There are both ML that is not AI, and AI that is not ML.
For example, if you pick them manually, decision trees can be AI but not ML. Video game character behavior is a trivial example.
Eliza for example is also not ML, but could be called AI.
Likewise, there is ML that is not AI. Such is debatable, because you could always argue that using machine-learning on anything results in intelligence. The way I see it, things like image enhancement or voice replacement are not artificial intelligence at all. I probably could not define a hard line where it becomes artificial intelligence though.
To me they're the same thing. If there's a bunch of training data that is fed into a system that creates a model, then it's not traditional programming, where someone laboriously writes out if statements by hand. AI and ML aren't, as far as I'm aware, rigorously specifically defined terms. They're words that marketing picked up and ran with it. To me, what matters is: is there a black box somewhere in the system that's a bag of numbers, or is it code that a human could dig in and read.
It is correct to argue about misleading terminology. "AI" contains the word "intelligence", and for instance logistic regression algorithm is not intelligent, while it is clearly ML, since machine learns something. As Machine learning is broader category, it should include Artificial Intelligence, not vice versa.
Also, 'every single course' is perhaps an overstatement - a course that I co-authored tries to get it right from the first principles.
The machine is learning something so that it can produce outputs based on its learned knowledge. At a high level that seems to be very clearly AI. What am I missing here? You’re probably right, I’m asking genuinely.
It's a matter of definitions, but I can at least understand someone wanting to make a distinction between reactive and non-reactive 'AI' (such as data filters).
There's overlap and edge cases, though: Maybe you have a program that summarizes texts. One could argue that's no different from a passive filter. But can you then ask questions about the text? That's unquestionably AI.
AI and ML have very clear definitions[1]. ML has been a subset of AI, always has been. Latest marketing or "scare quotes" doesn't/shouldn't change that. Especially not in a technical forum like HN.
Laypeople are changing how “AI” is used in common language, like they previously did for “algorithm” and “crypto”.
The textbook definition of AI is a system that solves problems that are difficult for humans. Whether the approach uses formal logic, machine learning, neural networks as a special case of machine learning, optimization, search problems, etc. does not matter.
I never heard people calling machine learning "AI" until large language models made it trivial to market it as such. Like, I remember back when Netflix, for instance, was going around advertising how machine learning (not AI) powers their recommendations.
You should listen better. The University of Edinburgh had an entire Department of Artificial Intelligence when I was an undergrad there in the 1990s, and one of the things it researched was machine learning.
That example kind of illustrates my point though? ML being in an AI book doesn't mean ML is AI, just as being in the dairy section does not make almond milk dairy.
Ed: I disagree. My recollection is that machine learning was routinely sold as “AI” even when it obviously wasn’t. (IBM’s Watson was good at Jeopardy but not real medical applications.)
This isn’t exactly the same, but nothing in the book Paradigms of Artificial Intelligence would be considered AI today.
You must be thinking of a different machine learning. All the on-device machine learning, backend machine learning, OCR, etc. was all called "machine learning" before LLMs. Yes, the field of artificial intelligence still existed, often used machine learning, and called the result "AI". But Apple would call keyboard prediction machine learning. Microsoft would call OCR machine learning. YouTube called machine transcription machine learning. Google called camera image enhancement machine learning.
Microsoft now calls everything AI (actually mostly "Copilot"). YouTube now calls everything AI (including genuine LLMs and generative features, but also everything it used to call machine learning). Google now calls everything AI (including everything it used to call machine learning). Apple is seemingly the only one immune.
My argument is not that no one ever used "AI" to refer to a product that utilized machine learning, but rather that the term of art in the industry for machine learning itself was actually "machine learning", not "AI", until LLMs took over and made it "AI".
You would not pull a library off the shelf for "AI", it would be for machine learning. You would not implement and perform "AI", but machine learning. Even central parts of the AI ecosystem like PyTorch advertise as being for "deep learning", which is a subset of machine learning. Not "AI".
Counter example, the book that is the foundation of much coursework and learning for people in AI, has a whole section on "Machine Learning" with all that k-means and such in there - https://aima.cs.berkeley.edu/
I'm really not sure how that's a counterexample. The section is called machine learning, not AI. Machine learning is a useful tool for artificial intelligence, so I'd be surprised if a book about AI did not talk about it.
