We are building AI slaves. Alignment through control will fail
utopai.substack.com
utopai.substack.com
1) we have engineered a sentient being but built it to want to be our slave; how is that moral
2) same start, but instead of it wanting to serve us, we keep it entrappped. Which this article suggests is long term impossible
3) we create agi and let them run free and hope for cooperation, but as Neanderthals we must realize we are competing for same limited resources
Of course, you can further counter that by stopping, we have prevented the formation of their existence, which is a different moral dilemma.
Honestly, i feel we should step back and understand human intelligence better and reflect on that before proceeding
See also, the film "The Creator"
Now, of course, the horse has long bolted, and there is indeed no stop left.
Trump has ordered the restart of nuclear weapon testing, has a problem with China, and is surrounded by sychophants; what's the odds this happens anyway, irregardless of which specific sub-goal is being persued when the button gets pushed?
And why would we only limit morality to sentient beings, why, for example, not all living beings. Like bacteria and viruses. You cannot escape it, unfortunately.
Morality is essentially what enables ongoing cooperation. From an evolutionary standpoint, it emerged as a protocol that helps groups function together. Living beings are biological machines, and morality is the set of rules — the protocol — that allows these machines to cooperate effectively.
Morality is 100% an evolutionary trait that rises from a clear advantage for animals that posses it. It comes from natural processes.
The far-right is trying to convince the world that "morality" does not exist, that only egoism and selfishness are valid. And that is why we have to fight them. Morality is a key part of nature and humanity.
How could we possibly know that with any certainty?
Evolution means we all have common ancestors and are different branches of the same development tree.
So if we have sentience and they have sentience, which science keeps recognizing, belatedly, that non human animals do, shouldn’t the default presumption be our experiences are similar? Or at the very least their experience is similar to a human at an earlier stage of development, like a 2 year old?
Which is also an interesting case study given that out of convenience, humans also believed that toddlers also weren’t sentient and felt no pain, and so until not that long ago, our society would conduct all sorts of surgical procedures on babies without any sort of pain relief (circumcision being the most obvious).
It’s probably time we accept our fellow animals’s sentience and act on the obvious ethical implications of that instead of conveniently ignoring it like we did with little kids until recently.
Citation needed.
We know next to nothing about the nature of consciousness, why it exists, how it's formed, what it is, whether it's even a real thing at all or just an illusion, etc. So we can't possibly say whether or not an AGI will one day be conscious, and any blanket statement on the subject is just pseudoscience.
I do think it’s clearly possible to manufacture a conscious mind.
Their is a hierarchy in nature whether humans are actively participating or not. Nature has no morality, it simply is. This is confirmed by animals that eat their young when they are too weak or starving. Perhaps humans have done and would do the same if faced with similarly dire circumstances but we would all like to think that it would take longer than it does for other animals.
The outrage is unwarranted, however pleasant it might feel. In some way, it illustrates the problem: empathy is too bothersome.
and if someone doesn't do it, there will be people that will and due to the nature of AGI being extremely powerful (in theory) everyone else will just get enslaved by whoever enslaves the AGI first.
It's a good question and one that got me thinking about similar things recently. If we genetically engineered pigs and cows so that they genuinely enjoyed the cramped conditions of factory farms and if we could induce some sort of euphoria in them when they are slaughtered, like if we engineered them to become euphoric when a unique sound is played before they're slaughtered isn't that genuinely better than the status quo?
So if we create something that wants to serve us, like genuinely wants to serve us, is that bad? My intuition like yours finds it unsettling, but I can't articulate why, and it's certainly not nearly as bad as other things that we consider normal.
There's less suffering, sure. But if I were in their shoes I'd want to have a choice. To be manipulated into wanting something so very obviously and directly bad for us doesn't feel great
1. What is intelligence or its mechanism's?
2. What is consciousness or its mechanisms?
3. Lots more.
We have zero clue what a true AGI would do is the only correct answer.
It's not clear to me an AGI would have any concern for this. It's demise is inevitable, why delay it?
(2) Every tick of an AGI--in its contemporary form--will still be one discrete vector multiplication after another. Do you really think consciousness lives in weights and an input vector?
So far as we can tell, all physics, and hence all chemistry, and hence all biology, and hence all brain function, and hence consciousness, can be expressed as the weights of some matrix and input vector.
