In my attempt to keep myself educated on AI and how it is shaping our world, or speculations on how it’ll shape our world in coming days, I picked up The Age of AI.

Written by Henry Kissinger, Eric Schmidt, and Daniel Huttenlocher, the book is somewhat unusual in the way it approaches artificial intelligence. Most books in this space tend to focus either on the technology or on predictions of the future. This one attempts something else. It tries to place AI within a broader philosophical context.

What I found interesting was not just that it references thinkers like Kant and Wittgenstein, but that it takes their concerns seriously. The book suggests that the questions raised by AI are not merely technical. They are about how we understand reality, how we reason, and what it means to know something at all.

What follows is not a review, but a collection of those moments where I paused, went back, and found myself thinking a little longer.

Beyond the Pinnacle

One of the first ideas that stayed with me was this suggestion that AI may not just be an extension of human intelligence, but something that has already moved beyond it. We have seen machines defeat humans at chess and Go, and that story is now familiar. But what interested me here was not the outcome, but the manner in which it happened. The moves were often described as non-human. Not just better, but different in a way that made them difficult to anticipate or even fully appreciate.

When I had first encountered this idea while reading Janelle Shane’s or Melanie Mitchell’s books, I had dismissed it as a function of computational strength. Given enough power, the machine explores more possibilities than a human can. But reading this again, I found myself less convinced by that explanation. It began to feel like something else was at play. Perhaps intelligence is not a single peak that we are all climbing, with AI simply moving faster. Perhaps it is a landscape with multiple paths, and we have been walking on one of them, assuming it to be the only one.

I remember reading earlier about certain AI outputs that felt slightly alien. They were not incorrect, but they arrived at correctness through routes that are difficult to explain. At that time, I treated it as an interesting quirk. Now it feels like a hint of something more fundamental. If intelligence can take forms that we cannot intuitively follow, then the idea of human intelligence as the pinnacle begins to look less certain.

The Limits of Definition

Another idea that stayed with me was about how we define artificial intelligence. John McCarthy’s definition, machines performing tasks characteristic of human intelligence, has a certain elegance to it. It gives us a frame of reference. The Turing test is built on this framework. But the more I thought about it, the more it started to feel like something I have encountered before in my earlier readings. The tendency to understand AI by mapping it onto ourselves.

In some of my earlier notes on AI, especially while thinking about anthropomorphism, I had realized how quickly we project human qualities onto these systems. We talk about them as if they understand, intend, or perceive in ways similar to us. McCarthy’s definition, while useful, seems to carry a similar impulse. It defines AI in relation to human intelligence, as if that is the natural benchmark.

I might be stepping into mystical territory here, but almost all of the eastern paths towards enlightenment speak about a world that is beyond the senses if we were able to perceive it. Human intelligence itself is deeply tied to our senses. What we know and how we know it is shaped by what we can see, hear, touch, taste, and smell. Our technological progress, for the most part, has been about extending these senses. The telescope extends sight, the microphone extends hearing, and even vehicles extend our ability to move through space. We have always been building outward from the body.

AI does not begin there. It does not share our sensory grounding, and therefore there is no inherent reason for its intelligence to resemble ours. This led me to a thought I had not fully articulated earlier. If AI is not constrained by human modes of perception, then its intelligence may not just be stronger or faster, but qualitatively different. And if that is the case, then defining it in human terms may not just be limiting, it may be misleading.

And perhaps more importantly, we don’t know what it might be capable of. We know that the engine of its thinking is formed by neural networks, but so is the thinking formed in every living being with neurons. We can see the output, but we don’t know how they think or what they perceive.

Plato or Wittgenstein

There was a section in the book that brought in philosophy in a way that I found particularly interesting. The idea that machines learn not by identifying an ideal form of something, but by observing many instances and finding patterns across them. To recognize a cat, the machine does not need a perfect definition of “catness.” It needs exposure to many cats, across contexts, and the ability to detect overlaps.

This is often described as being closer to Wittgenstein than Plato. Instead of ideal forms, we have family resemblances. Meaning is not anchored in a perfect definition, but in patterns of use and similarity.

When you look at the large language models, you can see similar phenomenon. The idea that meaning emerges from usage rather than from fixed definitions. It felt like an interesting philosophical parallel. But here, the machine is not merely illustrating the idea. It is implementing it. It is building a working system on the basis of these patterns.

And that is where I found myself pausing. Because once meaning is reduced to pattern recognition, the question of understanding becomes more complicated. Do we say that the machine understands what a cat is, or do we say that it has learned to recognize patterns that correspond to what we call a cat? And more importantly, is there a meaningful difference between the two?

I do not have a clear answer to that. But the fact that the question can be asked at all suggests that our traditional distinctions may not hold as firmly as we assume.

Truth Without Understanding

The final idea that stayed with me is getting an answer without even knowing the question. A bit of a 42 scenario. Serendipity might have found a boost.

The idea was that AI systems may produce results that are true and useful, but beyond our ability to fully explain. The comparison to Alexander Fleming and the discovery of penicillin is helpful, but only up to a point. In Fleming’s case, the mechanism was not immediately understood, but it was eventually explained. The gap between observation and understanding was temporary.

With AI, that gap may persist.

In my post on the illusion of understanding, I had reflected on how we tend to attribute understanding to these systems, even when they are operating on correlations rather than comprehension. That tendency becomes more problematic in this context. Because now, the system is not just generating plausible outputs. It may be producing insights that work, even if we cannot fully account for them.

This raises a difficult question. What do we do with knowledge that we cannot explain? If a system consistently produces results that are effective, do we accept them as valid, even in the absence of understanding? Or does the lack of explanation limit their legitimacy?

I find myself leaning in both directions. On one hand, science has always moved forward through observation, often ahead of explanation. On the other, the idea of relying on something we do not understand feels like a departure from that tradition. Not because it is wrong, but because it changes our role. We move from being creators of knowledge to interpreters, perhaps even auditors of something that lies outside our conceptual reach.

Where This Leaves Us

I do not think these ideas form a single argument. They feel more like fragments that point in a similar direction. Intelligence that may not be centered on us, perception that is not bound by our senses, meaning that emerges from patterns rather than definitions, and truth that may not require understanding.

In my earlier readings on AI, I often found myself thinking about what these systems can and cannot do. Their limitations, their occasional absurdities, the ways in which they resemble and differ from us. This felt like a shift away from those questions. Less about capability, and more about position. Where do we stand in relation to something that may operate beyond our intuitive frameworks?

I do not have a conclusion to offer here. Only a sense that these are questions worth returning to. And perhaps, the next time I come across an impressive claim about AI, I will pause a little longer. Not to decide whether it is correct or incorrect, but to ask whether it is something I understand, or simply something I am learning to accept.