbage in, garbage out.”
Sixth idea: AI is fundamentally backward-looking, since it infers the future from observed regularities. This is the limitation of induction. When the context changes abruptly—crises, technological disruptions, political changes, pandemics, surprise attacks—models can fail. In these situations, the uncertainty is not only statistical, it is structural: the rules of the game themselves change. AI does not handle this type of radical unknown well, because it is unlike anything that has been observed before. Such sudden changes are undoubtedly rare, but their impact is very significant, which makes failure to anticipate them catastrophic. These are the famous black swans identified by Nicholas Taleb.
Furthermore, sophisticated processing of large amounts of data is useless if the basic assumptions, and mental models more generally, are flawed. In October 1962, the US intelligence services were alerted to significant Russian activity in Cuba. The Russians were building ballistic missile bases, but despite extensive observational data, the CIA refused to consider this possibility until almost the last minute, convinced that the Russians would never do such a thing. In decision-making, the beliefs of the decision-maker (or analyst) count for more than the data, and AI is of no help here. This is an issue I analyzed at length in my book Constructing Cassandra, which deals with strategic surprises.
One final thought: AI can shift the way we take responsibility in the face of uncertainty. When a decision is supported by a model, it becomes tempting to rely on the algorithmic recommendation, even when the real context no longer exactly matches the implicit assumptions of the data. This can reduce the anxiety associated with uncertainty, but at the cost of weakening critical judgment and human deliberation. The danger here is to believe that the tool is magical and that it frees us from effort and judgment.
In short, AI helps to transform some of the uncertainty into quantifiable risk, to develop scenarios and to explore options more systematically than a human could do alone. In this sense, it is a useful decision-making tool. But it remains limited by the digital and imperfect nature of data, by the ambiguity of reality and by the emergence of discontinuities.
A question of attitude
More generally, there is a question of attitude. Often, “resolving uncertainty” means being able to predict better. This explains the wild hopes placed in AI, which, based on its mountain of data, leads us to believe that this is possible. However, more data does not enable us to predict better, let alone understand better. It is the basic stance that is flawed. Uncertainty cannot be “resolved” by prediction because it does not result from the difficulty of finding or analyzing data, but from the absence of data due to the unprecedented nature of the situation in question.
We need to develop a different, two-pronged approach. First, we need to engage with reality: instead of viewing the world as an abstraction from the top of our mountain of data, we need to be on the ground and accept its ambiguities. We need to get our hands dirty. Second, we need to prioritize creation over prediction. It is not a question of predicting, or trying to predict better, but of inventing a new response to a new situation.
Culture of uncertainty
AI is a powerful tool that significantly improves decision-making. But as with all tools, it is important to understand how it works in order to know its limitations. No tool is ever the ultimate, one-size-fits-all solution. Believing that the world can be reduced to a mathematical equation and a calculable data set is a scientistic illusion that is deeply harmful to decision-making. Abandoning this belief would contribute to the development of a culture of uncertainty, which is essential in today’s world.
One thought on “Resolving uncertainty with AI, or the scientist illusion of management”