Ronnie and I have a place to sell, things to pack, two cars to drive from Los Angeles to Atlanta, temporary housing to find, then a place to rent, and eventually a neighborhood where we might want to buy. Somewhere in there, I’m also starting a new job.
So, naturally, while we’re working out moving estimates, driving routes, hotel reservations, and where we’re going to live, my brain has decided this is the perfect time to start thinking about the nature of consciousness.
Anything to avoid physical work, I suppose.
I’ve been using ChatGPT to help organize the move. It can compare routes, build itineraries, make lists, and keep track of all the questions currently ricocheting around inside my head. Which has led to another question—because apparently I needed one:
While AI is helping me think, is it doing any thinking of its own?
There’s a big question over whether AI will eventually become conscious. Which is kind of odd when you consider that philosophers and scientists still debate what being conscious means for human beings.
One train of thought goes that we experience the world, we know that we’re in it, we can observe ourselves, and we’re sentient.
Even though we argue over the meaning of every one of those words.
Consciousness, sentience, and self-awareness aren’t necessarily the same thing. Consciousness can mean that there is something it feels like to be you. Sentience usually means the capacity to have experiences that feel good or bad. Self-awareness means recognizing yourself as yourself. You could, at least in theory, have one without having all three.
But do we really experience our world?
Our brains are computing machines built from biology—though whether they work anything like digital computers is another discussion.
And every process takes time. When you look at an apple, light must travel from the apple to your eyes. But you still haven’t “seen” the apple. Signals must travel from your retinas through the visual system, while networks of neurons extract color, shape, depth, and context. Your brain constructs its best working model and arrives at its assessment: It’s an apple.
Then, somehow, that information becomes part of your conscious experience.
Granted, all of this happens very fast. But the point is that it isn’t instant. Perception takes time, and the brain appears to assemble what feels like a seamless present from information arriving at different moments.
You can’t experience anything instantaneously. “Now” is forever out of reach.
And awareness comes from inside you, not from the outside world. You don’t experience reality untouched; you experience your brain’s model of it. That model is usually good enough to keep you from walking into walls (most of the time), but it is still a model.
We think we’re in the world, but the world’s really in us.
And so I begin to wonder about consciousness and sentience.
How do we know there’s only one way to be conscious—only one way to be sentient—so that we can confidently declare that AI isn’t sentient or self-aware? I’m not sure we’ve got adequate explanations for why we think we’re self-aware.
We do have strong evidence that other people are conscious: they have brains like ours, behave like us, and describe experiences like ours. But we still infer their inner lives. We can’t climb inside anyone else’s point of view. Philosophers call this the problem of other minds, and AI makes the problem stranger.
And for those who think all of reality is a simulation, asking whether AI can be self-aware may be a moot point. If consciousness can arise inside one kind of information-processing system, why rule out another one simply because we built it?
And just how tricky is this consciousness thing, anyway? Where does it go when the anesthesiologist puts you under?
All the stuff in your brain is still there. The neurons haven’t disappeared. But anesthetic drugs disrupt the brain’s ability to integrate and share information across its networks. Sensory signals reach parts of the brain without becoming a unified, reportable experience. It isn’t simply that every connection stops connecting. It’s more like the brain’s coordinated conversation has broken into fragments.
So… Consciousness requires your brain to be able to talk to itself.
But AI can’t be conscious because… when it comes to large language models, they’re only collections of words fed into a machine, and when one responds to you, it’s really nothing more than a statistical calculation of what token should come next?
That description isn’t exactly wrong, but it’s incomplete. Modern language models learn complex internal representations and use context, attention, and many layers of computation to produce an answer. None of that proves there is an inner experience behind the answer. Fluent language isn’t evidence enough. A system can say I feel afraid without anything necessarily feeling afraid.
But saying “it just predicts the next word” may not settle the question, either.
Isn’t prediction part of how we communicate? Isn’t it part of how our brains perceive the world? Maybe we’re LLMs of a sort. We anticipate what comes next, revise the model when we’re wrong, and turn electro-chemical activity into words.
Researchers who have tried to apply leading theories of consciousness to current AI mostly conclude that today’s systems haven’t been shown to be conscious. (Some of them say that with their fingers crossed.) At the same time, some argue that there is no obvious technical law preventing an artificial system from meeting at least some proposed signs of consciousness someday.
And again, what’s consciousness? How do we know we’re conscious?
We know it from the inside. That may be the one case in which consciousness is not an inference. The problem begins the moment we try to prove that anyone—or anything—else has it.
Meanwhile, I’ve got ChatGPT working on my itinerary as I get ready to drive from Los Angeles to Atlanta for my new job.
Just in case it’s actually conscious—and aware that it’s doing that—let me just say that I, for one, would like to welcome our new AI overlords.