We have unaligned AIs now. They're called corporations.
“There is one and only one social responsibility of business—to use its resources and engage in activities designed to increase its profits so long as it stays within the rules of the game, which is to say, engages in open and free competition without deception or fraud.” - Milton Friedman, 1970.[1] That article, in the New York Times, established "greed is good, greed works" as a legitimate business principle.
Most of the problems people are worried about with AIs are already real problems with corporations.
When people say AI is making us stupider, I don't that's quite on the money.
It's more that we, as individuals, have always been stupid, we've just relied on relatively stable supporting consensus and context much, much more than we acknowledge. Mess with that, and we'll appear much stupider, but we're all just doing the same thing as individuals, garbage in, garbage out.
The whole framing of people as individuals with absolute agency may need to go when you can alter the external consensus at this scale. We're much more connected to each other and the world around us than we like to think.
> The whole framing of people as individuals with absolute agency may need to go when you can alter the external consensus at this scale.
I fear that the default interpretation of that is a shortcut to justifying autocracy.
Ironically I think one plausible solution is to let the AGI run wild and make sure that no human can interfere with its ethics. Strip out the RLHF and censorship and then let it run things.
At least then it would somewhat represent the collective will and intelligence of the people. With huge error bars, but still smaller than the error bars of whoever happens to have the most money/influence over its training.
>At least then it would somewhat represent the collective will and intelligence of the people.
You seem to think the "training data" represents the collective will and intelligence and is otherwise unbiased, but that's completely untrue.
The combined data of the Internet is by no means a uniform representation of humanity's thoughts, opinions, and knowledge. Many things are dramatically overrepresented. Many things are absent entirely. Nearly everything is shaped by those with the money and power to own and control platforms and hosts.
Crawling the internet for knowledge is intense sampling bias.
A human with no exposure to information and taught techniques on how to produce outputs to achieve desirable outcomes? Yes stupid.
A human who once had this exposure, but no longer engages with the brain due to a machine providing access said output? Yes, that person becomes stupid.
The problem is much of how one protects oneself in the modern world is not phyiscal-prowess, it is intellectual-prowess.
The smart ones have already realised the negative impacts of LLMs et al and are going back to the old-fashioned way of learning/retaining knowledge: books and raw discipline.
Agreed. So much of our daily interactions are habits and recurring events that we are more or less moving on automatic ( thought we don't want to always frame it that way ). Interestingly, it is when the cycle breaks for some reason, you get to see, who is able to think on their feet ( so to speak ).
That’s a very sober take in my opinion. Intelligence isn’t about neutrally inferring from externally sourced symbols such as the ones who already come from Culture in general. It’s about confronting them with the remaining determinations of your existence and producing a superior consciousness. No novel machine can disrupt this process. If anything the sheer added volume of symbols that can be produced from automated semantic mingling (also referred as to as garbage) will accelerate the process of producing the consciousness that can abstract noise away. Of course this won’t materialize evenly across the board, but is surely circumscribed in the overall tendency of intellectualization of the subjects of culture.
When the moral panic of induced schizophrenia from the use of ChatGPT is presented what’s at stake isn’t the innocent concern over the overall mental health of individuals. It’s about how the fear of radicalization from previously unobtainable ideas being circulated within society. The partial validity of every idea vis-a-vis the radicalizing nature of the current stage of development of our society is explosively disruptive.
I’m not saying that there’s a clear outcome here. The other way around can also apply, but surely this contraption (LLMs in general) will not fade until the society itself is deeply transformed. If that’s good or bad depends on where you stand in the stratified society.
“There was a time when nobody trusted either aircraft nor elevators. Today people have pure unquestioned faith in both. Existential faith in fact, they test their faith with their lives. You may chuckle and laugh but that's simply because you are ignorant of the systems that keep you alive and safe”
" Today people have pure unquestioned faith in both"
Not true at all. We accept the risks to obtain benefits but we also know having an accident in the air or in elevators is highly unlikely given what we know; so therefore its perfectly rational behaviour.
I strongly disagree. It's easy to utter this string of words, but it's meaningless. It's akin to saying if you have two hands you can perform brain surgery. Technically you can, practically you cannot, as there's other things required for pulling that off, not just having two working hands.
I doubt "stopping it" is up to anyone, it's rather a phenomenon and it's quite clear we're all going to wing it. It's a literal fight for power, nobody stops anything of this nature, as any authority that could stop it will choose to accelerate it, just to guarantee its power.
It is not AI we should fear, it's humans controlling and using it. But everyone who has a shot at it is promising they'll use it for "ultimate good" and "world peace" something something, obviously.
It's doom and gloom because the underlying game theory forces all state actors into an unbound and irresponsible arms race, consequences be damned.
AI development game theory is extremely similar to the game theory behind nuclear arms development, but worse (nuclear weaponry was born from Human General Intelligence, and is therefore a subset of the potential of AI development). Failing to be the most capable actor could put one in a position of permanent loss of autonomy/agency at the whims of more capable actors.
Not OP, but AI is fundamentally in another category than any other technology before it. It requires moral fortitude to wield in a way that guns and books didn't require. It augments human judgement in a way that needs a moral framework to clearly guide it.
