If you are, like me, generally opposed to how Large Language Models are being positioned and used (and not regulated currently) one very simple thing you could do is to stop using the term “AI” to describe them. When you do, you’re buying into the linguistic and conceptual framework of your enemies.
Companies name their LLMs Claude or whatever because they want you to think of them as things with consciousness. They do not have consciousness. They are sophisticated, energy-gobbling, programs. They’re tools.
These tools can mimic (to some extent, with many faults) the outputs of human consciousness: language, images, etc. But they do not have the actual consciousness that created those things. They do this by stealing the work of conscious beings and pastiching it.
I’m not saying these tools don’t have legitimate uses. Just like using Timesmachine or newspapers dot com is easier than a microfiche. But we should be careful to describe them accurately.
a world-renowned scientist I've worked with says "it's neither artificial nor intelligent."
It's just supercharged autocomplete
I've been taking that approach, but I think there's a reasonable case, as others have suggested, for using the phrase "generative AI" (insofar as "AI" has long been used to describe any computer algorithm).
Don’t be reasonable! Fight to win!
You could explain that AI in this case stands for Averaged Inanity - taking the entirety of text posted on the internet without doing any quality review, and using it to predict what word comes next does indeed give a kind of Inanity Of The Crowds.
in fairness, LLMs come out of the field of AI computing research, which goes back many, many decades, to the '50s, so it's a bit hard to separate the terms
Sure, but they are not actually artificially intelligent. They do not have intelligence. The goal of AI computing research is to eventually create actual artificial intelligence, which it has not yet.
It’d be like if I was in sandwich research and then put cheese on a slice of ham without the bread and called it a sandwich. You’d be like “no, this might be a step towards a sandwich but it isn’t a sandwich”
I think it's more like you're trying to make a sandwich and you create a really accurate 3D printed model of a sandwich. It would fool lots of people who see it, but could never actually become a sandwich. LLMs are an impressive simulation of verbal intelligence, but not really a step toward it.
"artificial intelligence" in the same sense as "artificial houseplant," or maybe as a more genuinely useful example, "artificial limb" the real concern becomes: are we cutting off our limbs to replace them with robotic limbs that we can't even really control? are we cutting out our brains?
that's a lofty goal that some have, but not really what the field is all about. it's just interdisciplinary theories of computing logic based on the ways human beings think rather than the traditional binary logic. grew out of the cognitive revolution. Chomsky came out of the same stew.
interestingly, critiques about the theoretical untrustworthiness of AI to determine facts, for example, were there almost from the beginning. the best argument was made in the '80s, that because symbols have no meaning to a computer, the logical mode can't really be compared to humans.
i tend to think the problem is more a function of sci-fi stories. science fiction directly connected AI with like robots that have emotions, when it was actually being used to develop speech recognition, language translation, data mining, search engines, video processing, chess software, etc.
I'm game -- but is there a short explanation of what the difference is and why the term "AI" is bad? I think a big part of the problem right now is a knowledge gap, & I admit I'm frustrated by those--not you--who think everyone shld already know what they know.
"AI" makes people think you're talking about a machine that is a person - it thinks, feels, has wants. That's how AI appears in movies and books. Think HAL in 2001. LLMs just predict and regurgitate text. Some of the recent horrible news stories stem from believing that LLM chatbots are people.
Yes, this I "know"--that is, I've heard this, but I don't really know what that means in practice short of resorting to the metaphysical. I'm not saying there isn't a difference--I'm saying the folks who know need to work on their explanation.
These “tools” are not intelligent. Full stop. They are a mathematical amalgamation of language which it uses statistical inference to “sound” human.
You and I may know they’re not “intelligent”, but based on my discussions with folks who don’t breathe tech everyday, they truly do believe these are intelligent systems. Words matter in those cases.
I don't breathe tech every day, or any day, and I confess that while I appreciate your response, I don't genuinely know what it means. I think many of us feel caught between groups asking us to take their word, & do so according to previous inclinations. A lefty, I take your word. Others don't.
I guess to make it a bit simpler. Think of the Computer from Star Trek. That would be artificial intelligence. It knows facts. It’s know’s the difference between true and false. It can intelligently learn. OTOH, LLMs are the autocomplete on your phone. But much more advanced. 🧵
It knows that the next words or phrase might be this or that. But it doesn’t know if this or that is the correct, factual value. It’s a mathematical guess. And it’s realllllllly good at guessing when it has a vast majority of humanity’s data in its brain. But it still doesn’t *know* the sky is blue
A little bit shorter - pretend we’re playing a ________ where I leave a word or phrase out. Your job is to guess what that word or phrase is. Did you guess game? Congrats, you’re a human LLM 😉
Unfortunately, I haven't really seen Star Trek, so I don't know its computer. I do know a friend asked some chatbot to write an essay as me about something they were wondering about. And it was... very, very good. Some sentences I wish I'd written.
In other words, you're speaking to how it comes to the result, perhaps under assumption that everyday people care about intention. I don't think they do that much. They don't care if their clothes are ethically made or by slave labor; only if they like the way they look.
🤷♂️
That 99 percent of US k-12 teachers have either been give no LLM guidance or have been targeted & marketed to use this crap makes me 😡. 23 years of unregulated use of the web in schools continues to have downsides, hence recent phone bans. Slop is way, way worse for both children and teachers.
OK. Sounds right. What is the language that I should use to reframe AI?