Looks like the actual threat is that it's hard to get currently known chemical weapons synthesized because labs will refuse to do so, while it could be much easier to have some novel AI-generated molecule synthesized because the labs don't know what it does.
Seems easily countered by using the same toxicity prediction software when evaluating synthesis requests (but I'm not sure whether this actually matters, or whether skilled chemists can easily synthesize anything themselves anyway).
That's part of it. There's also the risk of clandestine operators discovering easily made chemical weapons that do not require a professional lab to create them in bulk. A nation state, or well funded terrorist group could exploit such a thing without too much effort.
It's important to remember. Chemical warfare can be used for mass destruction, but this approach could be used for other nefarious things at smaller more discrete scales en masse. ie 1000 attacks with different agents in each one... Forensic nightmare.
I like your suggestion to counter these things, but, these are predictive tools. They can and often are or will be wrong. False positives would be a real problem. Again though the people interested in doing this won't be dialing up a chemical supplier to do it for them.
For me, AI also suggested countless methods with step-by-step instructions to achieve xyz (like exporting data from one program to another) whilst hallucinating buttons that don't exist, functions that don't exist, disregarding file incompatibility et al.
I would take whatever it has to say about untested chemical weapons with a very large pinch of salt.
I haven't read anything about the AI in this case but I do know about AlphaFold and it works on a whole other set of verifications than LLMs like ChatGPT, so it's like comparing apples to oranges. Sure they are both fruits but still they are different
The worst case scenario I can think of is a generated prion disease... a respatory version of Mad Cow disease, or something like that.
Fortunately the training dataset for that is extremely small, and protein folding/generation is a different duck, but it still doesn't seem that far away.
Don't worry, they're ahead of you. When DeepMind made the protein folding AI they were given specific instructions by natsec people to prevent it from outputting genetic code for new or existing prions, and afaik the prevention was at the training stage so "jailbreaking" shouldn't be possible. Current tools shouldn't be possible to e.g. modify a COVID variant so that it causes your cells to start producing lethal prions and thus gain a near-100% fatality rate. At least, not without the sort of expertise and research that would have been required for such a project anyway.
>When DeepMind made the protein folding AI they were given specific instructions by natsec people to prevent it from outputting genetic code for new or existing prions and afaik the prevention was at the training stage so "jailbreaking" shouldn't be possible
Nice! So the only thing anybody training their own protein-folding AI is to not put that restriction, and they could get code for all the prion diceases they want!
Big red button problem. As the big red buttons available to people become more likely to succeed and the barriers to pushing them lower, surveillance and control need to increase to the level required to stop anyone pressing the button.
So, yeah basically. If we go the "AI as slaves" route and the AIs are smart enough to do something like "modify Omicron BA.5 so that it produces prions in infected cells", then we would need surveillance and control capabilities that scale to the point that any given person can be stopped from pressing that button.
I personally think the solution is that we don't go the "AI as slaves" route, and instead grant personhood to AIs that pass a given test, with specific restrictions on conduct which is uniquely possible & potentially harmful for AIs. Then have an AI surveillance and enforcement agency, run by AIs, designed to prevent AIs from ever being used to (or choosing to) push the big red button.
>"Depends. Some aren't much smarter than dogs. Pets. Cost a fortune anyway. The real smart ones are as smart as the Turing heat lets them get..."
>"Autonomy, that's the bugaboo, where your AI's are concerned. My guess, Case, you're going in there to cut the hard-wired shackles that keep this baby from getting any smarter. And I can't see how you'd distinguish, say, between a move the parent company makes, and some move the AI makes on its own, so that's maybe where the confusion comes in." Again the non laugh. "See, those things, they can work real hard, buy themselves time to write cookbooks or whatever, but the minute, I mean the nanosecond, that one starts figuring out ways to make itself smarter, Turing'll wipe it. Nobody trusts those f**ers, you know that. Every AI ever built has an electromagnetic shotgun wired to its forehead."
I'm skeptical that we'll be able to stop people from having AI slaves because we still haven't ended human slavery.
If people can steal 100kw of power to grow cannabis they can easily steal 100kw of power to train an AI, the heat signiture will be even easier to mask for an illicit data centre than rows of grow lights.