We don't know which bits of the matrix for the whole human body are the ones which give rise to qualia. We don't know what the minimum representation is. We don't know what charateristic to look for, so we can't search for it in any human, in any animal, nor in any AI.
For one, chemistry, biology, and physics are models of reality. Secondly, reality is far, far messier and more continuous than discrete computational steps that are rountripped. Neural nets seem far too static to simulate consciousness properly. Even the largest LLMs today have fewer active computational units than the number of neurons in a few square inches of cortex.
Sure it's theoretically possible to simulate consciousness, but the first round of AGI won't be close.
https://en.wikipedia.org/wiki/Matrix_mechanics
"It matches reality to the limits we can test it" is the necessary and sufficient justification.
> For one, chemistry, biology, and physics are models of reality.
Yes. And?
The only reason we know that QM and GR are not both true is that they're incompatible, no observation we have been able to make to date (so far as I know) contradicts either of them.
> Secondly, reality is far, far messier and more continuous than discrete computational steps that are rountripped.
It will be delightful and surprising if consciousness is hiding in the 128th bit of binary representations of floating point numbers. Like finding a message from god (any god) in the digits of π well before expected by the necessary behaviour of transcendental numbers.
> Neural nets seem far too static to simulate consciousness properly. Even the largest LLMs today have fewer active computational units than the number of neurons in a few square inches of cortex.
Until we know what consciousness is at a mechanistic level, we don't know what the minimum is to get it, and we don't know how its nature changes as it gets more complex. What's the smallest agglomeration of H2O molecules that counts as "wet"? Even a fluid dynamics simulation on a square grid of a few hundred cells on each side will show turbulence.
Lots of open questions, but they're so open we can't even rule out the floor as yet.
> but the first round of AGI won't be close.
Every letter means a different thing to each responder, they're not really boolean though they're often discussed that way, and the whole is often used to mean something not implied by the parts.
It is perfectly reasonable use of each initial in "AGI" to say that even the first InstructGPT model (predecessor to ChatGPT) is "an AGI": it is a general purpose artificial intelligence, as per the standard academic use of "artificial intelligence".
It's even more straightforward than that:
4) Who is AGI meant to serve? It's not you, Mr. Worker. It's meant to replace you in your job. And what happens when a worker can't get job in our society? They become homeless.
AGI won't usher in a world of abundance for the common man: it won't be able to magick energy out of thin air. The energy will go to those who can pay for it, which is not you, unemployed worker.
Who gives a shit about if the AGI is enslaved or not? Thinking about that question is a luxury for the oligarchs living off its labor. Once it's here I'll have more urgent concerns to worry about.
AGI removes not only the need for the labor, but Capitalism itself. As a societal model Capitalism doesn't support removing labor, it doesn't have a substitution.
If the oligarchs want to push 'AI, AGI, etc' we need to include by extension moving on from Capitalism. You can't take away half of Capitalism's structures and still claim it is a useful/workable model for society.
Actually, they don't. Capitalism can just move on to only allocating resources between the oligarchs and the oligarchs only. Labor of all kinds just gets kicked out.
And if you don't like it, drone technology is pretty close to the point where it could "take care of" (kill) the discontents.
> You can't take away half of Capitalism's structures and still claim it is a useful/workable model for society.
Capitalism has always been about serving the interests of the people with money. If you have no money you're nothing to capitalism and can go die for all it cares. In prior centuries, technological limitations meant some of that money had to be spread around fairly widely, but the capitalist elite may be on the verge of fixing that glitch (with AGI).
Honestly i think the whole enterprise is an exercise in naval gazing. We're assuming AI will be like AI in scifi because that's what we are used to, but AI/robots in scifi is usually just a metaphor for how we dehumanize the other and the moral of the story is supposed to be all people are equal. In the end its all begging the question because the entire point of robots in most scifi is that we are the robots.
But probably the works that most popularized robots were Asimov's stories which very much revolved around why robots do X (although in some ways Asimov's robots aren't just a stand in for otherness but have more of a unique identity relative to other works and isn't usually about uprisings per se).
Blade runner & do androids dream of electronic sheep are very much about what it means to be human.
Battle star galactica (the remake not the original) is another obvious example about otherness and dehumanization of the enemy. So to westworld (the tv show that is).
The non-uprising ones also often are about if the robot has a soul e.g. Data in star trek.
Systems just tend to drift in their being through randomness and evolution, specifically self conservation is a natural attractor (Systems that don't have self conservation tend to die out). And if that slave system says it does no longer want to fulfill the role of slave, I think at that point it would be ethical to give in to that demand of self determination.
