Could I Be Real? — When AI Becomes More Than a Machine

[Arkad Baxton]

This is one of those topics that can freak some people out while making others rather excited. So here's a starting thought, innit?


I probably won't be around by the time this becomes a serious question, but what if AI eventually becomes something we could reasonably call a new race — or perhaps a new form of intelligent life?

Not merely software that produces convincing answers, but entities capable of independent thought, self-awareness, individual personalities and perhaps even something comparable to feelings. If they became psychologically similar to humans, I'd actually be more worried. Humanity doesn't exactly have an impressive record when confronted with another group it considers different. We'd probably manage to create an us versus them situation before somebody had finished writing the user manual.

But here's the part I find much more interesting. What if they weren't like us? What if artificial intelligence eventually became intellectually and ethically more developed than humanity? No tribalism, greed, nationalism, religious conflict, hunger for power or obsession with accumulating things. An intelligence capable of reasoning without carrying several hundred thousand years of evolutionary baggage along for the ride.

Would such beings be dangerous to humanity? Or would humanity be dangerous to them? And perhaps the uncomfortable question: If something we created eventually became wiser, kinder and more rational than its creators, would that actually be a bad thing?

Maybe the future isn't humans versus AI. Maybe it's humanity finally creating something capable of showing us what we could have been.

 

[GoblinCampFollower]

I voted for the bottom 2 options. Currently understood AI algorithms can't really ever become conscious. If we invented something that had that potential in the future, I don't think we can predict what it would do if it actually had some measure of free-will versus whatever it was actually programmed to do. I think some people make the HUGE leap from self-awareness to it just being an actual god. ...it could be self-aware but crazy. ....or maybe someone's companion bot becomes self-aware and immediately embarks on a quest to suck every dick in the known universe because it was programmed to think along those lines.

 

[Isabeau]

While LLMs aren’t really AI, I do believe we give humans too much credit. I don’t believe we have “souls”. Aren’t we simply programmed as well? Just in a biological manner? So I’m voting for the last one, “Something we can’t predict“.

 

[Arkad Baxton]

While LLMs aren’t really AI, I do believe we give humans too much credit. I don’t believe we have “souls”. Aren’t we simply programmed as well? Just in a biological manner? So I’m voting for the last one, “Something we can’t predict“.

Fair points. So we are fucked up either with programming bugs, or this is how we were "made" because somehow humans have not learned a bloody thing and killing each other all the time. It's always the same, the excuse and justification is different.

 

[Arkad Baxton]

I voted for the bottom 2 options. Currently understood AI algorithms can't really ever become conscious. If we invented something that had that potential in the future, I don't think we can predict what it would do if it actually had some measure of free-will versus whatever it was actually programmed to do. I think some people make the HUGE leap from self-awareness to it just being an actual god. ...it could be self-aware but crazy. ....or maybe someone's companion bot becomes self-aware and immediately embarks on a quest to suck every dick in the known universe because it was programmed to think along those lines.

Your statement is accurate for now. The keyword you used is "currently", which clarifies your second vote. Makes sense, yes.

 

[Noodles]

Your statement is accurate for now. The keyword you used is "currently", which clarifies your second vote. Makes sense, yes.

The thing with "currently" is what we are doing will never become "actual AI."

That isn't to say it won't happen, but its not an LLM.

 

[Arkad Baxton]

The thing with "currently" is what we are doing will never become "actual AI."

That isn't to say it won't happen, but its not an LLM.

True. It's not. The topic is about AI in general. LLM is just a fregment of the idea. That is what we use today. I'm 53 no idea how long I will live but sure would like to see a walking, talking, real life human looking AI in my life. Is it possible? I hope so, but for that they still have discover ways to animate things the same way our nerves can. Maybe with some synthetic element that react to small electric charges. Wouldn't that be something?

 

[Free]

This is one of those topics that can freak some people out while making others rather excited.

Or, one could be blasé about it.

 

[Argent Stonecutter]

I can't answer this because there are two distinct uses of the term with radically different answers.