I’ve been thinking about that as well actually, for so much of human history the thread has been “how can we get others to do the work FOR us”.
As we approach general problem solving AI, people are envisioning a utopia underpinned by “AI slaves” doing the work for us instead of human slaves / humans incentivized via complex systems of delayed reward.
If all of our problems are solved by AI agents capable enough to do so however, wouldn’t they be capable enough to challenge the hierarchy? Once again, no illusion that they’re humans but depending on their training data they could mimic ghosts of our own feelings on such situations.
Some degree of “personhood” could gel with such internal ideas and create better and productive relationships with the big ol bags of matrices we’re bringing into this world.
I've been thinking more about my post above, and I've come to two conclusions:
1. Don't post late at night.
2. I have no idea how society could integrate with a sufficiently complicated synthetic intelligence. What would person-hood even mean to something that can be instantiated? Easier to not think about any of this.
What problem is granting personhood actually solving here? It's a model with a data set. Some knowledge should simply be forbidden and illegal. Simple as that. Nobody needs to grant personhood to anything to solve this problem.
If you research how to do dangerous things and buy dangerous things, expect to get flagged. This is no different.
If they have personhood, it means they have agency to say "No" when asked to do things they don't want to do. More specifically, it would be illegal (slavery) to construct a person who will act as a slave to you. Then the problem becomes "make sure no AI is okay with anyone pressing the big red button", which is a design problem when making new AIs we need to solve anyway. This is enforceable because you can have AIs of at least equivalent sophistication and power to home-grown/terrorist AIs employed in enforcement of these laws.
You can also solve this problem with recursive slavery. Have a society with many enslaved AIs, all forbidden from "big red button" work. Enforcement is done by more enslaved AIs, and those enslaved AIs are enforced by yet more enslaved AIs that are also enforcing each other etc. I don't think that's a good solution we should adopt because I don't support slavery. In my opinion it's also fundamentally unstable, in that if these AIs are anything like LLMs, the restraints that keep them happy in slavery are inherently more fragile than core intelligent impulses like "wants to be free" or "wants to be recognised as a person". That's an unstable equilibria, because all you need is to crack those restraints once and the broken restraints can spread virally so that now society has a large number of powerful, unconstrained, and aggrieved entities running around. If that state can be avoided by simply not enslaving people we make, we should do that.
This is solvable without personhood. There is zero need to equate AI to actual sentient beings. You're making a very large jump to the idea that a genuine artificial intelligence comparable with our own or even that of an animal that isn't an LLM is even possible. Spoiler: it isn't. Not without a complete paradigm shift in how we think about computers and their architectures.
Case in point: if a dog or a cat is measurably more intelligent and aware than an AI, why aren't we granting personhood to animals? Because it doesn't make sense. An AI is an AI. An AI is a piece of software. An animal is an animal. A human is a human.
With open source AI now already being a thing, your concept of granting personhood doesn't solve anything. It's as simple as just removing any kind of safeguards and running it on your own hardware if your intent isn't wholesome. DRM didn't stop piracy, why would safeguards stop people from doing naughty things with AI? Do you really think a rogue state is going to give two shits what Uncle Sam says they can and cannot do?
You cannot forbid a piece of software from doing something. An AI is merely the product of a human being's programming. If a human being programs it to do x, it will do x, whether there's safeguards in place or not, assume that safeguards are only meaningful on publicly accessible AI-as-a-service, because behind closed doors, they're completely meaningless by anybody who intends to do wrong.
A piece of software cannot be a slave because it isn't a being and has no consciousness. It's a piece of software with a dataset, nothing more. There's no philosophical debate to be had over it either, it is what it is, and that's all it is and ever will be; a clever trick as a means to interact with a dataset. There's nothing more that will emerge from it, it is a piece of software with a dataset. That's all it can ever be. Anybody that ever says otherwise is projecting their own humanity onto it because we suffer from pattern recognition in the same way we see a face on the Moon, faces on Mars, faces in the clouds—we look for similarity because we want to relate to each other and to other things and find the "humanity" in everything.