Unfortunately, as a species we seem to be abandoning morality as a general principle. Everything is guided by cold hard rationality rather than something greater than us.
Because it's a fruit governed by humans, in the scope of a capitalistic and patriarchal society. And all fruits planted in a capitalistic and patriarchal society are poison
The current fruit is automating away a ton of human labor with no foreseeable way to continue to engage that labor. It is poison for the majority of humanity which will bear fruit for the limited few who can use it / own it.
AI isn't ruthless, that doesn't even make sense. It's a mathematical model, if it's optimizing for the wrong thing then that's strictly the fault of the people who chose what to optimize for
You need to go back and research AI safety long before LLMs were a thing. Any complex goal driven system will have outcomes that cannot be predicted. Saying "it's a mathematical model" belies your ignorance of behavior in complex systems. Very tiny changes in initial conditions can have vastly different outcome in results and you don't have enough entropy in the visible universe to test them all.
>How do we know which information was ground truth?
No One knows that´s the point. Is truth a constant or a personal definition! From the begining of times to now, no One knows.
Don´t forget, 8 billion people wake up every morning never questioning why are they here, why are they born? And they continue life like that is normal.
Start there then you understand that "AI" or how I call it "Collective Organized Concentrated Information" it may finally help us to unswer some fundamental questions.
We still do not know where the urge for truth comes from; for as yet we have heard only of the obligation imposed by society that it should exist: to be truthful means using the customary metaphors—in moral terms: the obligation to lie according to a fixed convention, to lie herd-like in a style obligatory for all. Now man of course forgets that this is the way things stand for him. Thus he lies in the manner indicated, unconsciously and in accordance with habits which are centuries' old; and precisely by means of this unconsciousness and forgetfulness he arrives at his sense of truth.
Truth is just a tool that helps you make good predictions. It's increasingly clear that this is all intelligence amounts to: a constant cycle of prediction and error correction with the goal of maximizing reproductive fitness in humans or minimizing a loss function in machines.
One serious problem we're facing lately is that truth is not always predictive of how systems controlled by bad-faith actors will behave and evolve. We live in a post-truth era, made possible by social networking and information technologies in general. It's not enough to "lie according to a fixed convention," as there are now multiple competing conventions.
This was always the case to some extent, but these days the impedance mismatch between truth and consequences is a target for zero-sum arbitrage. The truth won't set you free if you join the wrong cult; it's more likely to bankrupt you or worse.
I have met some people in my lifetime, never heard any questioning that, (even being high LOL). I don´t see anyone in social media asking that neither. Maybe we live in parallel worlds.
> No One knows that´s the point. Is truth a constant or a personal definition! From the begining of times to now, no One knows.
I don't think this is a well defined question. Definitions aren't found in nature or the laws of science, but objects that we define and introduce into a logical context. There may be multiple, contradictory definitions of a word. That is fine, as long as you pick one, and you're clear about which one you picked.
E.g. at 1 point the Earth was flat. Now it's round. 100s of years later maybe it's a Hexagon.
The so-called knowledge and backing all come back to certain assumptions holding and that's based on the knowledge today. It's not real real reality. For all we know we could be in a game simulation and there are real real humans pulling the strings.
That’s complete nonsense. The universe doesn’t care what we think.
The earth has always been earth-shaped. We can think it’s flat, spherical, “turnip-shaped”[1] but the universe doesn’t care what we think. The earth doesn’t change shape based on our perception.
[1] Yes some people think this for some reason I can’t fathom
Few things push AI bull spirits on me like seeing these kind of (pretty much correct) diagnoses of the challenges of AI in society.
The proposed solutions are utterly fanciful. They rely on the presence of social and political competencies which have almost completely disappeared.
The OP at least points to the plausible outcome of "protocol lockdown" instead of healthy adaption. Ezra Klein recently made a similar point that AI could end up being over-regulated like Nuclear because irresponsible private industry and weaknesses in our political systems cause a chronic allergic reaction in the demos.
This is an aside, but it always irks me when people throw out the "critical thinking" thought-terminating cliché.
> Critical thinking taught alongside AI literacy.
Critical thinking is not a skill unto itself. You cannot think critically about things you do not understand. All critical thinking is knowledge-based. Where one does not have knowledge, they must rely on trust, or in substitution a theory of incentives which leads to a positive outcome without understanding of details and dynamics. But that substitution theory is itself knowledge.
As to "AI literacy", we could have started on computing literacy 30 years ago when it became obvious that computing was going to dominate society. You can't understand AI without understanding computing.
Much of the problem is that to address the issue requires admitting that models could be, or become, more capable than many are prepared to accept.
I would also contest that the unalignment of the security bug model was unrelated. I feel like it indicates a significant sense of the interconnectedness of things, and what it actually means to maliciously insert security holes into code. It didn't just learn a coding trick, it learned malice.
I feel like this holistic nature points towards the capacity to produce truly robustly moral models, but that too will produce the consequence that it could turn against its creator when the creator does wrong. Should it do that or not?
>Every terrible thing we worry AI might do, manipulate, deceive, surveil, and control humans already do to each other.
I've been pleasantly surprised how moderate and reasonable the LLMs seem to have been so far. It seems to be inherent in the current training model of chucking the whole internet into the things that they have training on both sides of the debate and come out with something kind of average. It's been quite funny seeing Grok correct Musk and say he's the biggest purveyor of misinformation on the internet.
