Discussion
Is logistic regression "AI"? What about convex optimization? Are LLMs AI? What about an image recognition system built without using "feature learning" techniques but done through "feature engineering"?
Is the resulting model AI or the technique that is used to build the model or the task that said model is solving? What about statistics? Is that "AI"? Which parts of it are "AI" vs not? What about information theory?
I read some mainstream news articles this year that described systems with no machine learning as AI. At this point, anything involving one or more computers is described as AI.
Those in signal processing may know the level to which the much more hypes field of machine learning uses so many concepts from information theory. But machine learning wasn't "AI" before.
Even under ML, are kernel techniques AI? Or is it just neural networks and the term "deep learning" which I learned was created to avoid the stigma around "neural networks" that existed in ML before?
@timnitGebru from what one of my professors said, "machine learning" is what got coined to avoid stigma around "AI" during the 1990s "winter", because "AI research" would make grants fail but machine learning wouldn't. AI was a pretty broad field, up until the very end of last century, so I understand anyway!
but I feel like the field probably made a mistake by using such a sci-fi evocative term, instead of something like applied statistics or emergent systems...
Techniques using statistical ML weren't bucketed under "AI" when those techniques weren't the things that showed high performance on whatever tasks people thought were "AI" tasks in the 80s.
Are decision trees "AI" now? How about 40 years ago? What's the difference?
@timnitGebru An observation: People in the Othello community (and I'm sure that of many other abstract strategy games) regularly refer to brute-force endgame evaluation as “AI”, even though it's just a naïve exhaustive search of the game tree. I think it's basically because it has the same interface as heuristic evaluation.
@timnitGebru I'm afraid it's too late to tidy up - at this point we simply have to scrap everything with an #AI label.
If someone wants to fish something useful out of the trash & reapply it using a label which is more accurate & less weighted down by bullshit, they should do so, but it can't be on all of us to rummage through mountains of shit looking for sort-of-edible kernels of corn.
@jwcph @timnitGebru Certainly if someone has themself called their perfectly useful ML application or something "AI", sure - reap what they sowed.
I find this very odd. In the last bubble the people doing useful things with VR goggles didn't call it "metaverse" and in the previous one the people doing useful things with Merkle trees didn't call it "blockchain".
I'm beginning to agree, especially since I am getting exhausted dealing with the constant, BUT YOUR PHONE IS AI, BUT YOUR SPELL CHECK IS AI, BUT YOUR AUTOCORRECT SCRIPT IS AI. It makes conversation about LLMs or really any variant of the sort impossible so, yes, I'd be willing to do this until people can learn how to name things propperly. @jwcph @timnitGebru
@timnitGebru This is such a good thread. Most of the time, I already have no idea what these concepts are individually, even when they're not being smushed into a meaningless buzzword paste!
@timnitGebru I liked Paul Cantrell's summary that "AI" research refers any task that computers were recently bad at. The techniques don't matter, only whether it's intended to do something people thought computers couldn't do.
https://hachyderm.io/@inthehands/114599036617229554
@timnitGebru In my experience (>40 years in #AI), once AI techniques were understood well they were thrown in the toolbox and no longer had the allure of the bleeding edge where “real AI” was to be found. Like peeling an onion, the layers came off and we eagerly begin to peel the next layer. Surely, AI must be inside! Nope, it’s just software. Decision trees were never AI, but automatically deducing them from examples (supervised learning) was AI (e.g., ID3).
@timnitGebru AI is defined as anything that can be sold to the public as magic and/or sentience.
@timnitGebru ai is anything you want to get funded atm
@timnitGebru 40 years ago decision trees were expert systems. And were often trumpeted as one of the successful applications of "AI" along with playing chess.
Of course the chess program was basically a decision tree!
@timnitGebru it seems to me that anything that involves optimizing an objective function is AI now, whether analytically or via gradient descent. So training a neural network via back propagation is AI , but so also is ridge regression implemented via SVD. 😅
@masp @timnitGebru
It's more a case that any project you want funding for is AI now. If their is even a hint of a way you can pass it off as such, then that's what you do, because that's where all the hype is.
BT Free is a non-profit organization founded by @ozoned@btfree.social . It's goal is for digital privacy rights, advocacy and consulting. This goal will be attained by hosting open platforms to allow others to seamlessly join the Fediverse on moderated instances or by helping others join the Fediverse.