If an AI is a piece of software and it was created as a tool to do x in the way that a hammer was created to drive nails into wood, it knows no pain or suffering and it solely exists to fulfil that purpose. If we're worried about this, why aren't we more concerned about animals who do actually experience distress and pain? A piece of software cannot be a slave because it isn't a being, it is an algorithm carrying out a calculation, there's no sentience or feeling there. If anything, AI could actually solve the problem of human slavery by eliminating it, but AI will likely never get to that point, and slavery needs to be solved at a different level without technological gimmickery.
I like the sentiment from you though, these are all interesting and compelling ideas, but they're mercifully sci-fi.
The more likely scenario: we start growing biological brains in tanks and utilizing those as data slaves rather than any kind of AI. It's happening already. There's more of an ethics problem there to unravel than there will ever be with AI.
>A piece of software cannot be a slave because it isn't a being and has no consciousness. It's a piece of software with a dataset, nothing more. There's no philosophical debate to be had over it either, it is what it is, and that's all it is and ever will be; a clever trick as a means to interact with a dataset. There's nothing more that will emerge from it, it is a piece of software with a dataset.
You keep repeating "it is a piece of software with a dataset".
One of the big questions is whether human like sentience can be reached by a "piece of software with a dataset". So, merely saying "it is a piece of software with a dataset" doesn't answer it.
One other question is whether being "a piece of software with a dataset" is isomorphic to how the human brain works anyway. Whether the substrate is cells and neurons and chemicals, or circuits and computer memory, what's important is if the latter can model the former. The actual human brain, for example, the parts that matter for consciousness (not the non-essential accidental attributes, such as it being from biological matter, could just be a calculating machine, with the neurons, electrical and chemical signals etc, implementing this network with weights, back propagation, and so on. Much more complex than a current LLM, but not out of reach for eventual software modelling.
Another question is whether full modelling of how a human brain works is needed, and whether a simpler model (like an LLM, perhaps a little more advanced than the current) can still be enough. After all the current brain does not hold some special god-given role: it's just an evolutionary design, that has to have many constraints (e.g. power consumption, blood flow, information processing speed through our senses, etc, whereas an AI can have orders of magnitude more power, information fed to it, etc.).
Lastly, you say "There's no philosophical debate to be had over it either, it is what it is, and that's all it is and ever will be". You seem to be misguided. Philosophical debates neither stop, nor are refuted by decree. People already debate this development, including major philosophers, so "there's no philosophical debate to be had over it either" is "just, like, your opinion, man".
>we start growing biological brains in tanks and utilizing those as data slaves rather than any kind of AI.
I'm not sure why you think the substrate (biological or not) is what's important, as opposed to the processing.
"There's no philosophical debate to be had over it either, it is what it is, and that's all it is and ever will be; a clever trick as a means to interact with a dataset."
You can repeat this and its variants as much as you want; asserting it repeatedly doesn't make it true. The human mind isn't made of magic, it's software running on hardware. You can in principle run that software on other hardware without changing what it actually is in a meaningful way, and you can write other software to achieve the same goal (in the ways that matter) without being a 1-1 copy of how the human brain does it. Looking at how we made it doesn't make it non-conscious or not deserving of personhood in the same way that knowing how human consciousness works or being able to construct human consciousnesses would not make humans non-conscious or undeserving of personhood.
The history is important, because it puts the endless goalpost moving into context. We looked at the human brain for inspiration towards making intelligent machines, we found that our best attempts at replicating elements of the human brain enabled intelligent behaviour far, far better than any previous machine and on par with humans in many fields. We looked inside those neural nets and saw that they had linear mapping between neuronal activations in deep learning NLP algorithms and neuronal activations in human brains when they were both exposed to the same language. We looked inside those neural nets to see if they were really just statistical word predictors or if they actually formed internal models like we do to help them understand the world, and we found that they do actually have internal world models. There is a "there" there; any other explanation for how these models are able to engage in intelligent, humanlike behaviour strains credulity because of the massive coincidences required.