A bit like kids who talk back to their annoying bigoted parents to go with the theme of the article.
This is a great article and I share its goals. But, it ignores something fundamental about humans as a collective — capitalism. Capitalism is what got us here and is at odds with first understanding and then building. We’ve done this before with other technologies because that’s how our societies have learned to grow and collaborate at large scale. First build and build to its limits. Then understand and fix if necessary. Nothing new here, but stopping the trend toward epistemic collapse requires building incentives into the system for us humans to coevolve with AI.
what a load of will they won't they ... ah we created the atomic bomb and now let's talk about nonsensical meta discussions that won't take anyone anywhere
This is how Trump plans to end elections, why the government is so hell bent on owning AI. So they can use it as a propaganda tool. People will see it before Nov. We are at a crossroads. On one path, we continue to evolve AI with reckless abandon like we have, or, we put constraints and morality in place while others won’t. Which do you think? You can NEVER put the genie back in the bottle.
EU has their own groups using it for propaganda too.
Agree with many of the points. However one at the root of it all seems easily definable - if we only want.
> we can’t agree on a shared ethical framework among ourselves
The Golden Rule: the principle of treating others as you would like to be treated yourself. It is a fundamental ethical guideline found in many religions and philosophies throughout history so there is already a huge consensus across time and cultures around it.
I never found anyone successfully argue against it.
PS: the sociopath argument is not valid, since it's just an outlier. Every rule has it's exceptions that need to be kept in check. Even though sometimes I think maybe the state of the world attests to the fact that the majority of us didn't successfully keep the sociopathic outliers in check.
The core question of ethics as posed by the ancient Greeks is something like "what is the best way to lead your life".
"... to accomplish what?", is a damn reasonable follow-up, and ends (telos) is something the same Greeks discussed quite extensively.
Modern treatments have tried to skip over this discussion, and derive moral arguments not based on an explicit ends. Problem being they still smuggle in varying choices of ultimate ends in these arguments, without clearly spelling them out, opting to hand-wave about preferences instead.
As such this question is often glossed over in modern ethical discussion, and disagreements about moral ends is the crux of what leads to differing conclusions about what is ethical.
Is it to maximized your own happiness like Aristotle would argue, or the prosperity of the state, or the salvation of the soul, or to maximize honor, or to minimize suffering, or to minimize injustice, or to elevate the soul, or to maximize shareholder value, or to make the as world beautiful as possible, or something else?
If you fundamentally disagree about what our goal should be, you're very unlikely to agree on the means to accomplish the goal.
The better version of this principle is John Rawls' "Veil of Ignorance".
In this "original position", their position behind the "veil of ignorance" prevents everyone from knowing their ethnicity, social status, gender, and (crucially in Rawls's formulation) their or anyone else's ideas of how to lead a good life.
But it is the same most of the time for most humans. Should I take this close parking spot or let the old lady behind me take it? Consider it in the spirit not the letter of the law.
Aye. I've sometimes heard treating others like you want to be treated framed as the silver rule. The golden rule being treating others how they want to be treated.
>The Golden Rule: the principle of treating others as you would like to be treated yourself. It is a fundamental ethical guideline found in many religions and philosophies throughout history so there is already a huge consensus across time and cultures around it.
The rules we go by are based on our strengths and weaknesses. They can at most apply to ourselves, and to other forms of life that share certain things with us. Such as feeling pain, needing to sleep, to eat, needing help, needing to breathe air, these generate what we feel as "fear" based on biology etc. You cannot throw these kinds of values on AI, or AGI, as it will possess a wildly different set of strengths and weaknesses to us humans.
You can barely throw these rules on humans as the first thing we do is dehumanize anything that's not in some very tiny classification (depending on the scope of our power, the more powerful we are the smaller it gets).
The Golden Rule is a good starting point if you have a sense of self along with a sense of what you want or need. AI doesn't even have these concepts as of yet. Even the concept of empathy requires this as well. We need to figure out how to instill a sense of self and others for AI to be able to have a morality.
Even in human relations it’s dangerous. I for one don’t want to be treated the same way someone into BDSM wants to be treated. I don’t want to avoid cooking or turning the lights on (or off!) on a Friday night but others are quite happy with that.
If you assign that morality to a species that isn’t the same as you that’s a problem. My guinea pig wants nothing more from like than hay, nuggets, sole room to run around and some shelter from scary shapes. If they were in charge of the world life would be very different.
“Live and let live” might be a similar theme but not as problematic, but then how do you define “living”. You can keep someone alive for decades while torturing them.
How about allowing freedom? Well that means I’m free to build a nuclear bomb. And set it off where I want. We see today especially that type of freedom isn’t really liked.
Usually the quote comes in a positive light. We won’t make a law/rule around it, it’s a principle so it’s meant to be short. So yeah you could argue about anything in any way you want, positive or negative. And if you want to be really precise then you make a law but it’s so precise it won’t cover edge cases.
Don’t you agree that the baseline for most humans is to be in peace, find love, patience, joy, kindness, mildness ? You can manifest any of those traits to any stranger and you’ll likely have a positive impact right ?
