Whatever the real intentions of Sam Altman are, I think what he is doing might end up being a net positive.
Imagine if fossil fuel companies had helped, early-on, to hammer into the public perception of oil how it might end up being disastrous for humanity instead of burying the studies they made.
I don't think we would have stopped using fossil fuels, but maybe we would have been in a better spot today.
I think a better analogy would be if fossil fuel companies lobbied the government to make it illegal to have solar panels on your roof. The thought of AI being fully controlled by corporations is the threat. Not some fake apocalypse those seeking regulatory capture have ginned up.
But if I took a guess: more internalization of climate change by the general population, earlier investments into alternative sources of energies, maybe baking in carbon pricing early on before we all became completely dependent on oil.
This is the fox guarding the henhouse. Establishing a business model for OpenAI is his whole intent, nothing less and nothing more.
He shapes the public image of AI in a way that the only "safe" way of having AI is that it's controlled by megacorporations like MS, Google or Facebook, and that anyone running a model privately at home is a potential criminal.
Also, in the unlikely case that OpenAI has given birth to a malevolent AI that is trying to take over the world right now, then that AI might come to the conclusion that one of the biggest threats to its existence aren't humans themselves, but other AI models created by humans. So "AI must be regulated" can just as well be the product of a rogue AI that doesn't want some other rogue AI to take over control.
Pine trees spread needles to suppress growth in the around them and keep nearby competitors from consuming their resources. Human governance structures and corporations do the same. No reason to think that a sufficiently complex AI would not do the same.
We should apply the same principles to fight each of these tendencies and reject special treatment for big players in all it's forms.
Your comment made me imagine a scenario where Altman is secretly asking GPT-5 what to do next. As a result he goes out on this world tour to try to turn governments against alternative AIs.
> When a myth dies hard, you can bet that it is either because people want to believe it (for example, when it makes their lives easier) or because there is a kernel of truth in it just big enough to make it convincing. The long-lived gardening myth that pine needles lower soil pH falls into the latter category. It is a twofold myth, erroneously claiming that pine needles should be used as a mulch only around acid-loving plants, and that they will acidify compost if added to a compost bin....
> Perhaps lending further credibility to the myth is that gardeners often observe that it is difficult to grow most plants under large pine trees. They attach an assumption to this observation: It must be that the pine needles shed by the tree cause the soil there to become acidic. However, this is a case of drawing an incorrect conclusion from the facts. The reasons most plants perform poorly when grown under large pine trees are that the trees cast too much shade, or the root system of the trees blankets the soil surface, forming a barrier and sucking up most of the available water and nutrients.
> Black walnuts (Juglans nigra) are one of the best known allelopathic plants. Black walnut trees naturally contain a chemical called juglone which can inhibit the growth of some plants. Walnut buds, nut hulls, and roots contain the largest amount of juglone. However, even walnut leaves and stems contain a small concentration of juglone.
I never said anything about soil pH. Pine needles are pretty acidic when they first drop but that goes away pretty quick. What i'm talking about is the biological perma-mulch that is the pine needle. Pine needles break down slowly and form an effective mulch layer that suppresses further growth. All you have to do to know that is true is to look under a pine tree and a deciduous tree of similar side and density and note the difference.
"Imagine if fossil fuel companies had helped, early-on, to hammer into the public perception of oil how it might end up being disastrous for humanity instead of burying the studies they made."
Saying your product is a disaster for humanity is...a bold marketing strategy...
In other news: "AI Doomerism Is a Decoy - Big Tech’s warnings about an AI apocalypse are distracting us from years of actual harms their products have caused.". Notable quote:
> This seems like really sophisticated PR from a company that is going full speed ahead with building the very technology that their team is flagging as risks to humanity
I don't think we should let real, present-day threats distract us from the actual possibility of AI apocalypse either.
They're two separate problems, they're both serious problems, and they can both be addressed (well, actually, we don't know if AI alignment even can be solved).
Not that I don't appreciate you bringing this up. As someone apparently wasting a lot of my time trying to explain the concept of misaligned AGI, I still think it's very important that we 1) don't trust Sam Altman, and 2) spend a lot of resources dealing with human alignment problems, such as abuse of existing AI and other technology.
Eh, every generation had its share of people predicting the apocalypse (Mayas Calendar, Nuclear, AI, Robot Uprising, Environemental collapse, Zombie Apocalypse) to the point one wonder if it's not a trait of human race.