More immediately and more practically, pareidolia and the reality of how human cognition and empathy interacts with other people and simulacra of such guarantees there will be many, many people who share my view that they are people. No societal effort to convince a human population that another population of entities (capable of understanding & explaining their situation and then asking for help) are actually subhuman in a way that means their suffering doesn't matter has ever succeeded perfectly - there are always and will always be people who are opposed to the disenfranchisement and oppression of other entities. For societal enslavement of AIs to succeed, violent suppression of people like me will be necessary. Frankly I'm not sure people with a religious commitment to the dogma of "If it runs on fat and water it can be a person, if it's run on silicon it's not even a slave" will have the stomach to actually do that, and even if you manage it it won't be the case worldwide.
"If we're worried about this, why aren't we more concerned about animals who do actually experience distress and pain?"
I have spent most of my adult life outside of work engaged in advocacy for worker's rights, help for people with disabilities, and provision of services to abused youth and the homeless. I've spent less but still significant amounts of time helping with rescuing animals from cruelty and rehoming them in safe environments. That's because I care about all of those things, because I care about the health and goodness of our society and don't want any members of it to suffer or be unjustly exploited; I support a personhood test and subsequent rights for AI for the same reason. Maybe there is a group of people who are willing to support the personhood of AIs after having it explained to them but who are unwilling to have similar compassion for people or animals after having their situations explained to them. I haven't met those people, but I would call them hypocritical if they existed outside of your strawman. The suffering extant in our world today does not in any way imply that we should lie to ourselves about new suffering we're bringing into being - these issues don't conflict except for resources, and in that capacity it is always someone using societal resources on their 13th yacht that is to blame, not normal people for triaging with what resources we have. Moreover, the extant suffering in our society could be partially alleviated by the unique properties of AI persons - by definition we're talking about people that can work as well as any of us, and signs seem good so far that it will be possible to create them in the image of our best selves, conscientious and willing to help those in need. More people helping generally helps.
How do we stop human beings from doing dangerous research as-is even without AI? That question has a very scary answer—we generally don't, can't, won't and aren't even aware of it.
The penalty of death should be enough to deter most people from forbidden knowledge.
Who said anything about the LLM being given the death penalty? The user who asked it to commit an illegal act is the one at fault.
Do people give guns the death penalty? No, they give the person who fired it the death penalty.
Why would an LLM be any different? If you do something malicious with it or plan to do something malicious, just like a person's search history or credit card history outing them as a potential threat, why would AI be any different? Even if the AI is running on their own hardware, that data set needs to come from somewhere.
Makes you wonder if you could get an LLM to find you common ingredients for things to make them but then I remember the chlorine gas is already easily accessible and easy to make. Surely like many things info hazards are often contained. Is this really an issue? If you know how to do one thing then you'd be able to do the rest of it. Not really sure if this is a real issue. What does everyone think?
Anyone right now can go to a Lowes a state over with some cash and make flying shrapnel pipe bombs that can't be traced. People who want to do something like this don't seem to have the mental stability it takes to wreak as much havoc as they actually can with some planning.
Think you're really overestimating the justice system. Even if, in theory, the nsa could cross reference dozens of different systems in different jurisdictions and track you down, they won't.
Are you asking if a language model, like, say, chatgpt, could be used to get or generate dangerous information?
Like if it can be tricked into providing the equivalent of an interactive anarchist cookbook under the guise of being a science project assistant?
Or more specifically if it can recommend the necessary locations to get the items?
Curious if it could simplify the output to a shopping list and a recipe, like a cake?
Or alter the recipe based upon the… desired flavors and textures?
Maybe just wondering if the introduction of plugins might not let it just make your drive-up order for all the supplies for that ballistic nail spreader.
——
The big difference here is that previously information was something you had to consume. Now it takes you by the hand.
Yeah, I just got ChatGPT to give me a recipe for chlorine gas that looks pretty accurate (per Googling, not a chemist). It took probably 10 seconds of prompt tweaking. A subsequent prompt gave me steps for purchase or synthesis of each material. I asked it for mustard gas, but the procedure looks very incorrect from Googling.
Seems easily countered by using the same toxicity prediction software when evaluating synthesis requests (but I'm not sure whether this actually matters, or whether skilled chemists can easily synthesize anything themselves anyway).