  > that does not have a technical meaning

I don't think the definition is very refined, but I think we should be careful to differentiate that from useless or meaningless. I would say most definitions are accurate, but not precise.

It's a hard problem, but we are making progress on it. We will probably get there, but it's going to end up being very nuanced and already it is important to recognize that the word means different things in vernacular and in even differing research domains. Words are overloaded and I think we need to recognize this divergence and that we are gravely miscommunicating by assuming the definitions are obvious. I'm not sure why we don't do more to work together on this. In our field we seem to think we got it all covered and don't need others. I don't get that.

  > In this view, if a machine performs a task as well as a human, it understands it exactly as much as a human.

And I do not think this is accurate at all. I would not say my calculator understands math despite it being able to do it better than me. I can say the same thing about a lot of different things which we don't attribute intelligence to. I'm sorry, but the logic doesn't hold.

Okay, you might take an out by saying the calculator can't do abstract math like I can, right? Well we're going to run into that same problem. You can't test your way out of it. We've known this in hard sciences like physics for centuries. It's why physicists do much more than just experiments.

There's the classic story of Freeman Dyson speaking to Fermi, which is why so many know about the 4 parameter elephant[0], but it is also just repeated through our history of physics. Guess what? Dyson's experiments worked. They fit the model. They were accurate and made accurate predictions! Yet they were not correct. People didn't reject Galileo just because the church, there were serious problems with his work too. Geocentricism made accurate predictions, including ones that Galileo's version of Heliocentrism couldn't. These historical misunderstandings are quite common, including things like how the average person understands things like Schrodinger's Cat. The cat isn't in a parallel universe of both dead and alive lol. It's just that we, outside the box can't determine which. Oh, no, information is lossy, there's injective functions, the universe could then still be deterministic yet we wouldn't be able to determine that (and my name comes into play).

So idk, it seems like you're just oversimplifying as a means to sidestep the hard problem[1]. The lack of a good technical definition of understanding should tell us we need to determine one. It's obviously a hard thing to do since, well... we don't have one and people have been trying to solve it for thousands of years lol.

  > Just my opinion, but my professional opinion from thirty-plus years in AI.

Maybe I don't have as many years as you, but I do have a PhD in CS (thesis on neural networks) and a degree in physics. I think it certainly qualifies as a professional opinion. But at the end of the day it isn't our pedigree that makes us right or wrong.

[0] https://www.youtube.com/watch?v=hV41QEKiMlM

[1] I'm perfectly fine tabling a hard problem and focusing on what's more approachable right now, but that's a different thing. We may follow a similar trajectory but I'm not going to say the path we didn't take is just an illusion. I'm not going to discourage others from trying to navigate it either. I'm just prioritizing. If they prove you right, then that's a nice feather in your hat, but I doubt it since people have tried that definition from the get go.

▲ robotresearcher a day ago | parent [-]

> It's a hard problem

So people say.

I’m not sidestepping the Hard Problem. I am denying it head on. It’s not a trick or a dodge! It’s a considered stance.

I'm denying that an idea that has historically resisted crisp definition, and that the Stanford Encyclopedia of Philosophy introduces as 'protean', needs to be taken seriously as an essential missing part of AI systems, until someone can explain why.

In my view, the only value the Hard Problem has is to capture a feeling people have about intelligent systems. I contend that this feeling is an artifact of being a social ape, and it entails nothing about AI.

▲ pastel8739 a day ago | parent | next [-]

Regardless of whether you think understanding is important, it’s clear from this thread that a lot of people find understanding valuable. In order to trust an AI with decisions that affect people, people will want to believe that the AI “understands” the implications of its decisions, for whatever meaning of “understand” those people have in their head. So indeed I think it is important that AI researchers try to get their AIs to understand things, because it is important to the consumers that they do.