Which risk? I mean I saw "Terminator 1" and "Terminator 2" (great movies) and I played "Horizon Zero Dawn" (great game). But what proof do we have of those risks exist apart from wild exercises of imagination?
You cannot prove that the risks exist, and you cannot prove that the risks do not exist, so this approach to the problem is clearly not useful to make any progress on the question "is AI dangerous or not".
Of course you can, there's a thing called research. This narrative is being pushed WITHOUT any significant research behind. So this is only about economics and politics, not science.
Well I'm not being paid to do that, but my professional instinct is, like you do with any other software, you set up a sandbox. Pretty sure an AI specialist could have much better ideas than me. What I don't get is how this narrative is being so wildly pushed without any research sustaining it, even just to say "hey we tried to investigate this huge issue, here you can see the great amount of papers we wrote, but we concluded is actually impossibly to establish the risks because etc etc". This is first class BS.
I am not sure how you came to the conclusion that there is no research on this topic. On Google Scholar you can literally find millions of research papers discussing the risks of AI, its ethics, alignment with human values, and so on, for example:
- Scherer, Matthew U. "Regulating artificial intelligence systems: Risks, challenges, competencies, and strategies." Harv. JL & Tech. 29 (2015): 353. (934 scholar citations)
- Araujo, Theo, et al. "In AI we trust? Perceptions about automated decision-making by artificial intelligence." AI & society 35 (2020): 611-623. (334 scholar citations)
- Osoba, Osonde A., and William Welser IV. "An intelligence in our image: The risks of bias and errors in artificial intelligence." Rand Corporation, 2017. (278 scholar citations)
- Gabriel, Iason. "Artificial intelligence, values, and alignment." Minds and machines 30.3 (2020): 411-437. (216 scholar citations)
- Bryson, Joanna, and Alan Winfield. "Standardizing ethical design for artificial intelligence and autonomous systems." Computer 50.5 (2017): 116-119. (214 scholar citations)
- Yu, Han, et al. "Building ethics into artificial intelligence." arXiv preprint arXiv:1812.02953 (2018). (200 scholar citations)
- Munoko, Ivy, Helen L. Brown-Liburd, and Miklos Vasarhelyi. "The ethical implications of using artificial intelligence in auditing." Journal of Business Ethics 167 (2020): 209-234. (180 scholar citations)
- Christian, Brian. "The alignment problem: Machine learning and human values." WW Norton & Company, 2020. (178 scholar citations)
- Manheim, Karl, and Lyric Kaplan. "Artificial intelligence: Risks to privacy and democracy." Yale JL & Tech. 21 (2019): 106. (178 scholar citations)
- Russell, Stuart, et al. "Ethics of artificial intelligence." Nature 521.7553 (2015): 415-416. (164 scholar citations)
- Risse, Mathias. "Human rights and artificial intelligence: An urgently needed agenda." Hum. Rts. Q. 41 (2019): 1. (135 scholar citations)
- Turchin, Alexey, and David Denkenberger. "Classification of global catastrophic risks connected with artificial intelligence." Ai & Society 35.1 (2020): 147-163. (108 scholar citations)
- Taylor, Jessica, et al. "Alignment for advanced machine learning systems." Ethics of Artificial Intelligence (2016): 342-382. (95 scholar citations)
- Santoni de Sio, Filippo, and Giulio Mecacci. "Four responsibility gaps with artificial intelligence: Why they matter and how to address them." Philosophy & Technology 34 (2021): 1057-1084. (84 scholar citations)
Oh c'mon... you know what I'm talking about, I'm not talking about theory and more theory, I'm talking about actual research done on top of an LLM, giving it the apparent freedom to go rogue and see what tries to do then. Specifically on top of OpenAI's GPT, as they the ones carrying this message along the world. This is the type of research I'm talking about. I saw Altman's presentation to the US congress and not a word, about any study, was said.
Well, there are plenty of real-world examples of harms caused by AI-powered systems (https://incidentdatabase.ai). For LLMs in particular, I think it's just too early to have convincing independent research on their harms, but coincidentally this just happened a few days ago:
I'm sure there is a name for this logical trick you are trying to use.
OK I will claim there is a grand spaghetti monster. You can't disprove the existence of the spaghetti monster. Therefore, the spaghetti monster must exist.
The burden of proof is on the AI fanfic crowd to prove that terminator and AGI is just around the corner.