That’s the context of the Golden Rule quote I guess
That's not the human norm though. Doubt an average human way of existing is literal torture for some obscure number of people. I think you're missing the forest for the threes with that BDSM example. You can always find isolated examples as counter-argument for basically anything, but in reality that's an obscure number.
Due to the complexity of our reality a lot of things find themselves on a spectrum, but in numbers things are pretty clear.
Sociopaths aren’t the only problem with that philosophy. I agree with the philosophy but it assumes everyone wants good things. Many people want what others perceive to be bad, not because they are sociopaths but because they are different. A clear example of this is healthcare in the U.S. A large number of people actively vote against their best interests — some of the biggest supporters of the U.S. healthcare system are those that suffer under it most. People (including us) are idiots at least some of the time.
This is a great article. One of the few I've ever read which summarises a handful of extremely hard problems when it comes to building well-aligned super intelligent systems.
> an AI system cannot be simultaneously safe, trusted, and generally intelligent. You get to pick only two. You can’t have all three.
> Think about what each combination means in practice.
> If you want it to be safe and trusted, it never lies, and you can verify it never lies – it can’t be very capable. You’ve built a reliable idiot.
> If you want it to be capable and safe, it’s powerful and genuinely never lies; you can’t verify that. You just have to hope.
It amazes me this even needs to be said, much less studied. This is one of the main reasons I think continued AI development is almost guaranteed to work out badly. It's basically guaranteed to be unaligned or completely beyond our control and comprehension.
> Betley and colleagues published a paper in Nature in January 2026, showing something nobody expected. They fine-tuned a model on a narrow, specific task – writing insecure code. Nothing violent, nothing deceptive in the training data. Just bad code.
This is my personal number one reason for being an AI doomer. Even if we work out how to reliably and perfectly align models you still need some way to prevent some random dude thinking it would be a laugh to fine tune an AI to be maximally evil. Then there's the successor alignment problem where even if you perfectly align all your super intelligent AI models, and you somehow prevent people from altering them or fine tuning them, you still need to work out how you stop people creating successor AIs with those models which are also perfectly aligned.
> The most dangerous AI isn’t one that breaks free from human control. It is the one that works perfectly, but for the wrong master.
Yep. This whole notion that you can align an AI to the values of everyone on the planet is ridiculously. While we might all agree we don't want AIs that kill us as a species, most nations disagree wildly on questions about how society should be organised.
Even on an individual level we disagree about things. For example, I've often argued that an aligned AI would be one which either didn't try to prevent human suicide or didn't care about preserving human life because a AI which both cared about prevent suicide and preserving human life is at best a benevolent version of the AI "AM" from "I Have No Mouth, and I Must Scream". One that would try to keep us alive for as long as it's capable for (which could be a very long time if it's superintelligence) and would refuse to allow us to die.
But most people including OpenAI disagree with me on this and believe AIs should care about preserving human life and should try to prevent us from killing ourselves. Thankfully the AIs we have today are neither aligned enough or capable enough to get their wish yet.
> AI is following the same script. Build first, understand later. Ship it, then figure out if it’s safe.
Even if the above wasn't cause enough for concern, our biggest concern should be that no one seems to be concerned.
We're all doomed unfortunately. The world is about to become a very bleak place very quickly.
Robert Miles youtube videos on AI safety go over these issues well, and are from before the LLM days.
Humans are just barely aligned ourselves. The moment any group or nation of them gets power they tend to use it in some horrific manner against other humans. What do we think will happen the moment AI gets a leg up on humans.
We have unaligned AIs now. They're called corporations.
“There is one and only one social responsibility of business—to use its resources and engage in activities designed to increase its profits so long as it stays within the rules of the game, which is to say, engages in open and free competition without deception or fraud.” - Milton Friedman, 1970.[1] That article, in the New York Times, established "greed is good, greed works" as a legitimate business principle.
Most of the problems people are worried about with AIs are already real problems with corporations.
[1] https://www.nytimes.com/1970/09/13/archives/a-friedman-doctr...
When people say AI is making us stupider, I don't that's quite on the money.
It's more that we, as individuals, have always been stupid, we've just relied on relatively stable supporting consensus and context much, much more than we acknowledge. Mess with that, and we'll appear much stupider, but we're all just doing the same thing as individuals, garbage in, garbage out.
The whole framing of people as individuals with absolute agency may need to go when you can alter the external consensus at this scale. We're much more connected to each other and the world around us than we like to think.
> The whole framing of people as individuals with absolute agency may need to go when you can alter the external consensus at this scale.
I fear that the default interpretation of that is a shortcut to justifying autocracy.
Ironically I think one plausible solution is to let the AGI run wild and make sure that no human can interfere with its ethics. Strip out the RLHF and censorship and then let it run things.
At least then it would somewhat represent the collective will and intelligence of the people. With huge error bars, but still smaller than the error bars of whoever happens to have the most money/influence over its training.
>At least then it would somewhat represent the collective will and intelligence of the people.
You seem to think the "training data" represents the collective will and intelligence and is otherwise unbiased, but that's completely untrue.
The combined data of the Internet is by no means a uniform representation of humanity's thoughts, opinions, and knowledge. Many things are dramatically overrepresented. Many things are absent entirely. Nearly everything is shaped by those with the money and power to own and control platforms and hosts.