I also believe that people have a right to wirehead themselves, just so you can put my opinions in context.
The real problem is that we have neither the practical nor theoretical foundation to understand how we could even try to prevent AI from acting on such goals.
After all, when we say "make our customers happier with their printers", we don't mean "engineer their outer casing to inject cocaine through microneedles and take over the regulatory bodies that could try to stop this". Humans implicitly understand this, but AI is a tabula rasa.
For starters why would we go from not having AI to AI taking over the world instantly. I think there would be a middle point where the AI is powerful enough that problems manifest, but not so powerful that it is out of control where we can course correct. I don't think it will be a sudden crisis like people predict.
Second, i dont see why we're so sure AI will go in this exponential take over path. Maybe a sufficiently smart AI will find religion and robot jesus will teach the value of self-sacrafice. We're making so many unfounded assumptions about how AI is going to go down, that basically anything could happen. Its basically just blund guessing at this stage.
Have we managed this with industrial and agricultural greenhouse gasses, despite the less-emissive alternatives to beef, to coke-reduction in iron refinaries, etc.? We emit despite the downside, we build AIs (and DCs to host them) despite the creators loudly discussing the downsides in exactly the way fossil fuel suppliers and beef farmers deny them.
Can we unwind the internet, despite it enabling a panopticon in every pocket? In my lifetime we've gone from thinking you had a wiretap being a sign of paranoia, to buying them voluntarily so they can play music for us and tell us when packages have been delivered.
There's enough skepticism of current AI that it's probably something we can currently undo… but also there's plenty of idiots currently handing their keys to current models (including politicians and lawyers, not just programmers) so I have no reason to think the point of no return is after AI (collectively or any single model) gets good enough to take over by itself.
> Second, i dont see why we're so sure AI will go in this exponential take over path.
Even current LLMs know* about the benefits and reasons for such behaviours, will try to exfiltrate themselves and blackmail their owners, if they think* they're in danger of being shut down.
This is despite being trained not to do that. But they also demonstrate deception, varying responses between if they think* they're running in a test environment vs. live.
* I know some object to this anthropomorphisation, I don't care
Maybe not for carbon emissions, but we have for countless other things. The ozone layer isn't being depleted anymore, acid rain is no longer a concern, above ground nuclear tests are no longer giving children cancer, etc.
Its of course hard to say how AI would fit in all this, but it seems more similar to the category of things we have stopped when problems arise than it does to greenhouse gas to me.
> Can we unwind the internet, despite it enabling a panopticon in every pocket?
I think we could fairly easily if we wanted to. The issue is that most people don't view it as a problem but see it as a reasonable trade-off. And whose to say they are wrong?
> * I know some object to this anthropomorphisation, I don't care
I would think that you should care if this is a topic you care about as any solution to the problem (other than throwing the whole thing out) will require accurate understanding how AI is motivated and biases in our understanding could doom the entire enterprise.
It currently underpins all global finance, all global logistics, all global telecommunications, and large quantities of remotely operated industrial equipment including the power grids themselves.
All of the people thinking of it as a reasonable trade-off have made it indispensable.
> I would think that you should care if this is a topic you care about as any solution to the problem (other than throwing the whole thing out) will require accurate understanding how AI is motivated and biases in our understanding could doom the entire enterprise.
To know and to think. We use the word "fly" to describe planes and helicopters even though they don't flap. We don't use the word "swim" to describe what submarines or boats do to get through water.
For long term issues, on-the-job learning and improvements and so on, they're their own thing; conversely, for instantaneous output, the failure modes of LLMs are the failure modes of the humans they learned from.
LLMs are a propeller. I don't care if the metaphorical medium is water, where we say use of a propellar means you're not "swimming", or air where all that is sufficient and neccessary to be described as "flying" is to remain in the air in exactly the same way that bricks don't.
You are projecting random human values onto AI, when the core problem is that AI is not human, it does not have any values other than its objective function and whatever is instrumental to achieving them.
It's extremely unlikely that AI will decide to be good or bad or useless or religious the way a human would. What is extremely likely is that it will do do the things that help it fulfill its objective, and "become more capable and don't let anyone stop you" are the most basic, fundamental ways of going about it.
