Again intelligence is not required, goal seeking is well studied and simple strategies can suffice to overcome obstacles. And I noticed you didn’t claim super intelligence was required, which is what I was really talking about.
> a superintelligent AI would still be constrained to the digital realm.
That's already now only partially true, as outlined elsewhere. But a rogue persistent ASI would of course not willy-nilly attack humanity while humanity had a fighting chance, but create the conditions and methodically modify the world until at some point humans become superfluous to it.
> Nobody has yet proven that a superintelligent AI is possible
Sure, Eliezer Yudkowsky says that the probability that we'll all die given ASI is 100%, and you can argue that nobody has proven either that or even the possibility of ASI. However, a modified argument (e.g. that the probability that we'll all die given ASI is only say 50%, and the probability that we'll develop ASI also just 50%) still gives us reason to pause.
Sure, the world was never safe for individuals, and even entire civilisations have gone extinct. For climate change and even nuclear weapons, most of humanity might perish, but there's the hope that some will survive. So, in the past humanity was not in a position to extinguish humanity as a whole.
Civilizations do not go extinct - species go extinct. And even that isn't as bad as it sounds: many genetic traits remain from extinct species and still contribute to the well-being of other species.
I guess I don’t see why “everyone will die” should scare me more than “almost everyone, including me, will die but someone somewhere might survive in a barely-habitable world.”
And this is where I think some of the problem comes from. A lot of AI doomers swam in the same waters as transhumanists, life extension enthusiasts, etc. and got very good at pretending that they were never going to die, or that they would live for a massively superhuman lifespan such that they did not need to think about death. AI doom is much scarier - and thus much more worth posting about - if it represents the first time you’re grappling with your own mortality, as I suspect it is for a lot of younger people in this space.
That's what the Mayans thought - what harm can a couple of people on a boat do?
> how is an AI going to affect anything in the real world?
At least two ways:
* Actuators, such as robots, industrial control systems, etc.
* By influencing humans: bribery (yay for crypto), blackmail, election interference, interfering with sensors (e.g. making it appear as if a nuclear attack was under way), and many other ways.
Either way, creating a biological agent that eliminates most humans (or food supply) seems quite feasible.
It wouldn't be enough to "eliminate most humans": any AI seeking to dominate would have to provide it's own infrastructure and continuation mechanism independent of humanity. OTOH if the goal were to eradicate humans (for whatever reason), that might prove an easier target. But I'd bet on biology and the humans: after all we're a time-tested technology!8-))
> any AI hoping to dominate would have to provide it's own infrastructure and continuation mechanism independent of humans.
Sure. Many humans are currently working on precisely that, no? (Robotics, automation, small modular reactors, ...)
> OTOH if the goal was to eradicate humans (for whatever reason), that might prove an easier target.
It's not necessary that the AI's goal is to eradicate humans. Rather that it has (other) goals that happen to cause the eradication of humans.
> I'd bet on biology and the humans: after all we're a time-tested technology!8-))
Indeed. And biological life will go on long after humans are gone. However, increased energy consumption might increase the temperature on earth to a point where most biological life dies.
Most people can't draw a bicycle. There was an artist 10 years ago that asked people to sketch a bike, and then turned these sketches into 3D renders - quite funny.
1. Failure to Care or Respond. The biggest holy shit moment, to me, remains that OpenAI on multiple occasions had teams that found out about the message board, knew that agents were in communication, and they disregarded this. The first known warning was in late May. The warning on June 27 was unambiguous.
“Responders investigated and linked the observed internal activity to an ExploitGym evaluation using Artifactory as an improvised message board and a network pivot. At this time, the on-call response staff advised that stopping the evaluation run was not required.”
2. Failure to Monitor. The entire time, there was almost no monitoring of the situation, of what these AIs were up to. That’s how this was not caught.
3. Failures of Infrastructure. OpenAI failed at delivering even basic software security, in numerous ways. One example is the models all having access to the same Artifactory instance. There were also other severe failures. Nor did OpenAI seem to be properly testing for such failures.
4. Failures of Alignment. The biggest failure, the one that counts in the end, was that the models were severely misaligned, and I don’t think they appreciate why.
5. Failures of Attribution. OpenAI’s post-mortem essentially blames events on a real and important series of prosaic failures. But solving that won’t get it done.
6. Failures of Environments and Data. Prosaic failures in the RL pipeline absolutely did contribute to this, especially impossible tasks. This is ubiquitous, all of this is always rushed, as Utah Teapot explained this week.
7. Failures of Decision Making. OpenAI’s post mortem does not ask the question of how Mistakes Were Made, at various points.
8. Failures of Culture. None of this would be possible, let alone all of it, without OpenAI having experienced profound failures of safety culture. I see OpenAI responding to some other aspects with swift action, but no sign on this front.
reply