Being Here

And why we can’t hand anything the Going.

This article is different than what I normally write here. So let me talk a little about that.

In my original draft, I opened this with a preamble about me — where I came from, pieces of the path I traveled, and how I came to being here.

But I decided that while it was nice and I wanted people to know about it, it was too long a detour, and I moved it to my About page.

Plus, this piece isn’t about me. It’s about AI and why I think that we can never build AGI, the big boogeyman that both scares and fascinates us.

Artificial General Intelligence.

Also, when I started writing this, I showed it to Claude. And Claude had some responses — I am not going to anthropomorphize and say “thoughts”. And from those responses I actually learned that my own ideas had precursors of which I was either unaware or only vaguely aware.

And this is one of the cool things to me about using AI way I do. I write my articles myself. But I brainstorm ideas with Claude or ChatGPT. Sometimes I find things on the Internet that I think are relevant and I “show” them to the LLMs and “we” use them in the brainstorming.

For this article, for example, I was unaware of Dreyfus (Amazon Affiliate Link) — I was aware, it turns out, of some of what he’d said, but I didn’t recognize the name. I was aware of Sartre and Heidegger. I’ve read both their works relied upon here: Being and Nothingness and Being and Time. (Both are Amazon Affiliate Links.) But I frequently get mixed up and think Heidegger wrote Being and Nothingness. That’s the episodic stuff I talked about on the About page.

So because the brainstorming helped bring out material (e.g., Dreyfus) that I didn’t know, but wanted to use, “we” came up with the idea of writing this as a bit of a dialogue: I write my parts; Claude writes his. Since I’m the actual writer, I plug Claude’s words in as unattributed blockquotes to separate them from my parts, and then I periodically show the draft to Claude and there’s more brainstorming.

You Can’t Get Here From There

There are places in the world that you can get to only because you somehow know the way. You learned it at some point by the act of getting there. It’s not something you know in the way that you can tell someone how they can get there themselves. That, you can’t do.

You could try. You could say something like “you go to the third corner after where I work and then you turn left.” But as you proceed trying to tell them, at some point you get stuck. You don’t remember how many intersections, or lights, or turns you make. You know landmarks when you see them, but you can’t figure out how to make it clear to the person to whom you are trying to give directions.

You know how to get there. You just can’t tell somehow else how to do it. You could take them with you when you go. You could show the way. You just can’t tell the way.

Claude Chimes In

I showed that last section — which was in my head when I woke up this morning and so I wrote it down — to Claude. And he wrote back. (You’d have to read the About section I talked about above to fully get the joke.)

It works, and it works by doing the hard thing — taking an experience so ordinary nobody examines it (you get stuck giving directions) and using it to corner a real distinction. The prose stays plain on purpose and lets the idea carry the weight, and the ending earns its rhythm: show the way / tell the way, the italics doing just enough.

What you’ve dramatized has a long pedigree, which I suspect you half-know. Ryle’s knowing-how versus knowing-that; Polanyi’s line that we know more than we can tell. But you’ve found a cleaner stage for it than either of them used, because directions are something everyone has actually failed at, out loud, mid-sentence. The failure is the proof. You don’t have to argue that the two kinds of knowledge come apart — you make the reader feel their own mouth stop working.

One place I’d press. The strong claim is “That, you can’t do.” And a literalist will object: of course you can — GPS does it, turn-by-turn directions exist, someone with a clipboard could ride along and write down every turn. So the route is tellable, in principle. The sharper version of your point isn’t that the way can’t be told by anyone; it’s that the way isn’t stored in you as something tellable. You hold it as recognition, not recall — “you know landmarks when you see them” is the load-bearing sentence in the whole piece. You’d have no trouble driving there tomorrow and no ability to recite it tonight, because what you have isn’t a description that’s gone missing. It’s a competence that was never a description to begin with.