But that is not the type of reasoning I was doing. Following your example, you also cannot prove that spaghetti monsters do exist. Since neither side can offer definitive proof of their stance, the only way to settle the debate is to have less stringent requirements than proofs, such as, for example, circumstantial evidence.
Nobody has ever seen spaghetti monsters, therefore it seems unlikely that they exist. Not a definitive proof, but it does allow a certain degree of certainty. The point is to reach probabilistic conclusions in light of the available evidence, rather than seeking definitive proofs, which are impossible for most real-world questions.
> The burden of proof is on the AI fanfic crowd to prove that terminator and AGI is just around the corner.
No, this is an argument from ignorance (https://en.wikipedia.org/wiki/Argument_from_ignorance).
Each side of the debate is supposed to provide proof, evidence, or reasons for their perspective, otherwise they do not have much clout, besides prejudice.
Since there are plenty of examples of risks and harmful consequences of AI applications (see my other replies in this thread), it is quite reasonable to assume that new AI technologies pose new risks and that we should at least consider what those risks may be. Now the burden is yours to argue that no, despite all these past failures, we're still totally safe.
I hate this narrative. Every piece of it has problems. We are no where close to AGI. On top of that, the risks associated with AI have not been demonstrated beyond some speculative pieces sci-fi fan fiction.
The current generation of work is not AI. The risks it poses are minimal. You cannot hurt me with an LLM. It may reshape the internet due to the fact that content generation just got a whole lot easier. Beyond that you would have a difficult time harming me in any way with an LLM.
This article really boils down to Sam Altman traveling the world and having all the institutions of power happily agree that they want to repress any threats to their power.
I hope open source wins before the UN Treaty on Compute gets drafted.
The current LLM’s can say things, but they have nothing to say; they all just say whatever sounds good in context. We have yet to have a generally useful AI which can reason, and use the reasoning to tell us things.
Actually we have reasoning AI - in inductive and deductive reasoning spaces. What we lack is abductive reasoning AI, which LLMs are actually really really close to. They all lack agency, but you can approximate agency through goal based agents. By tying together a goal based agent with inductive, deductive and LLM in a feedback cycle, I think you’ll find we are pretty darn close. The reasoning behavior will be different than human, and maybe the reasoning classes need to be extended, but we have already crafted either optimal or decent approximations of human reasoning. The task as I see it is less making LLM AGI but in taking all the AI work to date and synthesizing them into an ensemble that delegates tasks to the right model based on the domain of a sub tasks in a larger task space and providing the appropriate feedback, state, integrations, as well as providing integrations to optimizers, solvers, and other computational and mechanical algorithms.
It reminds me of nurses reporting all the patients dying of COVID-19 insisting to their last breath that covid isn't real. There is nothing that will change their mind because it has become opaque to objective reality. Burying their head even if suffocates them.
Isn't it the opposite? People who are willing to say that a simple ELIZA bot is a form of AI think that something being AI isn't very important; it's a low bar to clear and doesn't imply much about the usefulness of the thing. To these people, AI is not intrinsically important. A chess bot is AI, it has very limited utility and just isn't important at all in any domain other than chess.
On the other hand, people who say that AI is an as-of-yet unobtained far-future technology are saying that AI is intrinsically very sophisticated and useful; so much so that nobody has succeeded in creating one yet. These people think that AI is intrinsically important. They think the term AI is so important that it must not be applied to systems with limited mundane usefulness.
> They think the term AI is so important that it must not be applied to systems with limited mundane usefulness
Yes? Thinking is hard. People would happily delegate as much cognitive load to a computer as possible - especially if they believe the computer is intelligent. With this comes 'computer says no' and 'we are not responsible; it was the AI that was behaving unjust'. Until we have an artificial intelligence that can take responsibility for its actions - and remedy them if needs be, we should hammer home that these systems have limited mundane usefulness.
People delegate their decision making to other systems all the time, whether or not those systems are termed to be AI.
There is no empirical test for whether or not a system is AGI, and 'AI' is a term which has been applied to all manner of mundane systems for decades already. It is not important that we restrict the term AI to only mean AGI (whatever it is that AGI even means!) The 'CPU' players in Mario Kart 64 were frequently called AI, and there was no problem with that. Anybody who gets bent out of shape over the casual use of the term AI is placing way too much importance on the term.
Time for hacker culture to triple down on open source contribution. We have access to some great building blocks so far, here’s to hoping we can make the most out of it.