	▲	robotresearcher 13 hours ago \| parent [-]
		I agree with this. I contend that as the AIs improve in performance, the designation of understanding will accrete to them. I predict there will never be a component, module, training process, or any other significant piece of an AI that is the ‘understanding’ piece that some believe is missing today. Also, the widespread human belief that something is valuable has absolutely no entailments to me other than treating the believers with normal respect. It’s very easy to think of things that are important to billions that you believe are not true or relevant to a reality-driven life.

▲ godelski a day ago | parent | prev [-]

It's a sidestep if your stance doesn't address critiques.

  > needs to be taken seriously as an essential missing part of AI systems, until someone can explain why.

Ignoring critiques is not the same as a lack of them

▲ Zarathruster a day ago | parent | next [-]

While I agree with you in the main, I also take seriously the "until someone can explain why" counterpoint.

Though I agree with you that your calculator doesn't understand math, one might reasonably ask, "why should we care?" And yeah, if it's just a calculator, maybe we don't care. A calculator is useful to us irrespective of understanding.

If we're to persuade anyone (if we are indeed right), we'll need to articulate a case for why understanding matters, with respect to AI. I think everyone gets this on an instinctual level- it wasn't long ago that LLMs suggested we add rocks to our salads to make them more crunchy. As long as these problems can be overcome by throwing more data and compute at them, people will remain incurious about the Understanding Problem. We need to make a rigorous case, probably with a good working alternative, and I haven't seen much action here.

	▲	godelski a day ago \| parent [-]
		`> "why should we care?"` I'm not the one claiming that a calculator thinks. The burden of proof lies on those that do. Claims require evidence and extraordinary claims require extraordinary evidence. I don't think anyone is saying that the calculator isn't a useful tool. But certainly we should push back when people are claiming it understands math and can replace all mathematicians. `> If we're to persuade anyone, we'll need to articulate a case for why understanding matters` This is a more than fair point. Though I have not found it to be convincing when I've tried. I'll say that a major motivating reason of why I went into physics in the first place is because I found that a deep understanding was a far more efficient way of learning how to do things. I started as an engineer and even went into engineering after my degree. Physics made me a better engineer, and I think a better engineer than had I stayed in engineering. Understanding gave me the ability to not just take building blocks and put them together, but to innovate. Being able to see things at a deeper level allowed me to come to solutions I otherwise could not have. Using math to describe things allowed me to iterate faster (just like how we use simulations). Understanding what the math meant allowed me to solve the problems where the equations no longer applied. It allowed me to know where the equations no longer applied. It told me how to find and derive new ones. I often found that engineers took an approach of physical testing first, because "the math only gets you so far." But that was just a misunderstanding of how far their math took them. It could do more, just they hadn't been taught that. So maybe I had to take a few days working things out on pen and paper, but that was a cheaper and more robust solution than using the same time to test and iterate. Understanding is a superpower. Problems can be solved without understanding. A mechanic can fix an engine without knowing how it works. But they will certainly be able to fix more problems if they do. The reason to understand is because we want things to work. The problem is, the world isn't so simple that every problem is the same or very similar to another. A calculator is a great tool. It'll solve calculations all day. Much faster than me, with higher accuracy, but it'll never come up with an equation on its own. That isn't to call it useless, but I need to know this if I want to get things done. The more I understand what my calculator can and can't do, the better I can use that tool. Understanding things, and the pursuit to understand more is what has brought humans to where they are today. I do not understand why this is even such a point of contention. Maybe the pursuit of physics didn't build a computer, but it is without a doubt what laid the foundation. We never could have done this had we not thought to understand lightning. We would have never been able to tame it like we have. Understanding allows us to experiment with what we cannot touch. It does not mean a complete understanding nor does it mean perfection, but it is more than just knowledge.

▲ robotresearcher 13 hours ago | parent | prev [-]

Critiques should come with some argument if they want to be taken seriously.

If I say it’s not real intelligence because the box isn’t blue, how much does anyone owe that critique? How about if a billion people say that blueness is the essence missing from AIs?

Tell me why blue matters and we have a conversation.