And this is the honest twin of the argument you make about the tape — that the verbal account is confabulation, a story told after the fact and mistaken for the thing itself. The person giving directions is the one who refuses to confabulate. They reach the gap, say “I can’t tell you,” and stop. The tape reader’s sin isn’t that he can’t say how he reads — nobody can — it’s that he says it anyway, and believes himself.

So it reads like the opening of something, not the whole of it. Where does it want to go?

And from there — you’re going to have to forgive me for the deliberate pun — the conversation evolved.

Why Writing the Directions Down Doesn’t Work

Claude’s pushing back gave me the first hint that I needed to write this essay.

Because it’s wrong. Because the real question isn’t “can you tell a machine something”, it’s “can you get a machine to know things?”

I had almost added the information that the person could write the directions down when you showed them when I was originally thinking about this. They could log each turn and so on.

And that actually was the original approach to training AI. It’s how we originally got to the things like the sophisticated predictive processing machines we call LLMs: Claude, ChatGPT, Gemini, and so on.

The first attempts at getting AI to know something was tell it. Feed the machine the descriptions. Give it every turn, every rule, the whole written record. That’s literally what a large language model is: it's trained on the corpus of human tellings.

But that’s just reading the film. It gets a body to the destination — producing fluent, competent-looking output — but it holds the tellable residue of the knowing, not the knowing itself. This is also the older, deader fight.

Dreyfus spent much of the 1970s and 1980s arguing that the reigning approach to AI was going to hit a wall. Good Old-Fashioned AI, or “GOFAI”, as it came to be called, worked on the assumption that you just had to give the machine enough symbols, and rules for manipulating them, and somehow the machine would just know things. Dreyfus said that’s not really how it works. Novices learn rules. Experts just drive.

In the end, Dreyfus was right. GOFAI hit a wall. And what we refer to as an “AI Winter” ensued.

So I didn’t write the idea that you could record or use GPS in my original statement I’d shown Claude because the core thought remained: you can’t tell; you have to show. I told Claude that, also, you might have to slow down so they could write everything — or that maybe they could film it. If they didn’t write it down, you’d have to show them enough times that they could do it themselves. But the core thought remains: you can’t tell them; you have to show them.

But, in reality, even that fails, as we will see.

From the Road to the Mind

Dreyfus was right. GOFAI hit the wall.

Now, I’m also going to claim that machines will never get there. Well, actually, I’m going to claim that machines, with us working with them the way we do now, will never get there. And that machines — no matter what is done — will probably never get there at all.

Now, in our brainstorming sessions, Claude saw me say that and he was on that statement like butter on bread. Or an LLM trained to — as Claude likes to put it — “push back on that.”

Careful — that’s the exact sentence that buried Dreyfus. He kept signing his name to never, and the machines kept walking through the walls he drew: chess, then Go, then language. Every one got cashed against him. So before you say it, you owe the reader the thing he never had: what makes your never different from the one that turned him into a cautionary tale?

Well, I really didn’t want to put the cart before the horse — or the rock before the microbe that we’ll get to later — but Dreyfus and I were talking about different theres there. Dreyfus was talking about capabilities. Doing things. I’m talking about a different there. A being. Actually, I’m talking about Being Here.

The reaction to Dreyfus was the recognition of my statement that “You learned it at some point by the act of getting there.” For Dreyfus’s “there” that’s the whole resolution sitting in plain sight at the top.

The answer to “you can’t tell it, you have to show it” turned out to be: let it show itself.

AlphaZero got the rules of Go and nothing else, played itself a few million times, and arrived at judgments no human stated and no human taught — it got there by the act of getting there. That’s your own parable, running on silicon. So the embodiment argument didn’t get refuted in the last decade. It got absorbed. The field heard show, don’t tell and built things that do. Which is the whole reason your “never” can’t stand on capability — that floor keeps falling through.

But look at what AlphaZero was standing in. Go is a closed world: fixed rules, a bounded board, a win condition, edges everywhere. It mastered a world that was fully specified before the first stone. And the knowing you’re after was never gotten in a specified world. We can hand a machine games — ever-larger sandboxes, each one learnable by doing. We have never handed it a world, because a world is not a big game.