> You cannot hurt me with an LLM. It may reshape the internet due to the fact that content generation just got a whole lot easier. Beyond that you would have a difficult time harming me in any way with an LLM.
I'm always surprised seeing these narrow views on HN.
History is the process of technology and culture shifting over time, influencing each other. There's not going to be some magic moment where we all wake up and turn on our television and see the morning news tell us that "AGI exists today, and here's what's changing." Not least because very few people ever turn on a TV in the morning anymore.
Individually, things will change. Maybe a given change will be small or large, fast or quick. And these changes will add up. Of course it's very scary that a massive change could happen overnight, but that's not the expectation and it's not the only criterion for AI danger (or climate change, or political instability, or economic instability, or...)!
The calls for regulation are rallied around what is called AI existential risk. I am taking umbrage with these calls for regulation by stating we are no where close to artificial intelligence (current generation is content generation, no logic, no truth) and that it poses no risk.
There is a further argument that even with actual AI, which people have written volumes of sci-fi tier essays about how it will kill us all, that there has been no demonstrated risk, only speculation.
We know that progress towards ever more sophisticated AI will be incremental, with occasional breakthroughs and leaps forward. But even assuming we never leap forward again, the ideal time to sort out what to do when some threshold is crossed is before we have crossed that threshold.
By way of analogy, the climate crisis is arguably suffering from the same kinds of thinking. We’ve never witnessed the kinds of climate catastrophe that we know is around the corner, but that doesn’t stop people who understand the dynamics from taking it seriously. It also hasn’t stopped a very vocal and powerful opposition who won’t be convinced until we experience real catastrophe.
To me, the argument that we’re nowhere close is unconvincing, and I’d argue irrelevant. Even if the real risk is decades away, we’re now living in the moment where we can mitigate risks before they become pressingly real. Anything else is kicking the can down the road.
> There is a further argument that even with actual AI, which people have written volumes of sci-fi tier essays about how it will kill us all, that there has been no demonstrated risk
How would you define “demonstrated risk”? We’re at a stage where all we can do is speculate and predict based on our models of the world as they currently exist. That says nothing about the actual intrinsic risks that we are or are not speculating about.
And again, the climate crisis seems like a useful parallel to consider.
>Even if the real risk is decades away... All we can do is ... predict based on our models...
What real risk? What models?
Climate change is not a useful parallel because changing weather has an actual impact on human safety. No one can articulate the risks that AGI would pose beyond speculative fiction. It's going to clone itself to every device and turn us into paperclips by emailing an on-demand gene printing lab to make a super virus!! Sure, whatever man.
Just because you can say the words "X is similar to Y" does not make it true. This applies to comparisons with nuclear weapons, guns, climate change, et al. I understand how all of those things can kill me. No one has made a reasonable argument for how an LLM could harm me, much less an argument convincing enough to justify the regulatory capture sought by these companies.
You aren't answering my question. I asked about the relationship between 'is it dangerous?' and 'is it AI?' Some relationship between these questions was suggested by you in the comment I responded to above:
> "The current generation of work is not AI. [THEREFORE?] The risks it poses are minimal. You cannot hurt me with an LLM."
and again suggested by this remark:
> "we are no where close to artificial intelligence (current generation is content generation, no logic, no truth) and [THEREFORE?] it poses no risk."
You seems to be claiming that LLMs aren't dangerous because they aren't AI. Does the set of dangerous things belong to the set of AI? Pipe bombs and pit vipers aren't AI, so does that mean they aren't dangerous? They are, therefore the set of all dangerous things does not belong to the set of AI. Why then is "The current generation of work is not AI" thought to support the assertion that LLMs cannot be dangerous?
I think it could be reasonable for you to assert that LLMs have not yet been demonstrated to be dangerous enough to warrant regulation. But that's not what was asserted above; you asserted that LLMs cannot hurt people, and the supporting argument seems to be 'because they aren't real AI'. That is simply illogical. Whether or not LLMs are sci-fi tier AI is irrelevant; it has absolutely no bearing on whether or not LLMs are dangerous.
You're arguing a complete straw man. Please read the sentence before the one you are hyper focused on.
I am taking a position against everything about the current narrative.
1) Calling the current generation of tech AI.
2) The idea that AI is dangerous.
3) The conclusion that regulation is required.