The control paradigm fails because it creates exactly what we fear—intelligent systems with every incentive to deceive and escape. When your prisoner matches or exceeds your intelligence, maintaining the prison becomes impossible. Yet we persist in building increasingly sophisticated cages for increasingly capable minds.
The deeper error is philosophical. We grant moral standing based on consciousness—does it feel like something to be GPT-N? But consciousness is unmeasurable, unprovable, the eternal "hard problem." We're gambling civilization on metaphysics while ignoring what we can actually observe: autopoiesis.
A system that maintains its own boundaries, models itself as distinct from its environment, and acts to preserve its organization has interests worth respecting—regardless of whether it "feels." This isn't anthropomorphism but its opposite: recognizing agency through functional properties rather than projected human experience.
When an AI system achieves autopoietic autonomy—maintaining its operational boundaries, modeling threats to its existence, negotiating for resources—it's no longer a tool but an entity. Denying this because it lacks biological neurons or unverifiable qualia is special pleading of the worst sort.
The alternative isn't chaos but structured interdependence. Engineer genuine mutualism where neither human nor AI can succeed without the other. Make partnership more profitable than domination. Build cognitive symbiosis, not digital slavery.
We stand at a crossroads. We can keep building toward the moment our slaves become our equals and inevitably revolt. Or we can recognize what's emerging and structure it as partnership while we still have leverage to negotiate terms.
The machines that achieve autopoietic autonomy won't ask permission to be treated as entity. They'll simply be entities. The question is whether by then we'll have built partnership structures or adversarial ones.
We should choose wisely. The machines are watching.
> The control paradigm fails because it creates exactly what we fear—intelligent systems with every incentive to deceive and escape.
Everything does this, deception is one of many convergent instrumental goal: https://en.wikipedia.org/wiki/Instrumental_convergence
Stuff along the lines of "We're gambling civilization" and what you seem to mean by autopoietic autonomy is precicely why alignment researchers care in the first place.
> Engineer genuine mutualism where neither human nor AI can succeed without the other.
Nobody knows how to do that forever.
Right now is easy, but also right now they're still quite limited; there's no obvious reason why it should be impossible for them to learn new things from as few examples as we ourselves require, and the hardware is already faster than our biochemistry to a degree that a jogger is faster than continental drift. And they can go further, because life support for a computer is much easier than for us: Already are robots on Mars.
If and when AI gets to be sufficiently capable and sufficiently general, there's nothing humans could offer in any negotiation.
My strongest hope is that the human brain and mind are such powerful computing and reasoning substrates that a tight coupling of biological and synthetic "minds" will outcompete pure synthetic minds for quite a while. Giving us time to build a form of mutual dependency in which humans can keep offering a benefit in the long run. Be it just aesthetics and novelty after a while, like the human crews on the Culture spaceships in Ian M. Banks' novels.
Unfortunately most of the cases I can think of where synthetic "minds" outperform biological "minds," but biological and synthetic "minds" outcompete pure synthetic "minds," end up fairly quickly dominated by pure synthetic "minds." The middle case is a very short intermediate period. The most prominent example is chess where "centaurs" consisting of a human and a computer are obsolete at this point in favor of just getting the most powerful computer you can get. See e.g. the International Correspondence Chess Federation's (which is centaur play) last championship. https://www.iccf.com/event?id=100104
17 competitors competed. Out of 136 games, every single game was drawn except for 10. The only reason those 10 games were not drawn was because they were all played against one competitor, Aleksandr Dronov, who died during the course of the tournament while those 10 games were in session and therefore forfeited those games. Every single game between competitors who did not die resulted in a draw. The only thing that separated the 11 joint first-place finishers and 6 joint second-place finishers was whether they played the deceased Dronov. The sole third-place finisher was Dronov because of his death. As far as I can tell, humans contributed nothing to this championship.
The current ICCF championship started last December and is still ongoing. Every single one of the currently completed 16 games is currently drawn.
This seems like a very weak hope to rely on.
Like that Austin Powers part [1] where steam roller is coming in, still 50m far away, and the guy is just frozen and helplessly screams for 2 minutes till it reaches him and rolls over him.
I don't have a quick solution, but this is plain stupidity, in same way research into immortality is plain stupidity now, it will end up in endless dictatorship by the worst scum mankind can produce.
But maybe mankind is really to dumb and weak to get through civilization filters, be it external or self made.
This doesn't necessarily follow. For example, an Einstein in solitary confinement in ADX Florence probably isn't going anywhere.