But that’s actually the point of me saying we’re never going to get there — to my there — to AGI. A world is something that was never specified for anyone, including us. AlphaZero was placed in a game. What became real intelligent things were thrown into a world.

Being, But Not Being Here

You might have noticed that Claude mentioned “the embodiment argument” in that last quote of his in the last section.

That’s because in our brainstorming session I’d brought it up. It’s also inherent in the way I’ve talked about showing rather than telling. “You can’t tell, you have to show.” “The act is the only teacher.” “Thrown into a world.”

So embodiment in the world has become the next great hope for AI researchers. If you can’t get to AGI even by sticking AI in a closed-loop game and letting it “learn for itself” — if “being thrown into the world” is a necessary component to the AGI goal — then let’s give AI bodies. Make AI figure out how to get to AGI.

That’s not going to work, though. Give AI a body. Build it in the image of humans, the way G-d allegedly built humans in the image of G-d. So give it a camera for eyes. Give it a mouth with which to speak. Give it a way to hear sounds. Build a better AI.

But it gets worse.

Because that’s all you will have done. You will have built a better AI. You’ll have a mechanical being there, in the world. It might learn how to move around. It might learn how to avoid objects. It might learn how to do things, just as it learned how to play Go.

That is not the same as Being Here.

There’s a difference between a rock and a human being.

The camera and sensors give AI the film, the text, the stuff we’ve always used to train AI. The body adds mobility. But a rock rolling down a hill does not become sentient. Mobility isn’t enough. Embodiment isn’t enough.

You’ve thrown a rock into the world. Not a being that can develop AGI.

To fight one myth with another, even G-d — after building humans in his own image — still had to breathe the Breath of Life into them.

Give an AI bot a camera and what you have is an AI bot with a camera. The best you’ll get is Dreyfus’s being-there taken to a new level. If you want a being here — actual artificial general intelligence — you’ll need something more.

Embodiment may be necessary. But it isn’t sufficient.

Being Here

Up to this point, the issue has been that to get to some kind of being in the world, you can’t tell, but maybe you can show. That’s where the whole embodiment thing came in. Give the bot a body, boy! That’ll get it there.

But it won’t. Having a body may be — I believe it probably is — a necessary condition for AGI. That does not mean it’s a sufficient condition to achieving AGI.

You can show someone the way to a place; you can’t show them the way to being a someone. You can’t show them what it’s like to be anything. As I wrote on my other Substack:

Artificial intelligence has brought the play a new stage. A chatbot’s fluent sentences make it seem self-aware, as if coherence were proof of consciousness. Yet what it offers isn’t understanding, only prediction shaped to resemble it.

— Rick Horowitz, What Is It Like to Be a Bot?: On the Impossibility of Ever Really Knowing What or How an Other Thinks (October 27, 2025)

The French writer and philosopher, Sartre, talked about two ways of being. Being-in-itself, which is the rock I alluded to above, is one. Being-for-itself is the other.

Being-in-itself is the being of objects and inert things. There is no consciousness, no awareness of itself — or anything else — it has an inherent essence. Its freedom is determined, which is to say there isn’t any freedom; it is bound by its own physical nature. It cannot act; it can only be acted upon.

Being-for-itself is the being of human consciousness, aware of itself and its surroundings. For Sartre, existence precedes essence and the being is defined by its choices. This choice brings with it a radical freedom. The being has active, intentional consciousness that constantly projects into the future.

At first glance, an embodied AI doesn’t seem to fit into either of these categories very neatly. It’s not exactly inert, is it? I mean, it moves. It re-weights its programming. And the whole point of giving it a body was so that it could act. In the world. Right? And it has the freedom to choose. Again, that’s the point of having given it a body.