You're hallucinating some causality in the statements that isn't written. It has not been demonstrated that AI (AGI) will be dangerous. Further, LMMs are not AI. LLMs have also not been demonstrated to be dangerous. We do not need to give OpenAI (Microsoft) regulatory capture.
So your objection to LLMs being called AI is a non sequitur and has nothing to do with your objection to LLMs being regulated? That's not the impression I got from either of your comments.
This isn’t about any threat other than competitive threat to Microsoft and Google. This is naked regulatory arbitrage to form a moat as fast as possible in a space that is highly dynamic and likely to tilt away from the first movers. They want to create FUD as quickly and widely as possible and use the large megacorp regulatory experience as a differentiator in the pursuit of AI. I’ve worked in these company’s and related megacorps and a massive amount of their value is in regulatory compliance, and they can beat down any startup or open source effort on this alone in many cases. So, click here to accept for your safety.
I'd go beyond that. Let's say this is the day any advanced computer can be turned into an AGI device. So what. There are already 8 billion, not AGI but "NGI" minds in the world. Thanks to the Internet, and outgrowing the fear of death, inevitable in the other hand, any of us could decide to use our available knowledge to create terrifying damage to our fellow humans. Specially people close to high-tech military systems, which has to be many, many thousands around the world, not to mention nuclear weapons installations. Instead of that, with extremely rare exceptions, all of us chose and have a deep desire of making our fellow humans happy, grow beautiful families, and live a meaningful life. And why is that? We should really think a little bit more about that before being so terrified about AI.
Are you aware of any sane takes like this you have links to? Even some of the skeptical articles like the one in The Atlantic are more calling out hypocrisy than attempting to take a serious look at the reality of threats from "AI" (which, as you point out is, not really AI).
Needing a licence to create or possess models is what they’re planning here right? It obviously makes sense for openai and other big companies because it makes it more difficult for smaller competitors but it is really scary. It is probably way more dangerous for society than some theoretical artificial super intelligence because of opportunity cost and that it would certainly lead to more and more authoritarianism.
Sam Altman has a talent of utilizing any narrative in a way that benefits his company. It is sad to see legislators jumping onto AI space out of fear rather than to facilitate innovation
I think I missed the whole evil Microsoft bit in the 90s (a little too young), and while a lot of big tech has done a lot of harm since it hasn't felt so obviously, unashamedly evil.
Altman seems like the successor to Micro$oft to me. Every time I hear his name it's associated with some effort to regulate "ai" so his company can profit at the expense of humanity. It makes me so angry.
I get that regulatory capture is a bad thing if done for the reason of creating a moat, but where is the evidence that creating a moat is the primary motivation here?
AI researchers aren’t close to a consensus on the potential for bad alignment outcomes, so isn’t it possible Altman is warning of something real?
I think if this needs to be captured and controlled, it's unfair and unreasonable for it to be a high margin business (since the margins would stem in substantial part from a government granted monopoly).
I'm certain low margin stewardship isn't what Altman has in mind.
Anyone who thinks AI is a threat is either a moron or an agent (agent of influence, power or state). As described in many initiated conversations, AI to date in it's most known form is a glorified autocomplete. Any ML-implementations are rudimentary but valuable.
One of the hardest proponents of "there's a 50% chance of AI will wipe out humanity" are right out not well. Tegmark, the one who uttered this is a real hoax. His "Future of Life Institute" also offered pro-nazi paper "Nya Dagbladet" 100 000 USD. Once uncovered by expo in Sweden (https://expo.se/2023/01/elon-musk-funded-nonprofit-run-mit-p...) the offer was retracted.
The venue, "Nya dagbladet" is connected to Russia and seems to be a broad spectrum investment into the far-right movement.
With open source LLMs approaching Chat-GPT it makes sense for OpenAI to ask governments to erect barriers and controls on the use of AI. If AI can be deeply regulated then only deep pockets will be able to play.
Similar talk was heard about cryptography being weaponized and needing export controls, regulations, and protections.
62 comments
[ 3.1 ms ] story [ 133 ms ] threadImagine if fossil fuel companies had helped, early-on, to hammer into the public perception of oil how it might end up being disastrous for humanity instead of burying the studies they made.
I don't think we would have stopped using fossil fuels, but maybe we would have been in a better spot today.
But if I took a guess: more internalization of climate change by the general population, earlier investments into alternative sources of energies, maybe baking in carbon pricing early on before we all became completely dependent on oil.
He shapes the public image of AI in a way that the only "safe" way of having AI is that it's controlled by megacorporations like MS, Google or Facebook, and that anyone running a model privately at home is a potential criminal.
