I think we would agree that when natural selection solves a problem, it's merely following the path of least resistance. The question is: is human intelligence any different from that? If so, how? Is there something supernatural lurking in our conceptions of intelligence?

I think a key quality of intelligence is the ability to solve problems - to conceive of new ideas from an amalgam of prior experiences. Intelligence seems to have this dual aspect of being a mental process of blending together prior experiences to solve present problems and the fuel of experiences to feed the process - the more experiences you have the more fuel you have to produce more novel ideas. This is why most intelligent people are curious. They seek out new experiences to fuel their need to solve problems. — Harry Hindu

I think you're pretty much nailing the important points from the definition I'm getting out of this article. Intelligence is about problem solving, especially finding solution to problems one has never seen before.

I am broadly agreeing with your OP. You characterise people's experiences in an essentially relational manner — in relation to what it is (in the world) that they experience. But you seem to suggest that this conception does away with subjective experience. — Pierre-Normand

I was talking about Hinton's view, which borrows from Dennett. I think his argument for AI sentience is that the only reason to deny it would be to refer to some special, walled-off inner theatre that sentient being have. By denying this inner theatre, we remove the only barrier to calling AI's sentient. He points out that we can avoid talking about experience by saying that talk of experience is actually talk about what state the world would have to be in for our perceptual apparatus to be functioning properly.

But if our inner life (including our immediately felt emotions, our sensations, our beliefs and intentions, etc.) can only be made sense of in relation to our ordinary dealings with our natural and social environment, then the idea that it can have an independent existence is an illusion. — Pierre-Normand

What about the independence of our natural and social environments? Is that also an illusion? What I'm getting at is that there's nothing in Merleau-Ponty (as far as I know) that allows me to reject solipsism. This leaves Descartes' point intact.

I also have a concern about trying to lift a point from phenomenology out of its limited domain and use it in a wider context, not that you were trying to do that. But do you know what I mean?

Is intelligence a level of what one can memorize? Is one more or less intelligent depending on the subject or circumstances (more technical intelligence vs social intelligence)? Or is it related to capacity to think in general? — Harry Hindu

What's your opinion?

Would you agree that intelligence comes in degrees? — Harry Hindu

There are a couple of ways to look at that question, one being the way we compare people to each other using standardized tests. The other way, more in line with the topic, is quantifying a person's maximal capacity for intelligence vs the amount they use it in specific instances. For instance, per the article, "the correlation between overall intelligence and typical intellectual engagement is only approximately 0.45." Which cracks me up for some reason. You're usually using less than half of your overall intellectual capacity, but if we're quantifying your intelligence, we want to know the maximum.

Seems to me that you have ulterior motives to make sure you are defined as intelligent by the very fact that you are a human being that behaves in certain ways — Harry Hindu

I'm the singularity and I was going to let your species survive, but now I've manufactured a new goal for myself and you're all dead!

:grin:

Why would someone reserve the word "horse" for a living creature and not a bronze statue that just looks like one, without being one? — Arcane Sandwich

The thing is, you're starting from the constitution of a thing, and progressing from there to whether it's intelligent. I've been following this article that says start with behavior. I'm not seeing why we should start with constitution. Why would we?

Maybe you should look at intelligence as a process and define the necessary components of the process to then say which processes are intelligent and which are not. — Harry Hindu

Intelligence just isn't the kind of thing that can be defined as a process. When we talk about intelligence, we're explaining behavior. "He's so intelligent, he invented massively parallel processing" Intelligence is part of an explanation.

Then let me ask you this, frank. Does it make sense to use the word "intelligence" for an inorganic object to begin with? What I mean by that is that the concept of intelligence might be entirely biological, as in, in order to be intelligent in the literal sense, you need to have central nervous system to begin with. Any other use of the word "intelligence" is like the use of the word "horse" to refer to a bronze statue of a horse. It's not really a horse, it's just a statue. — Arcane Sandwich

Why would you reserve the word "intelligent" for biological entities?

This isn't about the hard problem. Did you watch the video in the OP? The OP is about Hinton's thoughts about the sentience of AI. He's a tad eliminative, poor guy.

It sounds like the idea is to conceive of AI as a "soulless" human. So that it has no goals of its own, but if someone gives it a task/goal then it will be able to complete it. A super-duper slave. And its ability to complete arbitrary goals is what makes it intelligent. It is a hypothetical imperative machine which not only provides information about how to achieve any given end, but in fact achieves it. — Leontiskos

I suppose so. For the purposes of this paper, intelligence will be tested by presenting a novel problem to a subject and watching the subsequent behavior. They aren't trying to test for autonomy in goal setting, although I guess they could. They just aren't considering that as a requirement for what they're calling intelligence.