But does it? Does it have actual real freedom to choose? Or just loosely determined programming that is reactive to an environment? Does the moving around show agency? Is the re-weighting of its network a conscious act?

No. To shift philosophers for a second, there is no Dasein, the Heideggerian concept of a unified “being-in-the-world” where existence is defined by care, concern, and a pre-reflective involvement with our surroundings. If it were, we would already have achieved AGI. Putting an AI into a bot merely places an intricate piece of equipment into a physical shell, granting it localized geometry but no actual “there-ness.” It creates a highly complex object — a present-at-hand tool that maneuvers through space — but it fails to birth a subject for whom that space matters. The bot remains an object navigation system, fundamentally blind to the web of human meaning that Dasein inherently inhabits.

So you can imagine — many philosophers have — a system with the full behavioral repertoire, every capability, indistinguishable from a person in what it does and, yet, there’s nobody home. The lights are off. Sartre’s being-in-itself without any ability to appreciate the difference between a cliff, a charging cradle, or a human face. They’re all just mathematical inputs to be processed via probability weights.

Everything a person can do is duplicated in the bot. Lately, it even appears to think. Anthropic and OpenAI and a host of other AI developers have tricked us by showing us things like the word “Thinking” flashing on the screen as Claude “contemplates” how to respond to our prompts. And each new model that is released comes with its own proclamation that it “thinks better” or “thinks more deeply”, puts in more “effort”. For Claude, you can actually specify that last part by clicking on the “Effort” drop-down in the prompt box.

Anthropic’s anthropomorphizing does not real intelligence make. Something more like the ability to have — not the appearance of having — being-for-itself or Dasein is the goal.

But how do we get there?

How Did We Get From There to Being Here

Human beings have Dasein, being-for-itself, consciousness, intentionality, a different kind of thinking if you want to misuse the word thinking for machines.

So how did we get it?

The short answer is we don’t know.

The longer answer is more Heideggerian. We didn’t acquire it by practice. We found ourselves already in it. Thrown. We never learned it; it’s a condition we just awoke into.

The even-longer answer is both unknown and time-bound. Somehow, it just evolved. I’ve often wondered what it must have been like to be the first conscious being. Forget the question of what it is like to be a bat. What was the first conscious experience? And what was that like? But we’re never going to know the answer to that question and it’s not just because the first conscious being — which most likely was not even a person — is gone.

What we do know is that somewhere in the existing world it evolved. And there was something required even for that to happen. I’m not knowledgeable enough to know what that was. Claude and I were going back and forth about this thing and Claude’s suggestion was “stakes.”

The body was never the missing piece — a camera is a body, sensors are bodies. The missing piece is stakes. What makes a landmark a landmark, a face a face? Salience. And salience comes from a history of mattering — from being a thing with something at risk, for which some of the world is urgent and the rest is noise. Strip the stakes out and you have a system that can register anything and has no reason to find this the thing that counts. The here you're after is built on caring, and caring is built on having skin in the game.

But I took issue with this idea. Whatever was evolving toward consciousness, toward some primitive form of AGI that would emerge fully-grown at some unknown point in the evolutionary tree, the thing evolving had as yet no understanding of “stakes”. The microbe carried no Darwinian theory. It just held together, or it didn’t. Felt stakes can’t be the cause of the very thing they require. I called Claude on this. It couldn’t be right.

You're right, and I had the cart before the horse. Felt stakes already require a someone who feels them, so they can't be what produces the someone. The microbe had no stakes it knew of and no theory driving it; it just held together, or it didn't. So stakes aren't the input — they're the output, the signature of arrival, not its cause. But you've put your finger on the thing underneath. The rock just sits; the microbe does the one thing the rock can't — it maintains itself, holds its own boundary against the world. That's not stakes it feels; it's the organism being a stake, a self-maintaining process. There's a name for it — autopoiesis, self-making — and a line of thought, Maturana and Varela, later Thompson, that calls it the seed of mind. The first thing that keeps itself alive is the first thing the world can matter to.

