• AI resists easy definition, but organizations can still measure how it performs—accuracy, errors, human intervention needed, business impact.
• We can frame AI as an unknowable "UFO" or as a manageable technology like electricity; that choice shapes how we govern it.
• AI adoption is outpacing the security and governance needed to catch hallucinations, manipulation, and faulty outputs.

The Paradox of Defining Time

“Quid est ergo tempus? Si nemo ex me quaerat, scio; si quaerenti explicare velim, nescio.”

(“What, then, is time? If no one asks me, I know; if I wish to explain it to someone who asks, I do not know.”)

Augustine wrote these lines between 397 and 400 CE, in Book 11 of his Confessiones (Confessions). In his eyes, time was one of the things human beings know most intimately — a reality we use, feel, and live inside every single day. Yet the question was simple: “What is time?” And it was precisely at that point that certainty gave way to confusion. Because describing something we experience is not nearly as easy as living it. What Augustine had stumbled onto was not just the mystery of time, but the limits of human knowledge itself: the things we think we know best are often the very things we find hardest to explain.

Today, roughly 16 centuries later, we see the same paradox playing out with AI. Like time, AI has become a reality we use every day yet struggle to define. We assume we know what AI is. This confidence is perhaps unsurprising. According to the Stanford AI Index 2026, generative AI reached 53% global population adoption within just three years, making it one of the fastest-adopted technologies in history. Yet, despite its remarkable diffusion into everyday life, our conceptual understanding of AI has not kept pace with its rapid adoption. But the moment someone stops and asks, “What exactly is artificial intelligence?” that shared certainty quickly falls apart. In its place, two powerful and opposing narratives emerge.

In one narrative, AI is a transformative technological leap on the scale of electricity or the Internet for humanity. In the other, it is a new source of risk whose scale and consequences we still do not fully understand.

Which is why the question worth asking about AI today isn’t “what is it,” but how we choose to look at it. Do we see AI as a UFO — or as a tool that, like every other technology before it, we are trying to understand and govern?

In the UFO frame, AI is seen as a force whose rules we don’t yet fully understand and whose control we fear losing. Debates about superintelligence, singularity, and existential risk are natural extensions of this perspective.

In the normal-technology frame, AI is transformative but manageable — much like electricity, the automobile, or the Internet. This is also the approach advocated by Arvind Narayanan and Sayash Kapoor of Princeton University. To them, what matters is not the theoretical capability of models, but how those capabilities actually get applied in the real world, which processes they transform, and which risks they create.

Both perspectives focus on the same technology. The difference between them lies entirely in which mental model we choose to apply.

While the philosophical debate continues, business has already moved. Companies have positioned AI as a strategic priority; AI agents and automation layers have been rolled out across customer service, software development, enterprise knowledge management, and decision-support processes. Call-center assistants, sales chatbots, enterprise knowledge assistants, and the summarization and content-generation tools used by legal and finance teams are no longer experimental projects — they are production systems directly shaping business outcomes.

But there is a striking asymmetry here: the pace of AI investment and adoption is leaving the pace of security, governance, and verification mechanisms far behind. While organizations boast that they have “gone AI,” they often push these systems into production without adequately testing under what conditions they will produce faulty outputs, when they will hallucinate, or how they can be manipulated. In other words, in the AI race, speed is most often outrunning trust.

This tension, in fact, isn’t new. A similar observation appeared decades ago in science fiction.

In 1953, Ray Bradbury, in his short-story collection The Golden Apples of the Sun, told the story of a man named Albert Brock in “The Murderer.” Brock lived in a world besieged by telephones, wrist radios, intercoms, and endless announcements. The “constant connectivity” that technology promised had become, in his eyes, not freedom but a kind of imprisonment.

Eventually he rebelled. He smashed the devices, silenced the screens, sabotaged the communication network. The system’s response was even more telling: it labeled Brock not a rebel, but mentally ill. The paradox Bradbury identified was this: once a technology spreads fast enough, our capacity to govern it always lags behind. Today’s chatbot lawsuits, prompt-injection incidents, and systems pushed live without proper testing are exactly the bill for that delayed governance.

Risk at Scale

As AI systems spread rapidly across every corner of life and business, over the past two years risks have moved out of theoretical debate and started showing up directly on balance sheets and in reputational damage.

• Chatbots that present themselves as ‘experts’ and expose companies to lawsuits because of misleading guidance.
• Enterprise assistants manipulated through simple prompt-injection attacks.
• Reputational and operational costs created by systems pushed live without sufficient testing.

What these examples have in common is not that AI makes mistakes — every technology makes mistakes. What’s new is that these mistakes can now scale instantly to millions of users, thousands of employees, and critical business processes.

The Question of Measurement

When Augustine tried to define time, the very thing he knew most intimately slipped through his fingers. Centuries later, physicist Richard Feynman, while explaining the fundamental concepts of physics, pointed to a similar idea: for some things, what matters most is not how we define them, but how we measure them.

This idea offers a powerful frame for the AI debate as well. The argument over what AI “is” will go on; but the problem organizations actually need to solve is not philosophical — it’s operational.

Albert Brock didn’t try to define the technology; he destroyed it because he could no longer bear it. The system, in turn, didn’t label him a “rebel” — it labeled him “sick.” We see a modern version of that same mistake in today’s world: AI either spreads everywhere with blind speed or gets rejected outright out of blind fear. Yet there is a third path: neither glorifying the technology nor demonizing it — simply measuring it.

For organizations, the question is no longer how impressive AI is, but how reliable it is. The success of an AI system shouldn’t be measured by adoption rate alone, but by its accuracy, how much human intervention it requires, the security incidents it generates, the impact of its errors on business outcomes, and the concrete business value it actually produces. You cannot manage what you cannot measure; and you cannot safely scale what you cannot manage.

In practice, it all comes down to a single question: is what you’re scaling actually creating value — or are you scaling risk at the same speed? For organizations that never ask this question, AI will stop being a manageable tool and remain a UFO quietly growing inside their systems — one whose moment of crashing down is anyone’s guess.