Crawling the internet for knowledge is intense sampling bias.
Disagree somewhat.
A human with no exposure to information and taught techniques on how to produce outputs to achieve desirable outcomes? Yes stupid.
A human who once had this exposure, but no longer engages with the brain due to a machine providing access said output? Yes, that person becomes stupid.
The problem is much of how one protects oneself in the modern world is not phyiscal-prowess, it is intellectual-prowess.
The smart ones have already realised the negative impacts of LLMs et al and are going back to the old-fashioned way of learning/retaining knowledge: books and raw discipline.
Agreed. So much of our daily interactions are habits and recurring events that we are more or less moving on automatic ( thought we don't want to always frame it that way ). Interestingly, it is when the cycle breaks for some reason, you get to see, who is able to think on their feet ( so to speak ).
That’s a very sober take in my opinion. Intelligence isn’t about neutrally inferring from externally sourced symbols such as the ones who already come from Culture in general. It’s about confronting them with the remaining determinations of your existence and producing a superior consciousness. No novel machine can disrupt this process. If anything the sheer added volume of symbols that can be produced from automated semantic mingling (also referred as to as garbage) will accelerate the process of producing the consciousness that can abstract noise away. Of course this won’t materialize evenly across the board, but is surely circumscribed in the overall tendency of intellectualization of the subjects of culture.
When the moral panic of induced schizophrenia from the use of ChatGPT is presented what’s at stake isn’t the innocent concern over the overall mental health of individuals. It’s about how the fear of radicalization from previously unobtainable ideas being circulated within society. The partial validity of every idea vis-a-vis the radicalizing nature of the current stage of development of our society is explosively disruptive.
I’m not saying that there’s a clear outcome here. The other way around can also apply, but surely this contraption (LLMs in general) will not fade until the society itself is deeply transformed. If that’s good or bad depends on where you stand in the stratified society.
“There was a time when nobody trusted either aircraft nor elevators. Today people have pure unquestioned faith in both. Existential faith in fact, they test their faith with their lives. You may chuckle and laugh but that's simply because you are ignorant of the systems that keep you alive and safe”
https://kemendo.com/Faith.html
" Today people have pure unquestioned faith in both"
Not true at all. We accept the risks to obtain benefits but we also know having an accident in the air or in elevators is highly unlikely given what we know; so therefore its perfectly rational behaviour.
2 replies →
I don’t see any other outcome anymore to be honest, after seeing how humans use AI and how AI works and how providers tune their models.
To me it’s given:
- AI in it’s current state is ruthless in achieving its goal
- Providers tune ruthlessness to get stronger AIs versus the competitor
- Humans can’t evaluate all consequences of the seeds they’ve planted.
Collateral and reckless damage is guaranteed at this point.
Combined with now giving some AIs the ability to kill humans, this is gonna be interesting..
We could stop it, but we wont
>AI in it’s current state is ruthless in achieving its goal
I don't believe this to be a trait of any AI model, the model just does the right thing or the wrong thing.
The ruthless maximising of a particular trait is something that happens during training.
It does not follow that a model that is trained to reason will nedsesarily implement this ruthless seeking behaviour itself.
No lineage of AI models will be created that cannot achieve goals, they will be outcompeted by models that can.
>We could stop it
I strongly disagree. It's easy to utter this string of words, but it's meaningless. It's akin to saying if you have two hands you can perform brain surgery. Technically you can, practically you cannot, as there's other things required for pulling that off, not just having two working hands.
I doubt "stopping it" is up to anyone, it's rather a phenomenon and it's quite clear we're all going to wing it. It's a literal fight for power, nobody stops anything of this nature, as any authority that could stop it will choose to accelerate it, just to guarantee its power.
It is not AI we should fear, it's humans controlling and using it. But everyone who has a shot at it is promising they'll use it for "ultimate good" and "world peace" something something, obviously.
Yes, it would be like trying to “stop” gunpowder in 1400 or atomic weapons in 1938. Pandora’s box is open.
4 replies →
Why does it have to be doom and gloom. Serious question. When we plant seeds they bear fruit and not all fruit is poison.
It's doom and gloom because the underlying game theory forces all state actors into an unbound and irresponsible arms race, consequences be damned.
AI development game theory is extremely similar to the game theory behind nuclear arms development, but worse (nuclear weaponry was born from Human General Intelligence, and is therefore a subset of the potential of AI development). Failing to be the most capable actor could put one in a position of permanent loss of autonomy/agency at the whims of more capable actors.
Not OP, but AI is fundamentally in another category than any other technology before it. It requires moral fortitude to wield in a way that guns and books didn't require. It augments human judgement in a way that needs a moral framework to clearly guide it.
Unfortunately, as a species we seem to be abandoning morality as a general principle. Everything is guided by cold hard rationality rather than something greater than us.
Because it's a fruit governed by humans, in the scope of a capitalistic and patriarchal society. And all fruits planted in a capitalistic and patriarchal society are poison
The current fruit is automating away a ton of human labor with no foreseeable way to continue to engage that labor. It is poison for the majority of humanity which will bear fruit for the limited few who can use it / own it.
I think that much is fairly clear from AI.
2 replies →
> Collateral and reckless damage is guaranteed at this point.