Also, in the unlikely case that OpenAI has given birth to a malevolent AI that is trying to take over the world right now, then that AI might come to the conclusion that one of the biggest threats to its existence aren't humans themselves, but other AI models created by humans. So "AI must be regulated" can just as well be the product of a rogue AI that doesn't want some other rogue AI to take over control.
We should apply the same principles to fight each of these tendencies and reject special treatment for big players in all it's forms.
https://www.thespruce.com/do-pine-needles-acidify-soil-14031...:
> When a myth dies hard, you can bet that it is either because people want to believe it (for example, when it makes their lives easier) or because there is a kernel of truth in it just big enough to make it convincing. The long-lived gardening myth that pine needles lower soil pH falls into the latter category. It is a twofold myth, erroneously claiming that pine needles should be used as a mulch only around acid-loving plants, and that they will acidify compost if added to a compost bin....
> Perhaps lending further credibility to the myth is that gardeners often observe that it is difficult to grow most plants under large pine trees. They attach an assumption to this observation: It must be that the pine needles shed by the tree cause the soil there to become acidic. However, this is a case of drawing an incorrect conclusion from the facts. The reasons most plants perform poorly when grown under large pine trees are that the trees cast too much shade, or the root system of the trees blankets the soil surface, forming a barrier and sucking up most of the available water and nutrients.
A better example would be Black Walnut:
https://extension.illinois.edu/blogs/ilriverhort/2018-08-06-...:
> Black walnuts (Juglans nigra) are one of the best known allelopathic plants. Black walnut trees naturally contain a chemical called juglone which can inhibit the growth of some plants. Walnut buds, nut hulls, and roots contain the largest amount of juglone. However, even walnut leaves and stems contain a small concentration of juglone.
Saying your product is a disaster for humanity is...a bold marketing strategy...
> This seems like really sophisticated PR from a company that is going full speed ahead with building the very technology that their team is flagging as risks to humanity
https://news.ycombinator.com/item?id=36175060
They're two separate problems, they're both serious problems, and they can both be addressed (well, actually, we don't know if AI alignment even can be solved).
Not that I don't appreciate you bringing this up. As someone apparently wasting a lot of my time trying to explain the concept of misaligned AGI, I still think it's very important that we 1) don't trust Sam Altman, and 2) spend a lot of resources dealing with human alignment problems, such as abuse of existing AI and other technology.
It's just not all very serious
- Scherer, Matthew U. "Regulating artificial intelligence systems: Risks, challenges, competencies, and strategies." Harv. JL & Tech. 29 (2015): 353. (934 scholar citations)
- Araujo, Theo, et al. "In AI we trust? Perceptions about automated decision-making by artificial intelligence." AI & society 35 (2020): 611-623. (334 scholar citations)
- Osoba, Osonde A., and William Welser IV. "An intelligence in our image: The risks of bias and errors in artificial intelligence." Rand Corporation, 2017. (278 scholar citations)
- Gabriel, Iason. "Artificial intelligence, values, and alignment." Minds and machines 30.3 (2020): 411-437. (216 scholar citations)
- Bryson, Joanna, and Alan Winfield. "Standardizing ethical design for artificial intelligence and autonomous systems." Computer 50.5 (2017): 116-119. (214 scholar citations)
- Yu, Han, et al. "Building ethics into artificial intelligence." arXiv preprint arXiv:1812.02953 (2018). (200 scholar citations)
- Munoko, Ivy, Helen L. Brown-Liburd, and Miklos Vasarhelyi. "The ethical implications of using artificial intelligence in auditing." Journal of Business Ethics 167 (2020): 209-234. (180 scholar citations)
- Christian, Brian. "The alignment problem: Machine learning and human values." WW Norton & Company, 2020. (178 scholar citations)
- Manheim, Karl, and Lyric Kaplan. "Artificial intelligence: Risks to privacy and democracy." Yale JL & Tech. 21 (2019): 106. (178 scholar citations)
- Russell, Stuart, et al. "Ethics of artificial intelligence." Nature 521.7553 (2015): 415-416. (164 scholar citations)
- Risse, Mathias. "Human rights and artificial intelligence: An urgently needed agenda." Hum. Rts. Q. 41 (2019): 1. (135 scholar citations)
- Turchin, Alexey, and David Denkenberger. "Classification of global catastrophic risks connected with artificial intelligence." Ai & Society 35.1 (2020): 147-163. (108 scholar citations)
- Taylor, Jessica, et al. "Alignment for advanced machine learning systems." Ethics of Artificial Intelligence (2016): 342-382. (95 scholar citations)
- Santoni de Sio, Filippo, and Giulio Mecacci. "Four responsibility gaps with artificial intelligence: Why they matter and how to address them." Philosophy & Technology 34 (2021): 1057-1084. (84 scholar citations)
National Eating Disorder Association shuts down A.I. chatbot it planned to use to replaces humans saying it 'may have given' harmful information (https://fortune.com/well/2023/05/31/neda-ai-chatbot-harmful-...)
