markdown content
The Genius Who Might Have Been Wrong
Okay, I need to share something that genuinely blew my mind this week, and it's got me thinking about artificial intelligence in a completely different way.
For decades, we've measured AI progress against a vision laid out by Alan Turing back in 1950. His famous "imitation game"—now known as the Turing Test—suggested that if a computer could fool humans into thinking they were talking to another person, it would essentially be "thinking."
Sounds reasonable, right?
Well, according to Peter J. Denning, a seriously respected computer scientist who's just released a new book called "Turing's Mistake: Escaping the Yoke of Unintelligent Machines," we might have been chasing the wrong dream for the past 75 years.
And honestly? After reading through his arguments, I think he might be onto something profound.
The Body Problem (Yes, It Matters)
Here's the first assumption Denning thinks Turing got wrong: the idea that intelligence can exist independently from having a physical body.
Think about that for a second. We’ve been trying to recreate human-level intelligence inside computers—pure software, floating around in digital space. But Denning argues this is like trying to understand swimming by studying water molecules. You're missing something essential.
When you and I navigate the world, we don't just process information. We feel things. We sense the temperature, feel the texture of a doorknob, experience the slight wobble when we step on a slightly uneven sidewalk. All of this physical experience shapes how we think, reason, and understand the world.
Denning points out that our bodies aren't just vehicles for carrying around our brains. They're integral to how we think.
And here's the thing—this isn't some fringe idea. Philosophers have been talking about "embodied cognition" for years. But AI research has largely ignored it in favor of pure computational approaches.
The Tacit Knowledge Puzzle
This is where things get really interesting (and a bit philosophical, but I promise to keep it grounded).
Denning introduces the concept of "tacit knowledge"—and it's honestly one of those ideas that, once you grasp it, you start seeing everywhere.
Tacit knowledge is all the stuff you know but can't easily explain. Like... how do you know exactly how much pressure to apply when unlocking a door? How do you catch a ball without doing complex physics calculations in your head? How do you know when someone is being sarcastic versus sincere based on the tone of their voice?
These things feel automatic to us. But they're incredibly complex.
Denning argues there are at least five major categories of tacit knowledge that machine learning simply cannot capture:
Common sense – The millions of tiny facts about how the world works that we never consciously think about.
Everyday interactions – How we read social situations, navigate personal relationships, understand unspoken expectations.
Emotions and perception – The way experiencing joy or sadness fundamentally changes how we process information.
Practical skills – The kind of knowledge a master craftsperson or virtuoso musician has that they can't fully articulate.
Cultural understanding – Values, norms, history, and the deeper meanings embedded in how communities work.
Why 40 Years of Cyc Didn't Solve This
To illustrate the challenge, Denning points to something called the Cyc project. If you're not familiar with it, Cyc was an ambitious attempt started in the 1980s to build a massive database of common sense knowledge—things like "if you drop something, it falls down" or "people need water to survive."
The project took four decades to build up roughly 25 million entries of common sense facts.
Twenty-five million pieces of knowledge!
And you know what? It still wasn't enough to create truly smart systems. Denning notes that "much of the knowledge that makes people experts cannot be articulated as propositions."
This is the quiet tragedy of the Cyc project. It demonstrated, through sheer effort, that trying to manually encode human common sense is essentially impossible. There's too much of it, and too much of it is implicit rather than explicit.
The Musician Who Can't Explain It
Let me share one of Denning's examples that really stuck with me.
Think about a virtuoso violinist. Someone who can play with breathtaking beauty and emotion. Now imagine trying to explain to a beginner exactly how to produce that sound.
They can't really do it. Not fully. They might offer some technical advice—angle the bow here, apply pressure there—but the knowing how versus knowing what gap is enormous.
The violinist possesses embodied knowledge that exists in their muscles, their nerves, their physical experience of creating sound. You can't download that into a computer. You can't transcribe it into a database.
And here's the deeper point: even if you built a robot that could perfectly imitate a violinist's movements, that robot still couldn't understand what it feels like to create beautiful music. It lacks the biological experience, the emotional context, the connection to an audience.
The Representation Problem
So what's actually going on here? Why is human knowledge so stubbornly resistant to being digitized?
Denning calls it the "representation problem," and it's genuinely fundamental.
Computers work with symbols—ones and zeros that represent things. When we feed information into computers, we're encoding real-world knowledge into these symbolic forms.
But here's the catch: behind every word, every concept, every piece of "knowledge" we possess, there's a deep well of tacit understanding that gives it meaning. Words are just symbols pointing to meanings—they're not the meanings themselves.
Denning puts it this way: "Behind every word is a deep well of tacit knowledge that gives it meaning. Words are but symbolic representations of meanings, not the meanings themselves."
This means when you talk to something like ChatGPT, Claude, or Gemini, you're interacting with systems that are incredibly sophisticated at manipulating symbols, but they don't truly understand what those symbols represent. They can pattern-match brilliantly, but pattern-matching isn't comprehension.
Context Is Everything (And Machines Don't Have It)
Here's another layer that AI struggles with: context.
When you and I have a conversation, we're drawing on endless layers of shared understanding. We know who we are, what we've been through together, what the current situation is, what's appropriate versus inappropriate, when someone is joking versus serious.
This context isn't just background noise—it's fundamental to meaning.
Denning describes it as fractal: each context rests on previous contexts, which rest on even earlier ones. It's an endless chain of shared understanding that we've built up over a lifetime of experience.
Culture adds yet another dimension. Denning describes it as encompassing "values, norms, judgments, history, communities, moods, and even relationships involving power and care."
All of this is invisible to AI systems. They can detect patterns in language, but they can't truly grasp the cultural weight behind words and actions.
What Does This Mean For AI's Future?
Now, I'm not saying AI is useless—obviously it's not. These systems are genuinely impressive at specific tasks, and they're incredibly useful tools.
But Denning's argument suggests we may be approaching a ceiling. That true artificial general intelligence—machines with human-level understanding and adaptability—might be fundamentally unachievable through our current approaches.
And here's where it gets a bit concerning: Denning warns that if we keep pursuing AGI while ignoring these fundamental limitations, we risk building technologies that introduce significant new risks without delivering on their promises.
Think about it. We're building systems we don't fully understand, using approaches that may be inherently limited, to solve problems that require kinds of understanding these systems can't access.
My Take
Honestly, I found Denning's arguments compelling, but not completely crushing.
Yes, there are fundamental challenges with capturing tacit knowledge. Yes, embodied experience likely matters more than we've acknowledged. And yes, the Turing Test might be a poor proxy for genuine intelligence.
But I also think we're discovering these limitations in real-time, and that discovery process itself is valuable. Every challenge Denning identifies is also a research question. Every wall we hit teaches us something new about human cognition and consciousness.
Perhaps the real lesson here isn't that AI can't be intelligent, but that human intelligence is more remarkable than we appreciated. Every time you understand a joke, catch a ball, or sense that something feels "off" in a conversation—you're doing something that our most sophisticated computers still cannot do.
Maybe the question isn't whether machines can think like humans. Maybe it's: what can they do well, and how do we build that into a world where human qualities matter more, not less?
That's a question worth sitting with for a while.
What do you think? Does Denning's critique resonate with your experience using AI tools? Or do you think we're underestimating what these systems can eventually achieve?
Drop a comment below—I genuinely want to know how this lands for you.
Source: ScienceDaily (https://www.sciencedaily.com/releases/2026/07/260713084850.htm)