Hacker News Reader

Comment by Closi | Hacker News Reader

> "What I know is that there is no evidence that we are going to get human level artificial intelligence any time soon"

Disagree with this statement - it's too broad. There are plenty of tasks where AI's can fully outperform humans, and the number of tasks where AI's can fully outperform humans seems to increase every week.

Then it depends what we mean by "human level" - the median human is different to the smartest person in the room.

View on HN

A calculator can also outperform a human, and there are many other devices that have outperformed humans for a very long time. Does that mean it is human-level intelligence?

The thing with AI though is that it is being trained to use tools, not to be a tool.

When they can outperform human infants in learning, eg data required to learn and versatility, we can talk business.

Not all world is "big data".

I suggest playing around with gemini's 1 or 2 million context window. You'd be surprised how much you can get from an LLM when you can drop a few textbooks on the topic into the context window before asking it anything.

Mind you, this is also all early gen stuff.

On the flipside, AI's have a super-human ability to consume knowledge that humans can't compete with.

This just seems like putting a human-constraint on AI systems so that we can still classify ourselves as special.

True, but that's also kind of the point - we have grown computers from adding number (calculator 1960s capability) to being able to beat humans at chess (1997) to beating humans at jeopardy (2013) to breaking the turing test (2020+), LLM's that can write code (and test it themselves!), and even creating art and music.

So IMO if we are saying that "there's no evidence that computers will have human level intelligence" we need to take a step back and define what that means - because if you were sitting in the 1960's and you defined it, a modern LLM may very well have already met that definition.

> to breaking the turing test (2020+)

You mean 1967, right? https://en.wikipedia.org/wiki/ELIZA

> However, many early users were convinced of ELIZA's intelligence and understanding, despite Weizenbaum's insistence to the contrary.

"Does X look like Y" is always a continum. In test environments, players are judged human with these rates:

* Real humans 66%

* GPT-4: 49.7%

* ELIZA: 22%

* GPT-3.5: 20%

https://arxiv.org/pdf/2310.20216

(I'm rather surprised by ELIZA beating 3.5, as were the researchers).

Turing's introduction of the test, was a 70% chance of spotting the AI after 5 minutes.

You can't define intelligence by how people in any one point in time would have responded to it. A calculator is AI to people of the 19th century, but not to us. It's only in the passing of time, if something that is thought to be intelligent still seems intelligent long after it is introduced, that we might have a contender.

A moving definition of intelligence isn't particularly useful though if we are trying to establish if something classifies as intelligent or not.

But I think we have at least moved from a state of computers that are less 'intelligent' than a fish, to more 'intelligent' than a dog (assuming by intelligent we mean 'ability to solve problems, and to apply knowledge to novel situations' - i.e. while GPT-4 will make illegal chess moves, it would make less illegal moves than a well-trained dog).

We have moved that quickly from fish to dog, so we need to be careful to not let hubris make us think we are that much more special! Dog-intelligence to human-intelligence is a big leap, but maybe not as much of a leap as fish to dog? Or at least dog to ape and ape to human?

The number of tasks where AI (or any machine) can outperform humans is not necessarily proof of human-level intelligence. I could easily imagine machines replacing the majority of current human occupations well before reaching human-level intelligence.

“The question of whether a computer can think is no more interesting than the question of whether a submarine can swim.” ― Edsger W. Dijkstra

The machine outperformed John Henry, does that make the machine human level intelligence or in the case of AI are there enough human tasks that are pretty easily accomplishable through protocol? A lot of people say things like “it’s so easy a baby could do it”. Nothing in AI means humans aren’t intelligent or that machines are now human intelligent.

The proof is on the pudding, and you cannot let Ai do anything more complex than a relative short prompt, the back and forth is bad and full of pot holes you need to check. It’s like working with a person with multiple levels of knowledge/intelligence sometimes 9/10 sometimes almost 0 and everywhere inbetween

100%.

I was using it with my daughter to help her with math review for high school finals, and was presented by ChatGPT 4o with an impossible triangle as an example problem for her to solve.

We "told" ChatGPT the problem couldn't be solved with this kind of triangle and asked it to come up with other examples NOT using impossible triangles and it couldn't NOT come up with workable examples.

This is always the dividing line in the debate and the place where folks talk past each other. Under normal circumstances people expect “human level” intelligence to be coupled with human level autonomy; but task focused AI often produces super human performance on that specific task with very little autonomy.

Moreover, ML has historically been very brittle compared to the generalization humans provide. A machine that reads addresses might be 10x better than a person, but it may not be worth it if you can’t do things like say “we don’t even sort the blue envelopes—there’s a specific rule for those” like you could with a human. (The specific case here is irrelevant—humans can handle arbitrary variation on the rules infinitely better than a bare CNN.)

