AI and the Illusion of Certainty: Why Machines Cannot Master the Unpredictable

I am worried for our future middle management —those who are rising in the ranks quickly, the go-getters. I’m talking about the ones who become directors after about five years of experience. They look to artificial intelligence (AI) with a degree of certainty. But in business, certainty does not exist. Despite its capabilities, AI cannot eliminate unpredictability.

I want to tell them, ‘Don’t get caught; heed the machine, but don’t take it for gospel.’

In an era where AI is embedded into everything from business analytics to personal assistants, there’s a prevailing misconception that AI can provide absolute certainty. Organizations seek AI to optimize processes, reduce risk, and make predictive decisions, yet AI fundamentally cannot manage certainty because certainty does not exist in the real world. You probably were waiting for me to say it, but uncertainty is the only thing that is certain in business.

As I have progressed in my executive career, I have really noted that what has set my great team and me apart is our ability to deal with uncertainty. I have yet to be part of a business that did exactly what the spreadsheet said or found a way to model all the assumptions that we need to truly predict our outcome. It just has not happened.  

The Nature of Certainty

Certainty implies an absolute, unwavering truth—an outcome that is fixed and inevitable. The real world, however, is governed by complexity, randomness, and emergent phenomena. No matter how much data AI processes or how advanced its algorithms become, it cannot eliminate the inherent unpredictability of life, human behavior, and evolving environments.  

Of course, as an engineer and a computer scientist, I am completely fascinated by AI and its ability to affect our lives, but I think we need to be cautious.  It bothers me a little bit that AI can seem like – and has been described as – a magical solution to our complex lives. It will undoubtedly be a game changer, but when it comes to predicting an outcome…. I am not sure I would be ready to trust a machine.

It bothers me a little bit that AI can seem like – and has been described as – a magical solution to our complex lives.

This must come as a surprise; surely a data-loving, computer scientist with engineering and business expertise would love an AI machine. Well sure, but even the most advanced models cannot anticipate unexpected shifts like geopolitical events or market crashes… these are the things that we still need humans for.  

Humans understand the concept of “unavoidable circumstances,” and we can begin the process of managing them better than any machine ever will. We have an inherent ability to think “outside the box” that works for us in times of crisis and uncertainty. We can be cautious and brave when we have to, and that helps us navigate when things don’t go as planned.

The Flawed Premise of AI Data 

AI models process historical data, identifying patterns, and makeprobabilistic predictions. However, they assume that past patterns will continue into the future. This is a flawed premise. 

Disruptions such as economic crashes, geopolitical shifts, or even spontaneous human decisions introduce factors that AI cannot anticipate with certainty. The inability to access perfect, all-encompassing data ensures that AI’s predictions always carry some uncertainty.  

Even more so, for a set of data to exist that can identify a pattern, there must be some form of repetition. An annual pattern may take at least two years, if not three years, to begin to emerge. And since many businesses may not have been capturing the data they need, the promise of predictive AI is realistically two to three years after data capture begins. 

Then, we are aware of macro phenomena in the stock market, such as cycles of five years, 20 years, and even 100 years. In that instance, the pattern may well take 300 years to identify. Real data scientists will be adept at identifying leading indicators of sets of data that may not require us to wait that long, but we will never really know if those models are accurate.  

Probabilistic, Not Deterministic

AI is inherently probabilistic, unlike traditional deterministic systems that yield the same result under identical conditions. Machine learning models provide confidence intervals, likelihoods, and risk assessments—not absolute answers. A self-driving car can estimate the probability of an accident, but it cannot provide a guarantee of safety. 

A predictive maintenance AI can indicate when a machine will likely fail, but it cannot state with certainty that failure will occur on a specific date and time. Does this mean that we should still listen? This is where human operators need to begin to understand why they are getting this alert. 

A self-driving car can estimate the probability of an accident, but it cannot provide a guarantee of safety. 

The information needs to be meaningful in a way that does not require a specialist to intervene. I liken this to the check engine light on your car. Many of us have had the experience of that little light going on but ignoring it for weeks or months. Why? It’s meaningless. What exactly should I check in the engine, and how severe is the issue? If I can’t see or feel a problem, I may not trust that there is one.

The Human Factor

Human behavior is among the most unpredictable variables AI must contend with. People change their minds, make emotional decisions, and act irrationally in ways no dataset can fully encapsulate. AI-driven hiring tools suggest an optimal candidate based on past data. Still, they cannot predict with certainty how that candidate will perform in a novel work environment or how they will respond to unforeseen challenges.

Complexity and Chaos

Fields such as finance, medicine, and weather forecasting illustrate AI’s limitations in managing certainty. An ever-changing web of economic policies, investor sentiment, and global events influences financial markets. Medical diagnoses depend on unique patient variables, emerging research, and individual responses to treatments. Weather forecasts, despite immense computational power, struggle to predict long-term conditions due to the chaotic nature of atmospheric systems. AI can enhance predictions, but it cannot eliminate uncertainty.

AI can enhance predictions, but it cannot eliminate uncertainty.

The Risk of Over-Reliance

Organizations that seek certainty from AI risk falling into a dangerous trap— overconfidence in predictive systems will inevitably fail at some point. The 2008 financial crisis, for example, was exacerbated by reliance on risk models that underestimated uncertainty. Similarly, AI-driven decision-making in critical sectors like law enforcement and healthcare raises ethical concerns when probabilistic judgments are mistaken for absolute truths.

Embracing AI Strengths Without Illusion

Rather than expecting AI to deliver certainty, we must acknowledge and leverage its strengths: improving efficiency, offering probabilistic insights, and identifying patterns at scales beyond human capability.

The key is integrating AI as a decision-support tool, not as an oracle of absolute truth. If that were true, I would submit that we have had an AI tool for decades in the Magic 8 Ball toy that could seemingly predict the truth. I am not saying you should put equal trust in AI and the Magic 8 ball; AI certainly deserves more. But not that much more.

Ultimately, AI cannot—and should not—be used to manage certainty because certainty itself is a myth. Instead, AI should be harnessed to navigate uncertainty with informed, adaptable strategies. The goal is not to eliminate risk but to better understand and mitigate it—a task that requires both machine intelligence and human judgment working in tandem.

I’ll shake my Magic 8 Ball and ask, ‘Should we trust AI in the short term?’ Hmmm…. “Outlook not so good.”