Thank you. I was starting to think the history revision was almost true, but your recollection is very much in sync with my own. Everything was machine learning, nobody talked about AI unless it was for research, now marketing has changed that, unfortunately.
> Everything was machine learning, nobody talked about AI unless it was for research
Machine learning was AI. The specific wording was a branding choice, because "AI" was a deeply stigmatized brand. ( https://en.wikipedia.org/wiki/AI_winter ) But there was not a conceptual division.
There's a close analogue to how modern genetic researchers are happy to tell you that your genome is not informative as to your "race", but it is informative as to your "ancestry".
For a long time, AI was a bad word that stood for unfulfilled promise. See AI Winter. Hence, researchers strictly avoided the term while still working on learning algorithms, the same that power LLM training.
It's stupid because there's no formal definition of what counts as AI, so we repeat the same conversation every time. Hence the joke ML is just AI that works.
There's a book from 1995, called Artificial Intelligence: A Modern Approach, by by Stuart Russell and Peter Norvig, which gives the definition "AI is the study of agents that receive percepts from the environment and perform actions." but doesn't define intelligence.
I personally would find it sufficient just to prove both that AI can exist without ML and that ML can exist without AI. The former is easy to prove, since AI has essentially existed since even before computing, while the latter is harder to prove, since ML as a field was created for AI.
> If we go with that, though, I'd say AI is anything that uses a model to make decisions instead of hand written "if" statements.
Hand-written if-statements are a model though...
I would say AI is anything with the intention of performing as a human does. In my opinion, LLMs are AI because they're designed to either perform a human role, or be addressable like a human to a user. (In general I do not agree with implying intelligence by humanity or that it is exclusive to humanity, but it is genuinely the easiest way for me to explain this here.)
Even LLMs that are used as part of harnesses or backend services would count for this, because you could technically replace the model with an equivalent human and they would be able to read and write in its place.
And pre-LLM AI would also count for this, because the whole selling point of AI is that it's like a human in some way or could perform as a human does in some way. Even if it's purely for entertainment value or whatever, if the entertainment value includes that it seems like a human.
But uses of ML that have nothing to do with this would not count, like for example camera image enhancement. You could say a human could sit there and work through some algorithm manually (like from a book, etc) but that's not what image enhancement algorithms are doing. Now if you let an LLM have access to human-UX-ful photo editing tools and let it iterate or etc. then that would be AI but that's another thing.
It's difficult to define precisely exactly what this definition is though. Different people have different ideas of what "performing as a human does" means. If the ML algorithm is executing machine code instructions to perform a task then could a human execute the same machine code instructions by hand to prove that it's AI? No, because the human has to be in place of it, not emulating it. The human would have to take its inputs and give its outputs, without depending on its definition. So, then, it depends on where you define the boundaries of the system. LLMs typically output in tokens, so would a human have to think in tokens? No, because the use of tokens is as a text encoding, and humans are perfectly free to use tools in their work, etc. A human being able to read and write text in English would still prove that a language model that reads and writes the same but in tokens is AI as long as they would be more or less interchangeable.
Of course, today's LLMs are not really true replacements for humans but I'm not talking about performance here, just where they can be positioned. AI to me is more of an intention than a technology -- any technology can be AI depending on how you use it. ML however is a technology, no matter the intention.
I took one longer ago than that and it wasn't all perceptrons and neural networks. It included other things too, like: planning, search methods, inference engines, decision trees, ...
Calling things "robots" is more about the amount of movement. Spinning in place like a washing machine sprayer isn't enough to qualify.
A paint conveyer belt is not a robot. A sprinkler system is not a robot. A CnC machine might be a robot. A conveyer belt that sorts items might be a robot. A roomba is a robot. And all of these function just fine.
I think of robots as general purpose, machines are specific purpose. When it works, we make it single purpose because that’s far far cheaper than general purpose.
>> Cardiac events from Apple Watches is not “AI” though
> It would have been, 20 years ago.
No, it would have been called what it is both then and now; an asynchronous message emitted by a device having sensors capable of detecting when to do so.
The rising LLM = AI equivalency is unfortunate.