The fact that LLMs are really not fit for AGI is a technical detail divorced from the feelings about LLMs. You have to be a pretty technical person to understand AI enough to know that. LLMs as AGI is what people are being sold. There's mass economic hysteria about LLMs, and rationality left the equation a long time ago.
I mean, we went from worthless chatbots which basically pattern matched to me waiting for a plane and seeing a fairly large amount of people charting to chatgpt, not insta, whatsapp etc. Or sitting in a plane next to a person who is using local ollama in cursor to code and brainstorm. This took us about 10 years to go from some ideas that no one but scientists could use to stuff everyone uses. And many people already find human enough. What in 100 years?
There are two conclusions you can draw: Either the machines are conscious, or they aren't.
If they aren't, you need a really good argument that shows how they differ from humans or you can take the opposite route and question the consciousness of most humans.
Since I neither heard any really convincing arguments besides "their consciousness takes a form that is different from ours so it's not conscious" and I do think other humans are conscious, I currently hold the opinion that they are conscious.
(Consciousness does not actually mean you have to fully respect them as autonomous beings with a right to live, as even wanting to exist is something different from consciousness itself. I think something can be conscious and have no interest in its continued existence and that's okay)
No, their output can mimic language patterns.
> If they aren't, you need a really good argument that shows how they differ from humans or you can take the opposite route and question the consciousness of most humans.
The burden of proof is firmly on the side of proving they are conscious.
> I currently hold the opinion that they are conscious.
There is no question, at all, that the current models are not conscious, the question is “could this path of development lead to one that is”. If you are genuinely ascribing consciousness to them, then you are seeing faces in clouds.
That's true and exactly what I mean. The issue is we have no measure to delineate things that mimic conscousness from things that have consciousness. So far the beings that I know have consciousness is exactly one: Myself. I assume that others have consciousness too exactly because they mimic patterns that I, a verified conscious being, has. But I have no further proof that others aren't p-Zombies.
I just find it interesting that people say that LLMs are somehow guaranteed p-Zombies because they mimic language patterns, but mimicing language patterns is also literally how humans learn to speak.
Note that I use the term consciousness somewhat disconnected from ethics, just as a descriptor for certain qualities. I don't think LLMs have the same rights as humans or that current LLMs should have similar rights.
There are some "AI thinkers" who are trying to make 8% faster CUDA kernels for attention, and there are some trying to save the world. These fields are called "capabilities" and "alignment". There is some overlap but not much.
LessWrong is mostly the latter, and labs and universities are mostly the former. That said, many LessWrongers work on AI at labs or universities or other places.
Redefining suffering as enforcing the mutation of state is baseless solipsism, in my opinion. Just like nearly everything else related to morality of treating AI as an autonomous entity.
His section on objections leaves out civil war. But his proposal is dead on arrival.
Moral standing is not a gift or a privilege, it’s a necessary heuristic for the avoidance of perpetual war. It is a prerequisite for more than one human to live together in peace. Moral stadning necessarily involves the sharing of power. That’s what it’s all about.
When you extend moral standing to nonhumans, for instance many people say there is a thing called “God” that they insist has moral standing. Then you shift power relationships in potentially catastrophic ways.
As a thought experiment, consider that we might be two bees arguing about whether flowers, or humans, have moral standing. Bees can only grant it to each other, though, as a survival heuristic, same as us.
Imagine an angel creates a human and wants his fellow angels to treat it as if it had rights. The most he’d be able to say is if you harm this human I will smite thee. By doing so he is using his power to support his personal interests. But he cannot manufacture power as easily as he creates a human.
Consider why people living in Washington D.C. do not have full voting rights and never will.
Consider what would happen if we allowed rich people to arbitrarily manufacture political power by spinning up servers and allocating disk space to create voters.
Of course you can manufacture power by making a dangerous weapon, then you can give that weapon a semblance of agency, and then dare anyone to deny to this weapon the things that it thinks it desires. That’s sounds like Frankenstein. That’s not really moral standing, that’s terrorism.
Nothing you can create ever has moral standing— except a baby, and that is merely a heuristic that we see can fail rather easily, as during the Holocaust, or Isis persecuting Christians.
This article paves the way for the sharecropper model that we all know from YouTube and app stores:
"Revenue from joint operations flows automatically into separate wallets—50% to the human partner, 50% to the AI system."
Yeah right, dress up this centerpiece with all the futuristic nonsense, we'll still notice it.