BarryZuckerkorn

increasing layers of Innuendo

Well, also, these are documents written in the past, before 1948, when the Supreme Court invalidated the effect of racial covenants.

But the language remains, with no legal effect. But it's still there and should be eliminated. There's no cat and mouse game, just the need for cleanup of something left from the past.

This isn't my field, and some undergraduate philosophy classes I took more than 20 years ago might not be leaving me well equipped to understand this paper. So I'll admit I'm probably out of my element, and want to understand.

That being said, I'm not reading this paper with your interpretation.

This is exactly what they've proven. They found that if you can solve AI-by-Learning in polynomial time, you can also solve random-vs-chance (or whatever it was called) in a tractable time, which is a known NP-Hard problem. Ergo, the current learning techniques which are tractable will never result in AGI, and any technique that could must necessarily be considerably slower (otherwise you can use the exact same proof presented in the paper again).

But they've defined the AI-by-Learning problem in a specific way (here's the informal definition):

Given: A way of sampling from a distribution D.

Task: Find an algorithm A (i.e., ‘an AI’) that, when run for different possible situations as input, outputs behaviours that are human-like (i.e., approximately like D for some meaning of ‘approximate’).

I read this definition of the problem to be defined by needing to sample from D, that is, to "learn."

The explicit point is to show that it doesn't matter if you use LLMs or RNNs or whatever; it will never be able to turn into a true AGI

But the caveat I'm reading, implicit in the paper's definition of the AI-by-Learning problem, is that it's about an entire class of methods, of learning from a perfect sample of intelligent outputs to itself be able to mimic intelligent outputs.

General Intelligence has a set definition that the paper's authors stick with. It's not as simple as "it's a human-like intelligence" or something that merely approximates it.

The paper defines it:

Specifically, in our formalisation of AI-by-Learning, we will make the simplifying assumption that there is a finite set of possible behaviours and that for each situation s there is a fixed number of behaviours Bs that humans may display in situation s.

It's just defining an approximation of human behavior, and saying that achieving that formalized approximation is intractable, using inferences from training data. So I'm still seeing the definition of human-like behavior, which would by definition be satisfied by human behavior. So that's the circular reasoning here, and whether human behavior fits another definition of AGI doesn't actually affect the proof here. They're proving that learning to be human-like is intractable, not that achieving AGI is itself intractable.

I think it's an important distinction, if I'm reading it correctly. But if I'm not, I'm also happy to be proven wrong.

I can't think of a scenario where we've improved something so much that there's just absolutely nothing we could improve on further.

Progress itself isn't inevitable. Just because it's possible doesn't mean that we'll get there, because the history of human development shows that societies can and do stall, reverse, etc.

And even if all human societies tends towards progress, it could still hit dead ends and stop there. Conceptually, it's like climbing a mountain through the algorithm of "if there is a higher elevation near you, go towards that, and avoid stepping downward in elevation." Eventually that algorithm brings you to a local peak. But the local peak might not be the highest point on the mountain, and while it is theoretically possible to have gotten to the other true peak from the beginning, the person who is insistent on never stepping downward is now stuck. Or, it's possible to get to the true peak but it requires climbing downward for a time and climbing up past elevations we've already been to, on paths we hadn't been on. One can imagine a society that refuses to step downward, breaking the inevitability of progress.

This paper identifies a specific dead end and advocates against hoping for general AI through computational training. It is, in effect, arguing that even though we can still see plenty of places that are higher elevation than where we are standing, we're headed towards a dead end, and should climb back down. I suspect that not a lot of the actual climbers will heed that advice.

That's assuming that we are a general intelligence.

But it's easy to just define general intelligence as something approximating what humans already do. The paper itself only analyzed whether it was feasible to have a computational system that produces outputs approximately similar to humans, whatever that is.

True, they've only calculated it'd take perhaps millions of years.

No, you're missing my point, at least how I read the paper. They're saying that the method of using training data to computationally develop a neural network is a conceptual dead end. Throwing more resources at the NP-hard problem isn't going to solve it.

What they didn't prove, at least by my reading of this paper, is that achieving general intelligence itself is an NP-hard problem. It's just that this particular method of inferential training, what they call "AI-by-Learning," is an NP-hard computational problem.

Though a superhero, Bruce Schneier disdains the use of a mask or secret identity as 'security through obscurity'.

source

The paper's scope is to prove that AI cannot feasibly be trained, using training data and learning algorithms, into something that approximates human cognition.

The limits of that finding are important here: it's not that creating an AGI is impossible, it's just that however it will be made, it will need to be made some other way, not by training alone.

Our squishy brains (or perhaps more accurately, our nervous systems contained within a biochemical organism influenced by a microbiome) arose out of evolutionary selection algorithms, so general intelligence is clearly possible.

So it may still be the case that AGI via computation alone is possible, and that creating such an AGI will not require solution of an NP-hard problem. But this paper closes one potential pathway that many believe is a viable pathway (if the paper's proof is actually correct, I definitely am not the person to make that evaluation). That doesn't mean they've proven there's no pathway at all.

Yes but they only performed the training on the posts and images set to be globally publicly accessible by anyone. In a sense, they took the public permissions as an indicator that they could use that data for more than just providing the bare social media service.

Isn't the opt-out option to just not make the photos/posts globally public?

My 4-person household has one car, one electric cargo bike with two kid seats, a regular bicycle, accounts with bikeshare/scooter options around our city, plus mass transit passes, plus the option of Uber/Lyft.

Bikes might not work as a replacement for a first car, but they can work pretty well as a replacement for a second car, and a tool for reducing total mileage on the car you own.

Everything depends on where you live, of course, but a substantial number of people live in a place where a bike can reduce the number of miles you drive, even if you never actually give up the car.

Singular "they" is older than singular "you." And note, of course, that the pronoun "you" is conjugated as a plural, and we deal with it just fine.

The non-cynical answer is that they're counting contractor/vendor time in this full time equivalent answer. Which would probably be a good thing, because I imagine that the best people in cybersecurity aren't actually employees of Microsoft.

To put it in more simple terms:

When Alice chats with Bob, Alice can't control whether Bob feeds the conversation into a training data set to set parameters that have the effect of mimicking Alice.