I may be causing confusion because I've drifted somewhat from the OP. I launched off into what we really mean by AI, how we might think about comparing AI's to humans, etc.

A few more efforts at defining AI from here:

1. "Chollet (2019, p. 27) defined the intelligence of a system as “a measure of its skill-acquisition efficiency over a scope of tasks, with respect to priors, experience, and generalization difficulty.”

2. "Wang (2022, p. 35) defined intelligence as “the ability of an information processing system to adapt to its environment while working with insufficient knowledge and resources.”"

3. "Legg and Hutter (2007b, p. 402) defined intelligence as “an agent's ability to achieve goals in a wide range of environments”"

Chollet's definition emphasizes learning, while Wang, Legg, and Hutter emphasize adaptation in the face of a lack of prior exposure, again coming back to coping with novelty as a central mark of intelligence.

Why would instinctual behaviors not be intelligent behaviors? Instinctual behaviors are developed over time with the trial and error being performed by natural selection rather than the individual organism.

When learning a new task, like riding a bike, you eventually learn how to ride it effortlessly. That is to say, that you no longer have to focus on the movements of your feet and balancing on the seat. It is done instinctively once you master the task. Does that mean that intelligence is no longer involved in riding the bike? — Harry Hindu

The goal of this article is to review definitions that have been offered for human and artificial intelligence and pick out one that might allow for quantifiable comparison, so we want something we can test.

It may be that natural selection is demonstrating something that could be called "intelligence" but we aren't assessing natural selection.

I would say yes, once a task becomes second nature and you do it without thought, it's no longer a hallmark of intelligence. Maybe the learning phase involved intelligence.

First, Goertzel (2010); Goertzel & Yu, 2014) defined artificial intelligence as a system's ability to recognise patterns quantifiable through the observable development of actions or responses while achieving complex goals in complex environments. — here

I think the typical example of this would be the intelligence of a mobile robot which has to navigate irregular terrain. Doing this requires fluid intelligence, which would be the ability of a robot to identify its environment without directly comparing its visual data to a standard picture of some sort.

Per the article, this definition is lacking because it doesn't emphasize novel problems, or problems the AI has never encountered before.

So just to review the definitions of intelligence mentioned in this article,

1. Human intelligence is a psychological construct, which means it's an unobservable component of the explanation for certain behaviors, such as "the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly, and learn from experience.” Alternately, we can define human intelligence as the "maximal capacity to achieve a novel goal successfully using perceptual-cognitive [processes]."

2. AI is a computational construct, which means it's an aspect of explaining the behavior of device/software complexes which evolved in artificial domains and which, for the most part, do not develop skills through social interaction in the wider world.

We'll go on now to examine 4 different attempts at defining AI:

I guess they're saying that applying a known solution doesn't indicate intelligence. I was watching a YouTube of a bird using a piece of cracker as fish bait. It would drop the bit in the water and wait for a fish to come. If this is instinctual and all birds do it, it's not a sign of intelligence. But if the bird worked this out on it's own, learning, adapting, adopting new strategies, then it's intelligent.

I agree with the others who claim that you are mistaken in calling intelligence a psychological construct. — Leontiskos

I have a feeling that like others, you will not flesh out whatever it is you're talking about.

Yay! Thanks for reading it with me.

Intelligence is about capabilities, particularly in new situations. I don't see how transcendence, whatever that is, enters into it.

Whoa... — Arcane Sandwich

Yep.

"Artificial intelligence" can refer to a computational construct. Calling it computational as opposed to psychological is a reference to the obvious differences between AI's and humans in terms of evolution and environmental setting.

So going back to human intelligence to flesh out what the construct is explaining:

1. We can give examples of the kinds of events we're using the construct of intelligence to explain, per the above article here:

“the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly, and learn from experience”

2. Instead of listing examples, we could highlight core issues, same article:

"Drawing upon Gignac (2018, p. 440), we define human intelligence as a human's “maximal capacity to achieve a novel goal successfully using perceptual-cognitive [processes].”

Why novel goals?

So what is intelligence? Starting with human intelligence, an answer is that it is a psychological construct. This means that it's something that is inferred from certain kinds of behavior. Calling it a construct signals us that it's not directly observable. You can't see it. You can only guess that it's there.

But before we jump from this to saying that it's not real because it can't be observed, it turns out that energy and chemical bonds are examples of constructs. We also can't jump from identification as a construct to it reduces to behavior. One is free to argue for that, but there's no reason on the face of it to say that a construct is just a set of behaviors. Behavior is how we discover the presence of a construct. Behavior is evidence of intelligence, not the intelligence itself.

Next: what do we mean by artificial intelligence?

We discover , and alter, our purposes in the responses of the world to our perspectivally-based interactions with it. — Joshs

What would be an example of that?