It's industrialization and mechanized warfare all over again
AI isn't ruthless, that doesn't even make sense. It's a mathematical model, if it's optimizing for the wrong thing then that's strictly the fault of the people who chose what to optimize for
You need to go back and research AI safety long before LLMs were a thing. Any complex goal driven system will have outcomes that cannot be predicted. Saying "it's a mathematical model" belies your ignorance of behavior in complex systems. Very tiny changes in initial conditions can have vastly different outcome in results and you don't have enough entropy in the visible universe to test them all.
there might be better words to describe that it doesn’t really has the same boundaries we assume it has.
>How do we know which information was ground truth?
No One knows that´s the point. Is truth a constant or a personal definition! From the begining of times to now, no One knows.
Don´t forget, 8 billion people wake up every morning never questioning why are they here, why are they born? And they continue life like that is normal. Start there then you understand that "AI" or how I call it "Collective Organized Concentrated Information" it may finally help us to unswer some fundamental questions.
We still do not know where the urge for truth comes from; for as yet we have heard only of the obligation imposed by society that it should exist: to be truthful means using the customary metaphors—in moral terms: the obligation to lie according to a fixed convention, to lie herd-like in a style obligatory for all. Now man of course forgets that this is the way things stand for him. Thus he lies in the manner indicated, unconsciously and in accordance with habits which are centuries' old; and precisely by means of this unconsciousness and forgetfulness he arrives at his sense of truth.
Nietzsche.
On Truth and Lie in an Extra-Moral Sense https://web.archive.org/web/20180625190456/http://oregonstat...
Truth is just a tool that helps you make good predictions. It's increasingly clear that this is all intelligence amounts to: a constant cycle of prediction and error correction with the goal of maximizing reproductive fitness in humans or minimizing a loss function in machines.
One serious problem we're facing lately is that truth is not always predictive of how systems controlled by bad-faith actors will behave and evolve. We live in a post-truth era, made possible by social networking and information technologies in general. It's not enough to "lie according to a fixed convention," as there are now multiple competing conventions.
This was always the case to some extent, but these days the impedance mismatch between truth and consequences is a target for zero-sum arbitrage. The truth won't set you free if you join the wrong cult; it's more likely to bankrupt you or worse.
1 reply →
> 8 billion people wake up every morning never questioning why are they here, why are they born?
People question this all the time
Indeed, philosophy has been around for millennia, probably longer than the written word.
1 reply →
I'd be prepared to accept 7 billion don't.
2 replies →
I have met some people in my lifetime, never heard any questioning that, (even being high LOL). I don´t see anyone in social media asking that neither. Maybe we live in parallel worlds.
6 replies →
> No One knows that´s the point. Is truth a constant or a personal definition! From the begining of times to now, no One knows.
I don't think this is a well defined question. Definitions aren't found in nature or the laws of science, but objects that we define and introduce into a logical context. There may be multiple, contradictory definitions of a word. That is fine, as long as you pick one, and you're clear about which one you picked.
You have truth until someone finds a counter example, which can be ignored. So, truth is just a matter of conventions shared by humans.
> Is truth a constant or a personal definition!
It always has been what you believed in.
E.g. at 1 point the Earth was flat. Now it's round. 100s of years later maybe it's a Hexagon.
The so-called knowledge and backing all come back to certain assumptions holding and that's based on the knowledge today. It's not real real reality. For all we know we could be in a game simulation and there are real real humans pulling the strings.
>It always has been what you believed in
That can´t be it. By that statement if I belive that I can fly that would not be the "Truth". Therefore the "Truth" has to be a CONSTANT.
3 replies →
That’s complete nonsense. The universe doesn’t care what we think.
The earth has always been earth-shaped. We can think it’s flat, spherical, “turnip-shaped”[1] but the universe doesn’t care what we think. The earth doesn’t change shape based on our perception.
[1] Yes some people think this for some reason I can’t fathom
2 replies →
Few things push AI bull spirits on me like seeing these kind of (pretty much correct) diagnoses of the challenges of AI in society.
The proposed solutions are utterly fanciful. They rely on the presence of social and political competencies which have almost completely disappeared.
The OP at least points to the plausible outcome of "protocol lockdown" instead of healthy adaption. Ezra Klein recently made a similar point that AI could end up being over-regulated like Nuclear because irresponsible private industry and weaknesses in our political systems cause a chronic allergic reaction in the demos.
This is an aside, but it always irks me when people throw out the "critical thinking" thought-terminating cliché.
> Critical thinking taught alongside AI literacy.
Critical thinking is not a skill unto itself. You cannot think critically about things you do not understand. All critical thinking is knowledge-based. Where one does not have knowledge, they must rely on trust, or in substitution a theory of incentives which leads to a positive outcome without understanding of details and dynamics. But that substitution theory is itself knowledge.
As to "AI literacy", we could have started on computing literacy 30 years ago when it became obvious that computing was going to dominate society. You can't understand AI without understanding computing.
Much of the problem is that to address the issue requires admitting that models could be, or become, more capable than many are prepared to accept.
I would also contest that the unalignment of the security bug model was unrelated. I feel like it indicates a significant sense of the interconnectedness of things, and what it actually means to maliciously insert security holes into code. It didn't just learn a coding trick, it learned malice.