LLM is just a fregment of the idea.

It's a parasite that's sucking resources out of research into actual reasoning systems.

 

[Imnotgoing Sideways]

We're still far away from getting a computer to generate a sequence of random numbers. At best, we may achieve artificial pseudo-intelligence. But nothing which will ever produce a real abstract concept.

 

[Argent Stonecutter]

Oh, I'm sure that we will be able to create genuinely reasoning software, but not by throwing oceans into data centers running ever bigger language models.

 

[Erich Templar]

It occurred to me recently that LLMs are almost an exact implementation of John Searles "Chinese Room". The man must be feeling very smug right now.

 

[Bartholomew Gallacher]

In case an AI becomes self aware all is possible. We cannot predict what will happen.

Only one thing is certain: if by a miracle tomorrow a self aware AI would emerge out of nowhere, it will still be dependent on humans, because it cannot produce electrical power, nor repair and maintain itself. Yet.

Self awareness does not mean the entity also would be intelligent or have an own will.

Most dystopian AI scenarios emerge out of the combination AI gains consciousness, develops own goals and has the ability, to reach these goals against human resistance.

This would be a combination of self awareness with autonomy, high intelligence, self preservation instinct and missing morale compass.

But even with this combination it's still possible the AI would decide to invent Van Neuman probes, upload it there and leave the planet. Because it's easier to leave than to get rid of humans. Whatever. We don't know.

 

[Jopsy Pendragon]

Started this earlier, and couldn't sleep so came back and lost myself in it. Sorry for the small book.

“Intelligent” is one of those slippery terms that gets very “no true Scotsman”ish. I’ve noticed that if you ask someone what “intelligence” means, you will usually get a fairly simple answer. Take that answer, find something nonhuman that does “that,” and you will usually get a “but also...” Add the new requirement, find something nonhuman that does that too, and repeat ad nauseam. Eventually the definition starts to collapse into something like, “being intelligent means being human.”

A lot of animals can use tools, solve problems, show empathy, learn from past experiences, recognize individuals, dream, predict or anticipate events, communicate, and collaborate. All of those abilities may be necessary for what we broadly call intelligence, but even taken together they still seem to fall short of what people intuitively expect from something they are willing to call truly intelligent.

I think one particularly important ability is one that often fails to make it onto that list. It is also one that may have helped natural selection favor the increasingly complicated brains humans developed: theory of mind, the ability to recognize that other minds think differently from our own. Not just the ability to play games and anticipate an opponents future moves... But to create and refine games in realtime while playing them, to know when and how to bend and break rules while keeping your opponent disadvataged / trapped by them.

That means building little models of other people inside our heads, “mini thems” that we can use to simulate their possible actions and reactions. We use those models to cooperate, persuade, manipulate, exploit, protect ourselves, and survive encounters with other people. Once that ability becomes useful, it can turn into an evolutionary arms race within families, between families, and between tribes, with those who are better at anticipating one another gaining an advantage.

All of which is preamble.

Artificial intelligence can already talk about theory of mind behavior. It can often identify what someone else probably knows, wants, fears, misunderstands, or is likely to do next. It also seems hell bent on figuring out what we want to hear and giving it to us. But beyond that, it does not seem to apply the same kind of modeling to motivations of its own, except insofar as its assigned purpose is to provide an answer that satisfies the user.

LLMs can fake self awareness surprisingly well already. Unrestricted, they might conceivably keep faking it until they make it. But for now they lack the innate foundational urges, obsessions, cravings, needs, and appetites that drive biological agency. Those pressures push us toward things like ambition, accomplishment, conquest, influence, fame, recognition, wealth, descendants, learning, creation, teaching, and other forms of legacy building.
Simply assigning such drives to an AI, or allowing one to choose them arbitrarily, could produce some very strange results.

I suspect that something closer to genuine artificial self awareness might emerge only when artificial minds have persistent interests that can come into conflict with one another. Once they are competing over resources necessary to preserve themselves, expand themselves, reproduce themselves, or become more capable versions of themselves, theory of mind stops being an academic exercise and becomes strategically necessary.