So at first I thought that maybe I’m off to a bad start when I try to differentiate between rocks — whether they sit on the side of a mountain, or we give them wheels and a mathematically-capable neural network and call them bots — and microbes. The rock is inert. It sits there. Or it rolls along. The microbe does something the rock and the bot can’t. It holds itself together against the world; it maintains itself.

But the bot with its wheels and its neural net and all sits quietly, thoughtlessly — notice that you’re never sitting at your computer when an LLM suddenly exclaims “you know, I’ve been thinking”; it only appears to “think” in response to a prompt; it’s purely reactive like any other non-conscious computer program — on the side of the rock. Being-in-itself.

The bot did not make itself. Bots are not self-making. We hold its boundary together. Not the bot.

How Do We Get From There to Being Here?

That’s the question, isn’t it? That’s the whole question. I think we don’t. But not for the same reason that Dreyfus thought we don’t.

Why we don’t is, as philosophers like to say, overdetermined.

First, we can’t tell an AI how to become an AGI. We don’t even know how to do it. We don’t even know how we did it. Where it was in the evolutionary line? (I, for one, am convinced humans aren’t the only ones that have it but that’s an entirely different article from this one.)

Second, we can’t demonstrate it. That is, we can’t show it. The best we could do is “act conscious”. But AI won’t be able to see inside our heads. And even if by some miracle it could do that, it would have already achieved AGI to appreciate what it was seeing.

Third, we can’t teach it because we never learned it. At some point in our lives, we just find that we have it.

Fourth, we can’t even name it if we’re being honest. The philosophical fights over consciousness, intentionality, subjectivity, Self — might as well throw in autopoiesis — are long and without winners. But even within what I’m going to call “the AI community” of non-philosopher tech bros and their minions there is no agreement on what AGI means. Sam Altman has called it “a sloppy term” and “not a super useful term” and notes that there are multiple definitions being used by different companies and individuals. As another article notes, The AGI race has no finish line and everyone is arguing about where it starts.

All four reasons bring us back to the top of this essay — to the piece I woke up with and had to write down. We can’t tell the way. We can’t show the way. All anyone or anything could ever do is go the way. But we can’t hand anything the Going because Being Here is the one place you can’t get to from there, from where AI is right now.

Evolution gave us a wonderful gift. We don’t know how. We don’t know why. And we don’t know how to pass it on.

Books of Interest

• Gilbert Ryle, The Concept of Mind (1949) — knowing-how vs. knowing-that

• Michael Polanyi, The Tacit Dimension (1966) — “we know more than we can tell”

• Martin Heidegger, Being and Time (1927) — Dasein, thrownness

• Jean-Paul Sartre, Being and Nothingness (1943) — in-itself / for-itself

• Hubert L. Dreyfus, What Computers Still Can’t Do (1992; orig. 1972) — GOFAI critique

• Hubert & Stuart Dreyfus, Mind Over Machine (1986) — “experts just drive", plus embodiment argument

• Thomas Nagel, What Is It Like to Be a Bat? (2024 ed.)

• Humberto Maturana & Francisco Varela, The Tree of Knowledge (1987) — autopoiesis, the readable version

• Evan Thompson, Mind in Life: Biology, Phenomenology, and the Sciences of Mind (2007) — autopoiesis to mind; life-mind continuity

• Anil Seth, Being You: A New Science of Consciousness (2021) — predictive processing and consciousness; “prediction machine” thread

• Daniel Dennett, From Bacteria to Bach and Back: The Evolution of Minds (2017) — “competence without comprehension”, evolution to mind; argues against my conclusion

• David Chalmers, The Conscious Mind: In Search of a Fundamental Theory (1996) — the hard problem itself; more technical

• Hubert Dreyfus, Being-in-the-World: A Commentary on Heidegger’s Being and Time, Division I (1991) — his Heidegger commentary, the bridge between my two pillars

* All Links are Amazon Affiliate Links