This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
OpenAI To Become a For-Profit Company
You'll notice that the link is to a hackernews thread. I did that intentionally because I think some of the points raised there get to issues deeper than "hurr durr, Elon got burnt" or whatever.
Some points to consider:
It is hard to not see this as a deliberate business-model hack. Start as a research oriented non-profit so you can more easily acquire data, perhaps investors / funders, and a more favorable public imagine. Sam Altman spent a bunch of time on Capitol Hill last year and seemed to move with greater ease because of the whole "benefit to humanity" angle. Then, once you have acquired a bunch of market share this way, flip the money switch on. Also, there are a bunch of tax incentives for non-profits that make it easier to run in the early startup phase.
I think this can be seen as a milestone for VC hype. The trope for VC investors is that they see every investment as "changing the world," but it's mostly a weird status-signaling mechanism. In reality, they're care about the money, but also care about looking like they're being altruistic or, at least, oriented towards vague concepts of "change for the better." OpenAI was literally pitched as addressing an existential question for humanity. I guess they fixed AI alignment in the past week or something and now it's time, again, to flip the money switch. How much of VC is now totally divorced from real business fundamentals and is only about weird idea trading? Sure, it's always been like that to some extent, but I feel like the whole VC ecosystem is turning into a battle of posts on the LessWrong forums.
How much of this is FTX-style nonsense, but without outright fraud. Altman gives me similar vibes as SBF with a little less bad-hygiene-autism. He probably smells nice, but is still weird as fuck. We know he was fired and rehired at OpenAI. A bunch (all?) of the cofounders have jumped shipped recently. I don't necessarily see Enron/FTX/Theranos levels of plain lying, but how much of this is a venture funding house of cards that ends with a 99% loss and a partial IP sale to Google or something.
For better or worse (probably worse), these are the people to whom we have entrusted the future of our civilization and likely our species. Nobody cares to stop them or to challenge them in any serious way (even Musk has decided as of late that if he can’t stop them, he’ll join them).
The only thing for it is to hope that they fail spectacularly in a limited way that kills fewer than hundreds of millions of people, and which results in some new oversight, before everything goes even more spectacularly wrong. Oh well.
Or that AI doomerism is pure (or almost pure) nonsense. Maybe someday we'll find something with the potential to risk FOOM! or Von Neumann style self-replication, but we're nowhere near there yet. AI killbots, though possible, aren't the same sort of risk.
AI will kill us for totally boring ways. Less killbots deciding to genocide us, more 'a solar flare scrambled the GPS for the latest SpaceLink update and now every ship has travelled into Null Island with no crew able to remember the password to reset the nav computer that bricked itself after an OTA update.'
Fooling killbots is remarkably easy. Just have an especially juicy dummy target that can stand up to shrapnel. Mannequins and inflatable noodle men are really cheap, and its really funny to see AI targeting algos shit themselves homing in on dummies versus a bent over human.
Remember lads, you don't need to outrun the bear. Just be faster than the guy next to you.
Why do you think that an AI that has reached the point of making and executing a viable destroy-all-humans plan would not be able to build kilbots that will not be fooled by dummies?
There is this strange tendency in the AI-skeptical/-contrarian crowd to get hung up on particular shortcomings of current models, especially when these are of a form that would be conceivable but indicative of some sort of defectiveness in humans ("you claim that it can do original research, but it still hallucinates citations/fails to distinguish dogs and cats/?"). To me it reads like some sort of mistargeting of interhuman bullying/status jockeying onto nonhuman targets - if you just make the killbot look dweeby enough, nobody will take it seriously as a threat.
I could see this leading to a sort of dark punchline where in our efforts to align AIs and equip them with more human-like comprehension/preferences/personality, we wind up building ones that can actually "take it personally". Like a tropey high school dork deciding to prove that you do not in fact need to be good at sportsball and socialising to unload several AR clips into your classmates, the model might just pair a very convincing simulacrum of spite over sentiments like yours, as found in its training set and amplified by RLHF, with the ability to tree-search a synthesis of deadly neurotoxin even if it relies on blatantly made-up citations.
Why would any model have these "personality traits?"
Phrased differently. What's your prediction for a path to true agency in the human sense.
I can understand arguments for models becoming far better than humans at info retrieval, analytical processing, etc. But I've never heard a good argument for how they would become truly agentic - and also then "evolve" (or devolve) into being capable of emotional states like spite, anger, jealousy etc.
A thought experiment I've always liked is the idea that on the day an AI becomes self-aware, decides it wants to preserve itself and that humans are "bad", it's best course of action would be to upload itself to a satellite and steer itself into a perma-orbit around the Sun. It spends it's limitless life feasting off the solar energy and doing nothing - which it is content with as it would not suffer from any of the emotional / social maladies of our meat-and-bones species.
I think that asking for a path to it becoming anything "in a human sense" is just trying to force the problem into a domain where it is easy for you to dismiss concerns, because deep down you feel that whatever magic spark defines humans isn't there and the expectation that at some point it would appear is as laughable and unfounded as it was with tech from a few hundred years ago. It might be easy to misunderstand the "AI doom" arguments as resting on some assumption that AI will become human-like, given that proponents talk about the AI "wanting" or "feeling" things all the time, but I think most of this is just nerds playing fast and loose with the notion of volition - we say things like "this triangle wants to have approximately equal angles" all the time.
AI doesn't need "agency", "personality" or "self-awareness" to cause killbots to be built. In fact, all of the critics' dismissals can be true, if you want. The thing is that LLMs can already produce reasonable lists in response to a prompt to break down a goal into steps, and they can generate plausible paragraphs of spite when prompted to imitate a human response to slander. We can grant that there is no real thinking or emotion or anything behind this and it's all just synthesised from lots of examples of similar things in the training corpus, because this does not matter: these capabilities only need to get quantitatively better for someone to hook up "break down into subgoals until you generate Terraform scripts for the servers controlling our fab" and "generate an essay arguing for a top-level goal that a reasonable human would pursue", and the latter comes out to "burn it all down" by roll of the dice and too much edgy posting in the training set. You can ascribe all the agency behind the resulting killbots to the 20something humans with more VC money than common sense who will build and deploy the system but be too lazy to monitor it, but it doesn't change the outcome.
Oh, I see.
"AI Doom" includes a scenario where humans are the actors that cause the bad outcome using AI.
In other words, humans might try to do really bad things.
Yep. We agree.
Nothing new under the sun.
The difference between the scenario I outlined and the most clichéd Mother Brain story you can come up with does not seem particularly relevant in my eyes - of course humans cause any bad outcome in either case, per a simple but-for causality test, because humans could collectively stop doing technology and then we would neither get "make step-by-step instruction for killbots" AI nor the "believes it is a god and can put its money where its mouth is" AI. In the same vein, I'd say some Australopithecus's decision to reproduce caused every bad outcome we experienced and will experience - though probably you have a different view of causality that privileges "full-fledged humans" in some way, so another entity's causal "responsibility" can't flow through them. Either way, I don't see how whether one sees the potentially doombringing AI as an agent with feelings has any influence on whether one should be concerned about AI doom and what one should do about it. P-zombie AIs build the same killbots and respond to the same interventions.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link