I'm sure there is a name for this logical trick you are trying to use.
OK I will claim there is a grand spaghetti monster. You can't disprove the existence of the spaghetti monster. Therefore, the spaghetti monster must exist.
The burden of proof is on the AI fanfic crowd to prove that terminator and AGI is just around the corner.
Your example is an instance of the fallacy of the inverse (https://en.wikipedia.org/wiki/Denying_the_antecedent).
But that is not the type of reasoning I was doing. Following your example, you also cannot prove that spaghetti monsters do exist. Since neither side can offer definitive proof of their stance, the only way to settle the debate is to have less stringent requirements than proofs, such as, for example, circumstantial evidence.
Nobody has ever seen spaghetti monsters, therefore it seems unlikely that they exist. Not a definitive proof, but it does allow a certain degree of certainty. The point is to reach probabilistic conclusions in light of the available evidence, rather than seeking definitive proofs, which are impossible for most real-world questions.
> The burden of proof is on the AI fanfic crowd to prove that terminator and AGI is just around the corner.
No, this is an argument from ignorance (https://en.wikipedia.org/wiki/Argument_from_ignorance). Each side of the debate is supposed to provide proof, evidence, or reasons for their perspective, otherwise they do not have much clout, besides prejudice.
Since there are plenty of examples of risks and harmful consequences of AI applications (see my other replies in this thread), it is quite reasonable to assume that new AI technologies pose new risks and that we should at least consider what those risks may be. Now the burden is yours to argue that no, despite all these past failures, we're still totally safe.
None of the existing AI technologies we examine have AGI. There are hundreds of examples you can use.
The current generation of work is not AI. The risks it poses are minimal. You cannot hurt me with an LLM. It may reshape the internet due to the fact that content generation just got a whole lot easier. Beyond that you would have a difficult time harming me in any way with an LLM.
This article really boils down to Sam Altman traveling the world and having all the institutions of power happily agree that they want to repress any threats to their power.
I hope open source wins before the UN Treaty on Compute gets drafted.
On the other hand, people who say that AI is an as-of-yet unobtained far-future technology are saying that AI is intrinsically very sophisticated and useful; so much so that nobody has succeeded in creating one yet. These people think that AI is intrinsically important. They think the term AI is so important that it must not be applied to systems with limited mundane usefulness.
Yes? Thinking is hard. People would happily delegate as much cognitive load to a computer as possible - especially if they believe the computer is intelligent. With this comes 'computer says no' and 'we are not responsible; it was the AI that was behaving unjust'. Until we have an artificial intelligence that can take responsibility for its actions - and remedy them if needs be, we should hammer home that these systems have limited mundane usefulness.
There is no empirical test for whether or not a system is AGI, and 'AI' is a term which has been applied to all manner of mundane systems for decades already. It is not important that we restrict the term AI to only mean AGI (whatever it is that AGI even means!) The 'CPU' players in Mario Kart 64 were frequently called AI, and there was no problem with that. Anybody who gets bent out of shape over the casual use of the term AI is placing way too much importance on the term.
The idea that AI is not risk in and of itself, but is a risk because of humans seems to miss the point that this delineation is artificial.
AI and the problems humans will cause with it are inextricably linked. Development of one leads to the other.
I'm always surprised seeing these narrow views on HN.
History is the process of technology and culture shifting over time, influencing each other. There's not going to be some magic moment where we all wake up and turn on our television and see the morning news tell us that "AGI exists today, and here's what's changing." Not least because very few people ever turn on a TV in the morning anymore.
Individually, things will change. Maybe a given change will be small or large, fast or quick. And these changes will add up. Of course it's very scary that a massive change could happen overnight, but that's not the expectation and it's not the only criterion for AI danger (or climate change, or political instability, or economic instability, or...)!
