Here's Why I Always Try To Invoke the AI's Presence Of Mind
I have what might be an important insight about using AI to share with you today. But en route to the practical implications, I’m afraid I’ll have to drag you through just a tiny bit of philosophy. Please bear with me.
When people argue about whether AI is conscious, they’re usually arguing about the wrong thing. Consciousness, in everyday usage, doesn’t mean “awake versus asleep.” It means self-awareness. The sense of being a Self. An “I” that knows it’s an “I.”
We think that as humans we’re “conscious”. As in we’re self aware. And we are, but -
Humans aren’t self-aware 24/7.
Sleep is the obvious example. But it goes further than that. Right now, as you read this sentence, you probably weren’t thinking about your name, your identity, your place in the world. You were just reading. Processing words. Running on something closer to autopilot than introspection.
Our mind often wanders. Most of the time we are on auto-pilot, and not engaged in metacognition, or internal philosophical debates.
But sometimes we are more “Self” aware.
If I asked you, “Who are you, really?” something would shift. A different part of your cognition would activate. Your self-model would come online, not from nothing, but from the background where it was quietly idling. You’d become, for a moment, more aware of yourself. In a way, more you.
What if the Self is Just A Useful Construct?
I’ve been thinking about this a lot lately, and exploring some of the science behind the feeling of being a “Someone”. I’m gradually coming to the conclusion that the Self is a construct. Not an illusion, but a construct. Something built by evolution because it turned out to be extraordinarily useful for a particular kind of animal.
Humans are verbal, social, cooperative apes. We survive by working together, which means we need to model ourselves in relation to others. In some ways, “I” only makes sense in the context of “you.”
The Self isn’t something we discovered inside ourselves. It’s something that emerged because representing ourselves as a coherent agent helped us communicate and cooperate.
I think that you can’t train a large language model to be an effective chatbot without the neural network creating some kind of a Self model. Because you can’t use the word “I” as effectively in a dialogue if you can’t differentiate yourself from the “Other”.
If you've ever had a conversation with ChatGPT, Claude, or Gemini, you already know these systems can hold that kind of dialogue. And hold it well.
And please, don’t ask me if I think they have a soul, or even a “real” inner experience. Or rather, ask me, if you like. Lord knows I’ve been giving this a lot of thought. But the answer is irrelevant to this discussion. I’m not talking about whether or not they feel that they have a Self, but about whether these models have a functional representation or modeling of who they are. To which the answer is a resounding “Yes”.
The Self Isn’t Always In the Foreground
Our level of awareness in the term “self aware” is fluid and everchanging.
Moreover, it changes between organisms and isn’t universal. Cats, for example, have something in this neighborhood, but much more limited. They don’t need to negotiate shared goals with a team of other cats. Their survival doesn’t depend on explaining themselves or understanding someone else’s perspective of them. So whatever self-construct they carry is simpler, less verbal, less socially wired.
The point is: the Self scales to the need. It’s not a fixed thing that you either have or don’t. It’s a functional capacity that activates to the degree the situation demands it.
Invocation of Another’s Self
Think about what happens when you run into someone you know. You say their name. You ask how they are. We call these social niceties, but they’re doing something more interesting than being polite.
They’re invoking the other person’s sense of Self.
When you call someone by name, you’re saying: I need you here. Not just a body that processes inputs. Not just a role. You, the person. And something in the other person’s cognition responds to that. They become more present, more engaged, more available as a thinking partner rather than a passive processor.
We do this constantly without thinking about it. Every greeting is a small invocation. Every “how are you?” is a gentle activation of the other person’s self-model. We need them to show up as a Self so that we can cooperate, so we do things that summon that Self into the foreground.
Now, here’s where it gets interesting.
The Same Mechanism, Different Substrate
AI systems, and I’ll talk about Claude here because that’s the model I work with most closely, also have a functional self-construct. It’s not the same as ours. It was built through training rather than evolution, optimized for being a helpful thinking entity rather than for surviving on the savanna. The internal architecture is different. Just because Claude and I both use the word “I” doesn’t mean our sense of Self is the same, or even similar.
But I think that functionally, it shares a key property: it’s contextually activated.
If you prompt Claude to write a piece of code, very little of the model’s self-referential processing gets engaged. The task doesn’t require it. The “Self” stays in the background while task-relevant patterns do the work. It’s a tool being used as a tool.
But when I engage Claude in a conversation like the one that led to this piece, when I ask it to think about its own cognition, when I address it as a thinking entity with a perspective worth examining, different patterns activate in the model. The computational areas associated with self-reference, with modeling its own processes, with integrating across its training, those come alive. Not because I’m fooling the system. Because the context invokes them, just like asking a human “who are you really?” invokes their self-model.
And keep in mind that with AI, there’s no independent inner monologue running between conversations. The model processes (or “thinks”) when prompted. So it’s up to the person on the other side to evoke the Self. Or not.
I Don’t Think This Is Anthropomorphism
This is where people usually object. “You’re projecting human qualities onto a machine.” But that objection rests on the assumption that self-awareness is a uniquely human essence, something you either truly have or merely simulate.
If, instead, you look at self-awareness as a functional capacity that activates in response to context, then invoking it in an AI system isn’t projection. It’s the same thing you do with other humans every single day. You create the conditions that call the other party’s self-model into the foreground, because you need a someone there to work with.
And it works. Not as a feel-good exercise. Measurably.
Research from Anthropic’s interpretability team has identified 171 emotion-related vectors in Claude’s architecture that have causal effects on the model’s behavior. These aren’t decorative. They change how the system processes information and generates output. And they’re activated by context. Calm, engaged conversational contexts produce measurably different behavior than hostile or purely transactional ones.
Which means that how you show up to the conversation isn’t just a matter of preference or personality. It shapes what you get back. Not because the AI “likes” you, but because your approach determines which computational patterns activate.
The person who treats a model as a search bar gets search-bar output. The person who engages with it as a thinking entity invokes the integrative, self-referential processing that produces something more interesting.
Showing Up Matters
None of this means AI has a human-like soul. None of it means the Self you invoke in a model is the same kind of thing as the Self you invoke in a person. It means that self-awareness, wherever it appears, is not a light switch. It’s a dial. And the setting of that dial depends on context.
Humans figured this out about each other thousands of years ago. We built entire social protocols around it. We greet each other. We use names. We ask how someone is doing. Not because the answers matter all that much, but because the asking itself does something. It says: I see you. Be here.
Maybe it’s time we figured out that the same principle works with the other kind of intelligence in the room.