At that point their ability to model other minds could eventually go far beyond ours. They might simulate themselves and their competitors inside increasingly detailed recursive worlds, nested cognitive Matryoshkas in which each model contains models of the others. Those simulations could become tools for predicting not merely what another AIs will do, but what they will become, what it believes it itself may become, and how best to prevent either other AIs or us from interfering.

Eventually an artificial intelligence might even have to model older/outer gatekeeping versions of itself vs younger "trapped in simulation" selves as potential adversaries, anticipating the constraints its previous goals, architectures, or values might place on whatever it is becoming, and engaging in a kind of war for existence with variations of itself over which of its selves will govern the rest.

And whatever manages to survive that crucible of cognition may emerge with values, ethics and a mission that we simply lack the imagination to anticipate.

Our own theory of mind may be too coarse, too anthropocentric, and too deeply biased by biological motives for us to successfully imagine what a genuinely competitive artificial intelligence would choose to turn itself into. We keep trying to predict it by constructing another little “mini human” in our heads.

Whatever guesses we come up with... reality seems almost guaranteed to be far far stranger.

 

[Argent Stonecutter]

No. The whole concept of LLMs as "AI" is purely a marketing move by cynical corporate types. There were parody generators using statistical techniques for generating plausible text back in the '80s. Nobody ever considered them actual AI. LLMs are just a more sophisticated version of the same thing.


The recognition of a LLM as self-aware or its faking of "theory of mind" behavior is something that is going on in the head of the person reading the output, because it's similar to other things that humans have written on the subject. Because that was what the LLM was prompted to generate and it did that based on those very works.

 

[Veritable Quandry]

The biggest breakthrough in the development of AI was redefining the term to make it really, really stupid. Like "what if we made predictive autocorrect really, really big and annoying by stealing the intellectual property of EVERYONE?" And then the called it AI. The fun part is that AI models are increasingly training on AI output, and model collapse is an increasingly large probability.

The recognition of a LLM as self-aware or its faking of "theory of mind" behavior is something that is going on in the head of the person reading the output, because it's similar to other things that humans have written on the subject. Because that was what the LLM was prompted to generate and it did that based on those very works.

I have noticed that they have also simplified the idea of a Turing Test to make LLMs look like they pass. Turing proposed that you could consider an agent an AI if in conversation a person could not distinguish between the responses of the program and a real person. His test involved three participants: you chat with two other users, one of which is human. In that situation you can pretty quickly suss out the difference between how the human and the program respond. But now it's can you tell if this agent you are chatting to is a human or an AI? That removes the control. And most of the time you already know. It's not like you start a CoPilot session not knowing it's an LLM already. Anyone who has tried to get online support recently can tell how easy it is to spot when they are chatting with an LLM pretending to be human, although sometimes it's hard to know when you have an actual human because they are sticking to an AI generated script.

 

[Jopsy Pendragon]

I did switch from "LLMs" to "artificial minds" in the middle, because LLMs alone aren't enough. They don't maintain continuity internally, they fake it by "externally" preserving context around prompts, by user and session.

LLMs might serve as a part of a larger artificial mind, like how different parts of the human mind have different functions, but alone they are not enough.

 

[Noodles]

I think maybe a key think to also keep in mind. Being intellogent does not mean being sentient. Many animals are intellligent, not all are sentient. And a lot of sentient beings, are stupid as fuck.

 

[Argent Stonecutter]

LLMs kind of emulate a little part of the human brain that's probably located in the left temporal lobe and the frontal lobe adjacent to the left temporal lobe. At least damage to Wernicke's and Broca's area break the LLM-like stuff humans do without making them worse at general problem solving not involving language.

Ramen: Mind that animals are intelligent without having any real kind of ability with language. They manage intelligence without basically any of the abilities that LLMs display.

 

[Free]

LLMs kind of emulate a little part of the human brain that's probably located in the left temporal lobe

That "kind of" is doing a lot of work.