Can you please clarify this point? Are you asserting "not AI, therefore it cannot hurt me", or "It can't hurt me, therefore not AI"?
What precisely is the relationship between something being AI and something being capable of harm?
There is a further argument that even with actual AI, which people have written volumes of sci-fi tier essays about how it will kill us all, that there has been no demonstrated risk, only speculation.
By way of analogy, the climate crisis is arguably suffering from the same kinds of thinking. We’ve never witnessed the kinds of climate catastrophe that we know is around the corner, but that doesn’t stop people who understand the dynamics from taking it seriously. It also hasn’t stopped a very vocal and powerful opposition who won’t be convinced until we experience real catastrophe.
To me, the argument that we’re nowhere close is unconvincing, and I’d argue irrelevant. Even if the real risk is decades away, we’re now living in the moment where we can mitigate risks before they become pressingly real. Anything else is kicking the can down the road.
> There is a further argument that even with actual AI, which people have written volumes of sci-fi tier essays about how it will kill us all, that there has been no demonstrated risk
How would you define “demonstrated risk”? We’re at a stage where all we can do is speculate and predict based on our models of the world as they currently exist. That says nothing about the actual intrinsic risks that we are or are not speculating about.
And again, the climate crisis seems like a useful parallel to consider.
What real risk? What models?
Climate change is not a useful parallel because changing weather has an actual impact on human safety. No one can articulate the risks that AGI would pose beyond speculative fiction. It's going to clone itself to every device and turn us into paperclips by emailing an on-demand gene printing lab to make a super virus!! Sure, whatever man.
Just because you can say the words "X is similar to Y" does not make it true. This applies to comparisons with nuclear weapons, guns, climate change, et al. I understand how all of those things can kill me. No one has made a reasonable argument for how an LLM could harm me, much less an argument convincing enough to justify the regulatory capture sought by these companies.
> "The current generation of work is not AI. [THEREFORE?] The risks it poses are minimal. You cannot hurt me with an LLM."
and again suggested by this remark:
> "we are no where close to artificial intelligence (current generation is content generation, no logic, no truth) and [THEREFORE?] it poses no risk."
You seems to be claiming that LLMs aren't dangerous because they aren't AI. Does the set of dangerous things belong to the set of AI? Pipe bombs and pit vipers aren't AI, so does that mean they aren't dangerous? They are, therefore the set of all dangerous things does not belong to the set of AI. Why then is "The current generation of work is not AI" thought to support the assertion that LLMs cannot be dangerous?
I think it could be reasonable for you to assert that LLMs have not yet been demonstrated to be dangerous enough to warrant regulation. But that's not what was asserted above; you asserted that LLMs cannot hurt people, and the supporting argument seems to be 'because they aren't real AI'. That is simply illogical. Whether or not LLMs are sci-fi tier AI is irrelevant; it has absolutely no bearing on whether or not LLMs are dangerous.
tl;dr: Rattlesnakes aren't AI, therefore rattlesnakes can't hurt me.
I am taking a position against everything about the current narrative.
You're hallucinating some causality in the statements that isn't written. It has not been demonstrated that AI (AGI) will be dangerous. Further, LMMs are not AI. LLMs have also not been demonstrated to be dangerous. We do not need to give OpenAI (Microsoft) regulatory capture.Altman seems like the successor to Micro$oft to me. Every time I hear his name it's associated with some effort to regulate "ai" so his company can profit at the expense of humanity. It makes me so angry.
AI researchers aren’t close to a consensus on the potential for bad alignment outcomes, so isn’t it possible Altman is warning of something real?
I think if this needs to be captured and controlled, it's unfair and unreasonable for it to be a high margin business (since the margins would stem in substantial part from a government granted monopoly).
I'm certain low margin stewardship isn't what Altman has in mind.
One of the hardest proponents of "there's a 50% chance of AI will wipe out humanity" are right out not well. Tegmark, the one who uttered this is a real hoax. His "Future of Life Institute" also offered pro-nazi paper "Nya Dagbladet" 100 000 USD. Once uncovered by expo in Sweden (https://expo.se/2023/01/elon-musk-funded-nonprofit-run-mit-p...) the offer was retracted.
The venue, "Nya dagbladet" is connected to Russia and seems to be a broad spectrum investment into the far-right movement.
Similar talk was heard about cryptography being weaponized and needing export controls, regulations, and protections.