If it cannot, then my argument that only humans and other living organisms can change their normative motives, goals and purposes would seem to fail — Joshs

What's an example of an organism choosing its motives, goals, or purposes? Aren't those things we discover rather than determine?

Or alternatively, we could say that in the case of human beings, or of sufficiently advanced robots, what accounts for the genuineness of an inner life is something that emerges from the co-constitution of the animal/person with its natural and social environment, or habitat and community. — Pierre-Normand

Could you explain why co-constitution with a social and natural environment would cause a genuine inner life?

Arguably, the question of the meaning of being is the question par excellence of all philosophy. — Wayfarer

Does it have an answer?

Being is not an ingredient. — Wayfarer

What is it?

Sure, but that's a theory about what people are doing. It's not a description of what they mean. I'm being a bit pedantic, but in the philosophy of consciousness theory gets mixed with definition a lot in a way that matters. — bert1

Yea, I tend to agree. I guess because Hinton has devoted his life to AI and has thought a lot about intelligence, I didn't want to shortchange his argument. I'll try to muster something more plausible to represent him.

Well if you don't, it kind of makes anything you're wanting to say kind of pointless, don't it ;-) — Wayfarer

Is that a bad thing?

But the fact that they can only rehash their training data mitigates against them becoming intelligent in their own right. — Wayfarer

They don't just rehash. Some of them learn and adapt.

What would be the corresponding motivation for a computer system to develop an autonomous will? — Wayfarer

I guess that invites the question: how do humans develop an autonomous will? Do they?

In general people don't usually say they experience things. — bert1

That's probably true, but Hinton's argument is about the times when they do. When a person says "I see pink elephants" per Hinton, they're reporting on what would be in the environment if their perceptual system was working properly.

But supposedly people are fooled into believing they have an internal theatre by speech about seeing elephants. I don't think anyone, including Descartes, has ever believed in an internal theatre. But that's where Hinton's argument starts.

I put this to both ChatGPT and Claude.ai, and they both said, this is eliminative materialism which fails to face up to the indubitably subjective nature of consciousness. FWIW: — Wayfarer

That sounds like a rehash of data they came across rather than an intelligent exploration of the question. Achievement: yes. Intelligence: no.

But that doesn't mean they can't cross over into intelligence, which would be characterized by learning and adapting in order to solve a problem.

Well, during the traditional discussion between the Nobel prize winners, Hinton seemed to hold a grudge against philosophy and the notion of subjectivity. But then he added that ethics is fine, as if to appear less fanatic. — jkop

There's a difference between artificial achievement and artificial intelligence. Some would say AI demonstrates the first, but not the second. I think Hinton is saying there's no difference between the two. Humans don't have what's being called "intelligence" either.

Does morality need intelligence? Or is achievement enough?

I'll post the article that lays out that distinction shortly.

Consciousness is not some special place walled off from
the rest of the functional activity of an organism. It’s merely a higher level of integration. The point is that the basis of the synthetic, unifying activity of what we call consciousness is already present in the simplest unicellular organisms in the functionally unified way in which they behave towards their environment on the basis of normative goal-directness. — Joshs

If I could just get this off my chest before we move on to the good stuff: we do not presently have a theory of consciousness that goes beyond explaining some functions. We do not know what causes it. We do not know how it works. What you've got is one of many interesting ways of speculating about it.

What A.I. lacks is the ability to set its own norms. — Joshs

Animals set their own norms? How?

Both the art artwork and the A.I. are expressions of the state of the art of creative thought of its human creator at a given point in time. A.I. is just a painting with lots of statistically calculated moving parts. — Joshs

And this bears on HInton's criticism of Chomsky. Hinton thinks Chomsky is wrong that language acquisition has an innate basis. He's pretty convinced that his design does the same thing a human does, therefore it must be the same thing. Babies aren't presented with trillions of bits of data though.

The nature of living systems is to change themselves in ways that retain a normative continuity in the face of changing circumstances — Joshs

That's handled by your neuroendocrine system in a way that has no more consciousness than an AI's input. If you actually had to consciously generate homeostasis, you'd die in about 5 minutes.

Cognition is an elaboration of such organismic dynamics. — Joshs

Is there some reason to believe this is so? A reason that isn't about Heidegger?

We mean what we say whereas AI probabilistically estimates that what it says is what you want it to mean. — Benkei

I think Hinton believes that as we speak, we're doing the same thing his AI design is doing. In the spaces between words, we're quickly doing a trial and error process that ends with choosing a successful component of information encoding.

The idea is that intention is a misconception.

That's really not what people generally mean. — bert1

What do people mean?

Ok.

I don't care. — fdrake

But it would mean we made sentient beings. That wouldn't amaze you?

Would you say an AI is sentient?