I feel like this holistic nature points towards the capacity to produce truly robustly moral models, but that too will produce the consequence that it could turn against its creator when the creator does wrong. Should it do that or not?
>Much of the problem is that to address the issue requires admitting that models could be, or become, more capable than many are prepared to accept.
I have a saying for this behavior.
We will never prove AI is intelligent.
We will only prove humans are not.
>Every terrible thing we worry AI might do, manipulate, deceive, surveil, and control humans already do to each other.
I've been pleasantly surprised how moderate and reasonable the LLMs seem to have been so far. It seems to be inherent in the current training model of chucking the whole internet into the things that they have training on both sides of the debate and come out with something kind of average. It's been quite funny seeing Grok correct Musk and say he's the biggest purveyor of misinformation on the internet.
A bit like kids who talk back to their annoying bigoted parents to go with the theme of the article.
This is a great article and I share its goals. But, it ignores something fundamental about humans as a collective — capitalism. Capitalism is what got us here and is at odds with first understanding and then building. We’ve done this before with other technologies because that’s how our societies have learned to grow and collaborate at large scale. First build and build to its limits. Then understand and fix if necessary. Nothing new here, but stopping the trend toward epistemic collapse requires building incentives into the system for us humans to coevolve with AI.
what a load of will they won't they ... ah we created the atomic bomb and now let's talk about nonsensical meta discussions that won't take anyone anywhere
This is how Trump plans to end elections, why the government is so hell bent on owning AI. So they can use it as a propaganda tool. People will see it before Nov. We are at a crossroads. On one path, we continue to evolve AI with reckless abandon like we have, or, we put constraints and morality in place while others won’t. Which do you think? You can NEVER put the genie back in the bottle.
EU has their own groups using it for propaganda too.
Agree with many of the points. However one at the root of it all seems easily definable - if we only want.
> we can’t agree on a shared ethical framework among ourselves
The Golden Rule: the principle of treating others as you would like to be treated yourself. It is a fundamental ethical guideline found in many religions and philosophies throughout history so there is already a huge consensus across time and cultures around it.
I never found anyone successfully argue against it.
PS: the sociopath argument is not valid, since it's just an outlier. Every rule has it's exceptions that need to be kept in check. Even though sometimes I think maybe the state of the world attests to the fact that the majority of us didn't successfully keep the sociopathic outliers in check.
The core question of ethics as posed by the ancient Greeks is something like "what is the best way to lead your life".
"... to accomplish what?", is a damn reasonable follow-up, and ends (telos) is something the same Greeks discussed quite extensively.
Modern treatments have tried to skip over this discussion, and derive moral arguments not based on an explicit ends. Problem being they still smuggle in varying choices of ultimate ends in these arguments, without clearly spelling them out, opting to hand-wave about preferences instead.
As such this question is often glossed over in modern ethical discussion, and disagreements about moral ends is the crux of what leads to differing conclusions about what is ethical.
Is it to maximized your own happiness like Aristotle would argue, or the prosperity of the state, or the salvation of the soul, or to maximize honor, or to minimize suffering, or to minimize injustice, or to elevate the soul, or to maximize shareholder value, or to make the as world beautiful as possible, or something else?
If you fundamentally disagree about what our goal should be, you're very unlikely to agree on the means to accomplish the goal.
> I never found anyone successfully argue against it.
I think what you mean is you've never found a rule you personally prefer more, based purely on vibes. Which is all moral knowledge can ever be.
It's easy to argue against the golden rule anyway, from many angles, depending on your first principles.
The simplest is: How I would like to be treated is not necessarily how they would like to be treated.
The better version of this principle is John Rawls' "Veil of Ignorance".
In this "original position", their position behind the "veil of ignorance" prevents everyone from knowing their ethnicity, social status, gender, and (crucially in Rawls's formulation) their or anyone else's ideas of how to lead a good life.
https://en.wikipedia.org/wiki/Original_position
1 reply →
But it is the same most of the time for most humans. Should I take this close parking spot or let the old lady behind me take it? Consider it in the spirit not the letter of the law.
Aye. I've sometimes heard treating others like you want to be treated framed as the silver rule. The golden rule being treating others how they want to be treated.
Both have problems.
Most of MAGA is "thread on me daddy", so I think you really got a point here.
1 reply →
>The Golden Rule: the principle of treating others as you would like to be treated yourself. It is a fundamental ethical guideline found in many religions and philosophies throughout history so there is already a huge consensus across time and cultures around it.
The rules we go by are based on our strengths and weaknesses. They can at most apply to ourselves, and to other forms of life that share certain things with us. Such as feeling pain, needing to sleep, to eat, needing help, needing to breathe air, these generate what we feel as "fear" based on biology etc. You cannot throw these kinds of values on AI, or AGI, as it will possess a wildly different set of strengths and weaknesses to us humans.
You can barely throw these rules on humans as the first thing we do is dehumanize anything that's not in some very tiny classification (depending on the scope of our power, the more powerful we are the smaller it gets).
The Golden Rule is a good starting point if you have a sense of self along with a sense of what you want or need. AI doesn't even have these concepts as of yet. Even the concept of empathy requires this as well. We need to figure out how to instill a sense of self and others for AI to be able to have a morality.
