First, let me say that what broke me from the herd at lesswrong was specifically the calls for AI pauses. That somehow 'rationalists' are so certain advanced AI will kill everyone in the future (pDoom = 100%!) that they need to commit any violent act needed to stop AI from being developed.
The flaw here is that there's 8 billion people alive right now, and we don't actually know what the future is. There are ways better AI could help the people living now, possibly saving their lives, and essentially eliezer yudkowsky is saying "fuck em". This could only be worth it if you actually somehow knew trillions of people were going to exist, had a low future discount rate, and so on. This seems deeply flawed, and seems to be one of the points here.
But I do think advanced AI is possible. And while it may not be a mainstream take yet, it seems like the problems current AI can't solve, like robotics, continuous learning, module reuse - the things needed to reach a general level of capabilities and for AI to do many but not all human jobs - are near future. I can link deepmind papers with all of these, published in 2022 or 2023.
And if AI can be general and control robots, and since making robots is a task human technicians and other workers can do, this does mean a form of Singularity is possible. Maybe not the breathless utopia by Ray Kurzweil but a fuckton of robots.
So I was wondering what the people here generally think. There are "boomer" forums I know of where they also generally deny AI is possible anytime soon, claim GPT-n is a stochastic parrot, and make fun of tech bros as being hypesters who collect 300k to edit javascript and drive Teslas*.
I also have noticed that the whole rationalist schtick of "what is your probability" seems like asking for "joint probabilities", aka smoke a joint and give a probability.
Here's my questions:
-
Before 2030, do you consider it more likely than not that current AI techniques will scale to human level in at least 25% of the domains that humans can do, to average human level.
-
Do you consider it likely, before 2040, those domains will include robotics
-
If AI systems can control robotics, do you believe a form of Singularity will happen. This means hard exponential growth of the number of robots, scaling past all industry on earth today by at least 1 order of magnitude, and off planet mining soon to follow. It does not necessarily mean anything else.
-
Do you think that mass transition where most human jobs we have now will become replaced by AI systems before 2040 will happen
-
Is AI system design an issue. I hate to say "alignment", because I think that's hopeless wankery by non software engineers, but given these will be robotic controlling advanced decision-making systems, will it require lots of methodical engineering by skilled engineers, with serious negative consequences when the work is sloppy?
*"epistemic status": I uh do work for a tech company, my job title is machine learning engineer, my girlfriend is much younger than me and sometimes fucks other dudes, and we have 2 Teslas..
Needling in on point 1 - no I don't, largely because AI techniques haven't surpassed humans in any given job ever :P. Yes, I am being somewhat provocative, but no AI has ever been able to 1:1 take over a job that any human has done. An AI can do a manual repetitive task like reading addresses on mail, but it cannot do all of the 'side' work that bottlenecks the response time of the system: it can't handle picking up a telephone and talking to people when things go wrong, it can't say "oh hey the kids are getting more into physical letters, we better order another machine", it can't read a sticker that somebody's attached somewhere else on the letter giving different instructions, it definitely can't go into a mail center that's been hit by a tornado and plan what the hell it's going to do next.
The real world is complex. It cannot be flattened out into a series of APIs. You can probably imagine building weird little gizmos to handle all of those funny side problems I laid out, but I guarantee you that all of them will then have their own little problems that you'd have to solve for. A truly general AI is necessary, and we are no closer to one of those than we were 20 years ago.
The problem with the idea of the singularity, and the current hype around AI in general, is a sort of proxy Dunning-Kruger. We can look at any given AI advance and be impressed but it distracts us from how complex the real world is and how flexible you need to be a general agent that actually exists and can interact and be interacted upon outside the context of a defined API. I have seen no signs that we are anywhere near anything like this yet.
The counter argument is GPT-4. For the domains this machine has been trained on it has a large amount of generality - a large amount of capturing that real world complexity and dirtiness. Reinforcement learning can make it better.
Or in essence, if you collect colossal amounts of information, yes pirated from humans, and then choose what to do next by 'what would a human do', this does seem to solve the generality problem. You then fix your mistakes with RL updates when the machine fails on a real world task.
No it's not. GPT-4 is nowhere near suitable for general interaction. It just isn't.
"Just do machine learning to figure out what a human would do and you'll be able to do what a human does!!1!". "Just fix it when it goes wrong using reinforcement learning!!11!".
GPT-4 has no structured concept of understanding. It cannot learn-on-the-fly like a human can. It is a stochastic parrot that badly mimics a the way that people on the internet talk, and it took an absurd amount of resources to get it to do even that. RL is not some magic process that makes a thing do the right thing if it does the wrong thing enough and it will not make GPT-4 a general agent.