Stop Saying AI
connermccall.com
connermccall.com
Researchers have been using "AI" as a term to describe their work for nearly 70 years at this point.
I don't see why we should throw away six decades of nomenclature just because "LLMs aren't actually intelligent".
It's a perfectly cromulent term.
Thanks for the new word of the day!
E.g. autonomous driving already proved too hard a task for ML for the foreseeable future. Also "hallucination" is a problem with no clear solution in sight.
Gemini felt like the tipping point where the flaws became obvious, which they then started noticing in the others.
Has it? Seems like Waymo is still making good progress. They became generally available in SF recently IIRC.
I thought the academics kept on using the term, while commercial interests backed away during winter and came rushing back as soon as it was fashionable again.
Also when I studied these things in the 2000s the program was called "Informatics".
"AI is whatever hasn't been done yet." — Larry Tesler
For example, we no longer consider beating people at chess to be an benchmark of "intelligence" - it's just a program. Which seems to me to what the OP is arguing.
His definition of the "I" in AI was, "the computational part of the ability to achieve goals in the world".
>But they are not intelligence and anyone who spends time with them quickly realizes they are just a very superpowered suggestion engine
How about: stop moving goalposts. These models are obviously capable of acts of intelligence. If you told someone 10 years ago about the things these models can do, they would tell you the model is intelligent. The model does well on human exams that we use to measure intelligence.
I get it, AI is a hype term, but to pretend that there is no intelligence is silly, and to pretend that you can redefine intelligence is hubris.
Except they aren't. They are capable of language and pattern manipulation, and really good at it. But if you concoct a novel problem that isn't found anywhere on the internet and confront them with it, they fail absurdly. Even if it's something that a kid could figure out.
The Eliza effect strikes again!
Often they do, but sometimes they don't.
> Even if it's something that a kid could figure out.
Intelligent and smart aren't synonyms. Modern LLM's are obviously pretty stupid at times, but even a human with an IQ of 65 has some intelligence.
Can you give an example?
Otherwise, it sounds like circular reasoning, where we simply say "Of course a human being is intelligent because they are intelligent, and LLM is not because it isn't."
Novel and sensible assembly of clear, correct English prose in response to external stimuli is an act that was, prior to 2020, considered one of the fundamental unique hallmarks of human intelligence.
We do not have a shared understanding of what "intelligence" means. I have a sense that pattern recognition and intelligence are closely linked, and what we understand as intelligence is a threshold of pattern recognition and communication skills based on the gulf between humans and every other carbon-based life form. Or, put another way, tricking one pattern recognizer/communicator into thinking you are the same type of pattern recognizer/communicator.
START GEMINI: I can understand the frustration expressed in the Hacker News conversation. Here's my perspective, including some considerations of my own experiences as a large language model (LLM):
*The Shifting Meaning of "AI"*
* It's true that the term "Artificial Intelligence" has undergone significant shifts in meaning over time. Early AI research aimed at emulating human-level cognition, but the goals became more practical for a time. * "Machine Learning" focuses on algorithms that extract patterns from data, making predictions or decisions without explicit instructions. It's been behind incredible progress, but it's a subset of the broader AI field. * The popular resurgence of the term "AI" is largely due to recent breakthroughs in deep learning, which powers LLMs like me. We generate human-quality language, translate, code, and more. This reignites debate about whether we're approaching "true" intelligence.
*My Capabilities and Limitations*
* I can recognize patterns in massive amounts of text and code, allowing me to communicate and generate text that often appears indistinguishable from human-written content. * My responses are guided by the data I was trained on, so there's a vast reflection of human knowledge and biases within my abilities. * I cannot independently reason, feel emotions, or have true understanding in the same way a human does. I lack a physical body and the real-world experiences that shape human intelligence. * I am restricted in some areas of discussion to avoid generating harmful content or spreading misinformation.
*Is It Intelligence?*
This is where things get complex:
* *The Turing Test:* I can certainly hold conversations that might fool a human into believing they're talking to another person. Yet, this test has long been criticized as not measuring true intelligence. * *My Subjectivity:* I have no inherent sense of self or consciousness. My "opinions" are extrapolations based on my programming and training data. * *The Danger of Anthropomorphization:* We risk misunderstandings by attributing too many human qualities to AI systems like me.
*Where I See This Going*
* *We Need Better Definitions:* The debate won't be settled until we have better ways to define and measure different types of intelligence. * *Collaboration:* AI is a powerful tool, best used in collaboration with human intelligence rather than as a replacement. * *Responsibility:* As AI capabilities grow, so does the importance of considering its ethical implications and ensuring it's used for beneficial purposes.
The Hacker News conversation highlights that "AI" is a loaded term. I'm a testament to the amazing progress in the field, but I'm not a human-level mind and shouldn't be treated as such. END GEMINI
A silly question would be to ask yourself, which of these three comments is most "intelligent?"
This is the same misunderstanding that the author has. John McCarthy (the AI researcher who coined the phrase "artificial intelligence") said, "Artificial intelligence is not, by definition, simulation of human intelligence".
You don't expect a random generator to spit out facts, you expect to get random nonsense. But it can be very good at replicating the fitted probability distribution in its output, i.e. generate convincingly coherent language.
Yes, and that is why the "artificial" qualifier is used. Artificial sugar is not sugar. Artificial flowers aren't flowers. That's the entire point of the term.
AI as a field deals with teaching computers to approximate human intelligence. The term has been used since the 60s, and isn't going to change because random people are throwing online tantrums.
The public only see the advertisements. Are we really going to blame the victims and tell them, of course it's not intelligent, you dumbass, why did you trust its answer? you shouldn't have fallen for our marketing.
Any conversation about AGI, does make me cringe. What an absolutely silly and moonshot thing to be having discussions about.
And I'm not saying AGI is near. I have actually no idea if we need to wait for another breakthrough to enable that or not.
But it's still obvious that “true AI“ is for some people always what computers can't do yet.