It’s this ability to cope with human-level ambiguity that has so much potential in LLMs. But even they don’t work like people—giving it a new rule could make it worse at all other tasks with no explanation as to why.

For me, it depends what is meant by "intelligence".

Even ignoring the people who insist it comes with baggage of consciousness or qualia.

AI can train on millions to billions of examples in a matter of months as transistors outpace synapses by the degree to which marathon runners outpace continental drift… but AI also need to do that well just to get up to the current level.

If AI could learn from as few examples as we require, Tesla would have achieved level 5 driving years ago — but as is, they've got millions of vehicles with years of real world experience and it's still merely "ok", not even "expert".

If we never learn how to make machines learn faster, then we'll have an economy of people whose jobs are essentially "give the machines examples to learn from"… and for some of those jobs, they could spend years providing those examples and still no machine will do as well.

The point I’ve seen raised that seems to escape many is that AI doesn’t need to be better than the best human to be profoundly disruptive - it only needs to be as good as a mediocre human carrying out a rote task to absolutely change the world.

ATM's killed bank tellers for the most part, people changed how they work in banks to move up the value chain. The same thing will happen.

Also, expect a backlash to AI taking human service jobs. Like having to scan a QR code at a restaurant. Its a bad experience that shows you're enraptured by a technology and don't care about the customer.

I don't disagree that AI will be disruptive. I just think it's going to find a place in technical fields and more people will want to remain talking and interacting with people like we've done for the last 100K years.

> ATM's killed bank tellers for the most part,

Has it really? I haven't noticed any drastic decline over the years. Maybe some, but I wouldn't describe it as "killed".

What universal ATMs and cards have really killed off is personal checks, IMO.

There was something like an order of magnitude delcine in the number of teller positions when banks rolled out ATM's. They've been around for so long that most people don't remember when they were uncommon.

But ask yourself this, when did you last go into a bank. I've done a home loan application on a smart phone.

Anyway, stuff is changing, There's always going to be jobs about. People are flexible and can learn things, AI ain't all that.

Yes, but the likelihood of it having similar capabilities to a human just seems really low. If AI develops “cognitive” capabilities, it won’t do so in any sort of human like way, not will it reach equilibrium at “mediocre human” levels of performance. It will rapidly evolve into something we can barely recognize because it’s not constrained by evolutionary boundaries in the same way we are.

> It will rapidly evolve into something we can barely recognize because it’s not constrained by evolutionary boundaries in the same way we are

It is constrained by physical infrastructure like compute and memory, just like the brain which had to become physically bigger to get better.

> it only needs to be as good as a mediocre human carrying out a rote task to absolutely change the world.

And "absolutely chang[ing] the world" is often assumed to be good, but that's not necessarily so. There are a lot of mediocre humans in the world, only capable of rote work. If we push automation to the maximum, eventually we'll reach a point were those people have nothing economically viable to do.

And then you reach a decision point: support them on welfare, forever? Immiserate them and hope they die off (what our current social framework will invariably choose)?

Capitalism requires consumers. Ergo, to keep the wheels turning, either everyone is going to have to become an artisan, or UBI is going to have to be a thing.

> Capitalism requires consumers.

I disagree with that premise. For certain businesses it's fundamentally true, and right now it's broadly true because of the limitations of our technology, but I don't think it's fundamentally true for capitalism.

If the AGI hype pans out, I think we'd see an transition into an economy with a lot fewer consumers. The capitalists would eventually amass all the valuable resources for themselves, and the positive externality that currently drives the consumer economy (the need for labor to make use of capital) would run down and eventually die. Then you'd have a small cadre of elites who own the automation and the resources, and use them chiefly for vanity projects and maybe some B2B activity between themselves (e.g. electricity sales), a small somewhat larger cadre of "middle class" providing luxury/vanity goods to the core elite (e.g. high end prostitutes, Mars colonists), and a great mass of economically useless people living at the margins.

> Ergo, to keep the wheels turning, either everyone is going to have to become an artisan, or UBI is going to have to be a thing.

So the former won't work because artists can't feed themselves by trading art among themselves, and the latter's not necessary because an Elon Musk focused on personal vanity projects doesn't need consumers, he just needs resources an an AGI bureaucracy with AGI robots to command around.

“Human-level” means we can drop the bullshit qualifiers.

Human-level intelligence should adapt to many uses.

It may require teachers, but not prompt engineers.

The article author is very valid in the stance that this is far from intelligence.

Yes yes, you wouldn’t criticize a fish’s intelligence for not being able to climb a chessboard.

But these things are stupid as hell. They are not even close.

Reply on news.ycombinator.com