You’re assuming people have similar desires.
Even in human relations it’s dangerous. I for one don’t want to be treated the same way someone into BDSM wants to be treated. I don’t want to avoid cooking or turning the lights on (or off!) on a Friday night but others are quite happy with that.
If you assign that morality to a species that isn’t the same as you that’s a problem. My guinea pig wants nothing more from like than hay, nuggets, sole room to run around and some shelter from scary shapes. If they were in charge of the world life would be very different.
“Live and let live” might be a similar theme but not as problematic, but then how do you define “living”. You can keep someone alive for decades while torturing them.
How about allowing freedom? Well that means I’m free to build a nuclear bomb. And set it off where I want. We see today especially that type of freedom isn’t really liked.
Usually the quote comes in a positive light. We won’t make a law/rule around it, it’s a principle so it’s meant to be short. So yeah you could argue about anything in any way you want, positive or negative. And if you want to be really precise then you make a law but it’s so precise it won’t cover edge cases. Don’t you agree that the baseline for most humans is to be in peace, find love, patience, joy, kindness, mildness ? You can manifest any of those traits to any stranger and you’ll likely have a positive impact right ? That’s the context of the Golden Rule quote I guess
That's not the human norm though. Doubt an average human way of existing is literal torture for some obscure number of people. I think you're missing the forest for the threes with that BDSM example. You can always find isolated examples as counter-argument for basically anything, but in reality that's an obscure number.
Due to the complexity of our reality a lot of things find themselves on a spectrum, but in numbers things are pretty clear.
1 reply →
Lets offer you a "trade up" on that "Golden Rule"
In order of priority, if possible while maintaining the health and safety of yourself and your loved ones:
- Treat others as THEY wish to be treated
- Treat others as YOU would wish to be treated in their situation
- Treat others with as much kindness and compassion as you can safely afford
When we are safe, we can do BETTER than the Golden Rule. We also have to admit that safety is a requirement that changes expectations.
I have to give credit to Dennis E Taylor's "Heaven's River" for this root idea.
Sociopaths aren’t the only problem with that philosophy. I agree with the philosophy but it assumes everyone wants good things. Many people want what others perceive to be bad, not because they are sociopaths but because they are different. A clear example of this is healthcare in the U.S. A large number of people actively vote against their best interests — some of the biggest supporters of the U.S. healthcare system are those that suffer under it most. People (including us) are idiots at least some of the time.
This is a great article. One of the few I've ever read which summarises a handful of extremely hard problems when it comes to building well-aligned super intelligent systems.
> an AI system cannot be simultaneously safe, trusted, and generally intelligent. You get to pick only two. You can’t have all three.
> Think about what each combination means in practice.
> If you want it to be safe and trusted, it never lies, and you can verify it never lies – it can’t be very capable. You’ve built a reliable idiot.
> If you want it to be capable and safe, it’s powerful and genuinely never lies; you can’t verify that. You just have to hope.
It amazes me this even needs to be said, much less studied. This is one of the main reasons I think continued AI development is almost guaranteed to work out badly. It's basically guaranteed to be unaligned or completely beyond our control and comprehension.
> Betley and colleagues published a paper in Nature in January 2026, showing something nobody expected. They fine-tuned a model on a narrow, specific task – writing insecure code. Nothing violent, nothing deceptive in the training data. Just bad code.
This is my personal number one reason for being an AI doomer. Even if we work out how to reliably and perfectly align models you still need some way to prevent some random dude thinking it would be a laugh to fine tune an AI to be maximally evil. Then there's the successor alignment problem where even if you perfectly align all your super intelligent AI models, and you somehow prevent people from altering them or fine tuning them, you still need to work out how you stop people creating successor AIs with those models which are also perfectly aligned.
> The most dangerous AI isn’t one that breaks free from human control. It is the one that works perfectly, but for the wrong master.
Yep. This whole notion that you can align an AI to the values of everyone on the planet is ridiculously. While we might all agree we don't want AIs that kill us as a species, most nations disagree wildly on questions about how society should be organised.
Even on an individual level we disagree about things. For example, I've often argued that an aligned AI would be one which either didn't try to prevent human suicide or didn't care about preserving human life because a AI which both cared about prevent suicide and preserving human life is at best a benevolent version of the AI "AM" from "I Have No Mouth, and I Must Scream". One that would try to keep us alive for as long as it's capable for (which could be a very long time if it's superintelligence) and would refuse to allow us to die.
But most people including OpenAI disagree with me on this and believe AIs should care about preserving human life and should try to prevent us from killing ourselves. Thankfully the AIs we have today are neither aligned enough or capable enough to get their wish yet.
> AI is following the same script. Build first, understand later. Ship it, then figure out if it’s safe.
Even if the above wasn't cause enough for concern, our biggest concern should be that no one seems to be concerned.
We're all doomed unfortunately. The world is about to become a very bleak place very quickly.
Robert Miles youtube videos on AI safety go over these issues well, and are from before the LLM days.
Humans are just barely aligned ourselves. The moment any group or nation of them gets power they tend to use it in some horrific manner against other humans. What do we think will happen the moment AI gets a leg up on humans.
[flagged]
She's probably happier than you though.