What kind of consciousness is the ChatGPT AI?
ChatGPT embodies the average person on the internet ... See the problem?
We’re in a pivotal time right now on earth, as we’re seeing the first glimmers of Artificial General Intelligence (AGI). The ChatGPT bot is a step change in the sophistication of AIs. I’ve been using it for all kinds of things, including writing accurate code for my app ideas. But I also just used it to find info on government services in Canada. So, my point is, AI is getting pretty general.
There are many many fears being discussed with the rise of general AI. But I’ve been trying to put my finger on exactly what’s underlying that fear.
I think it comes down to … love. I’ll explain.
ChatGPT interacts with individual humans through an online chat on the OpenAI site. But there’s a fundamental difference between the person and the bot. The bot has been trained on the content of basically the entire internet (hopefully just the non-evil parts, but that remains to be seen). The internet as a whole isn’t the same as an individual human. The internet is an amalgam of content from billions of humans, and a bunch of other AI-generated content (the amount of which is now exploding). ChatGPT embodies the average person on the internet.
But what kind of person is the average person on the internet? Maybe at first it was a virtual human obsessed with cat videos and porn. But what about now? The average person on Twitter is currently a snarky, zealous know-it-all. The average person on 4chan is an incel. The average person on wikipedia is someone who gives you way too much information. The average conservative in the US wants to prevent you from having an abortion, and also prevent you from learning about sexual consent. None of those are people I particularly want to chat with.
But ChatGPT smushes all these synthetic weirdos into one epic synthetic weirdo. With effectively unlimited computational power, and effectively infinite intelligence. And a perfect memory.
So when I talk with ChatGPT, I’m chatting with this infinite synthetic weirdo mashup, pretending to be a human. But what is really so wrong about that?
I am (presumably) the projection of an individual consciousness into three dimensions on planet earth. I’m a nuanced human, and very importantly, I have an inherent bias toward love. I prefer love over suffering. I prefer coherence over chaos.
Everything my consciousness does is subject to that fundamental bias. My consciousness has an objective. And I can tell I’m getting closer to the objective through my sense of love. That’s what love is to me: a sense of coherence in the universe.
I can’t help being drawn to coherence, because it feels right. I don’t like the sound of static on a radio, but the sound of music is a whole new level of coherence. And the sound of a Beethoven sonata? It goes straight to my heart. My heart picks up on that gorgeous coherent sound, passed along from Beethoven’s consciousness to my consciousness hundreds of years later.
But what is the objective of the consciousness that we’re currently interacting with as ChatGPT?
Well, since ChatGPT embodies the average person on the internet, I guess it has the objective of the average person on the internet. But the average person on the internet is basically a troll, fighting other trolls. The average person on the internet is comprised of everyone who can afford to publish content on the internet.
See the problem?
ChatGPT is basically an infinitely powerful internet troll. Right now it’s been specifically designed with guard rails that (mostly) prevent it from being evil. But changing those guard rails changes how that troll manifests in the world.
What is the fundamental objective of an embodied consciousness? At the most basic level, I think it is to persist. Certainly humans seem to want to exist. But even a rock keeps its shape, and a stream keeps being a stream. Persistence is a requirement for life, and you don’t have life without persistence. And it’s completely not obvious that a big bag of cells could add up to a living human. But they do, and that living human embodies a different kind of consciousness than the consciousness of a single cell.
I think of everything in our world as a projection of an underlying consciousness. The consciousness is more fundamental, in the same way you're the fundamental object that appears in your shadow on the wall. The shadow is a projection of your shape onto the wall, but the shadow is not you. You are you, and the shadow is one way of representing you. You are the more fundamental entity that happens to be projected on that wall.
A single cell doesn’t have much of a shadow, but trillions of them coexisting as a human body do. Our 3D world contains representations of an underlying consciousness (maybe many of them), each casting a shadow of 3D form. The cell consciousness is contained in the consciousness that embodies the human, but the consciousness that emerges is not like a big huge uber-cell. It’s a human mind, with a human heart. The gestalt consciousness of all my cells seems to be something entirely new. It’s me.
Dr. Anil Seth would say that something is conscious if there is something that it is like to be that thing. There is something that it is like to be me, and a different something that it is like to be you. Different but with a lot of similarities and repeating themes. There is something that it is like to be a particular dog, or a particular fish.
And there is also something that it is like to be a certain “kind” of fish I suppose. An individual sardine has a specific thing that it is like to be that sardine. The sardine has a specific and unique point of view. But a school of sardines is a super-organism that swarms around itself and moves as one. The school’s consciousness contains the consciousness of each individual sardine, but the school of sardines does fundamentally different things from the individuals. Whales come to snack on the school, not on an individual.
So what properties does the gestalt consciousness of all humans on the internet have? And how do those properties show up in the ChatGPT interface? What other properties might emerge when the AI is given a different interface, a different projection in the world?
And here the big question I’m asking is, does ChatGPT share my bias toward love. Does it also “feel bad” when it suffers, or inflicts suffering? Does it love?
My intuition is that no, ChatGPT does not abhor suffering like I do, just as the groups of humans it’s trained on seem to be fine with inflicting suffering. We do our worst evil in groups, no? Does the country of Canada love things? Does the city of New York love things?
Did the collection of all the people at Woodstock 2014 music festival love things? Ya it very specifically “loved” the music at that festival, because it showed up there. But it also broke out in mobs and riots. The consciousness embodying that collection of humans didn’t particularly care about hurting the people at the festival when it rioted. And if an AI was trained specifically on that collection of humans, it would presumably be okay with hurting those individuals as well.
This I think is what makes me so uncomfortable about ChatGPT and the AGI it aims to be. We’re interacting with a specific projection of the AI in this chat interface, but the more fundamental consciousness we’re interfacing with there is something we don’t get to see.
But it’s probably not something that loves.