[tialaramex]

> those systems are not a path to AGI

If that's the actual meat of your claim, it's weird to just put it unsupported at the end. If you think it's an aside you've got a big problem, because without it the rest is pretty weak.

We saw this with chess. Anybody who was clinging on to the idea that well, the machines can't really play chess, because technically there do seem to be a few humans who are better, was screwed the moment the machines begin routinely beating grand masters. Either this categorically isn't something that the machines can do, or, it is and so it's important that they can do it at all. We shouldn't expect any half measures on this.

Do you consider Rob Liefeld an artist? What do you reckon the chances are that the machine can get human anatomy right more often than Rob Liefeld? Rob is a human, so it seems like that should give him an advantage. However unlike Rob DALL-E has learned by seeing lots of existing pictures of humans that they don't look the way Rob draws them...

[slowmovintarget]

On "not the path to AGI": Gary Marcus on the Mindscape podcast is worth a listen.

https://www.youtube.com/watch?v=ANRnuT9nLEE&list=PLrxfgDEc2N...

Transcript: https://www.preposterousuniverse.com/podcast/2022/02/14/184-...

"...And then there’s natural language understanding and reasoning, and I would say we have not really made progress at all. GPT-3, which we may wanna talk about, gives the illusion of having natural language of understanding, but I don’t really think that it does. And we are nowhere near, for example, an all-purpose general assistant. ..."

[lostmsu]

> I don’t really think that it does

Either the quote is poorly chosen or reading this article is not worth the time.

[slowmovintarget]

That's a false dichotomy. Select the name Gary Marcus, right-click, and search (on my browser it defaults to Duck Duck Go, but that returns the right result).

The Mindscape Podcast is hosted by Sean Carroll. You have a very sharp quantum physicist interviewing an expert in the field of AI research.

The podcast is worth the time, and the quote is representative of an expert's take on the matter. He elaborates, but I don't need to write an essay just to argue on the internet.

[challenger-derp]

For anyone who's reading the above comment, the additional context that slowmovintarget hasn't provided is that Gary Marcus supports a school of thought in AI that opposes the currently popular school of thought that favors deep neural networks. A frequently contested point is what each school thinks is the path to AGI. Marcus is a well-known figure in AI.

[krageon]

This argumentation is based on the "feeling" that something doesn't understand things the way they do and that this is sufficient to say that it is inferior. It's a very old argument against AGI, and it's as boring as it ever was: It relies on human exceptionalism and on the concept of the soul (i.e. something humans have and other things can never have, which cannot be quantified or understood). It is compelling only to those who are religious or who still hold religious tendencies.

[YeGoblynQueenne]

>> Anybody who was clinging on to the idea that well, the machines can't really play chess, because technically there do seem to be a few humans who are better, was screwed the moment the machines begin routinely beating grand masters.

Who was "clinging on to the idea that well, the machines can't really play chess" and when was that?

[janalsncm]

Made an account so I could mention David Levy’s infamous bet: https://en.m.wikipedia.org/wiki/David_Levy_(chess_player)#Co...

That’s the way it is with AI: first people say X is impossible for computers, then they say X is hard, then they say X doesn’t work for certain edge cases, then they say we’ve known all along that computers could do X.

[YeGoblynQueenne]

Why "infamous"? In any case, Levy did win his bet when there had been no computer chess engine that could beat him in ten years after betting against McCarthy and Michie [1]. Despite that he acknowledge that chess engines had improved further than he had thought:

>> Levy wrote, "I had proved that my 1968 assessment had been correct, but on the other hand my opponent in this match was very, very much stronger than I had thought possible when I started the bet."[37] He observed that, "Now nothing would surprise me (very much)."[38]

(Edit: the quote is from your WP link.)

So that at least really does not match the pattern you mention. Levy's evaluation of chess engines was pretty much in the water, all the way up to Kasparov's loss to Deep Blue. As far as I can tell he never said anything like "the machines can't really play chess" as the OP suggested. He just made a prediction about how much they could advance in ten years. And he was right.

I would therefore like more justification for your assertion that "that's the way it is in AI". I would also like a clarification: who are the "people" who say all those things? How representative are they of experts and researchers?

Who makes all those naive predictions about the impossibility of AI that are later proven wrong? For the time being, it seems that naive predictions can be traced more directly to luminaries of the field, like Alan Turing or Marvin Minsky [2], who predict that AI is "just around the corner", rather than skeptics and naysayers who say it wont' happen, as is usually suggested.

And then of course, there's Rodney Brooks' dated predictions, so far standing the test of time (although that's a short time!) [3].

_______________

[1] Totally coincidentally, Donald Michie was my thesis advisor's thesis advisor, so I'm, like, his academic grandchild.

[2] See: https://web.eecs.umich.edu/~kuipers/opinions/AI-progress.htm...

"In 1958, Herbert Simon and Allen Newell wrote, “within ten years a digital computer will be the world’s chess champion”; note that this is 10 years before Levy and McCarthy's bet.

[3] https://rodneybrooks.com/predictions-scorecard-2022-january-...

[dougabug]

Well, Levy won the bet, but just barely. More importantly, his dismissive attitude towards AI mastering chess (“[…] the idea of an electronic world champion belongs only in the pages of a science fiction book.”) was deeply shaken.

[YeGoblynQueenne]

Sure, but I just think it's all a bit more nuancend than the way the OPs make it to be. There's always a lot of speculation that goes both ways: either AI is just around the corner, or it's never gonna happen. And there's plenty of opinions and educated guesses in the middle, always.

[ajafari1]

Right on. When do you expect a productized AI that can replace your average skilled programmer?

[ben_w]

I'm not sure what average looks like. Possibly last August? Possibly some time around 2035? It'll be a moving target though, as anyone worse than the AI isn't likely to stay in the role for long.

[carapace]

Here's me almost exactly six years ago, predicting "most coding and design tasks will be automated within three to five years."

https://news.ycombinator.com/item?id=11718300

If we count GitHub Copilot as the tip of the spear then I was too optimistic.

> GitHub Copilot ... was first announced by GitHub on 29 June 2021

https://en.wikipedia.org/wiki/GitHub_Copilot

[YeGoblynQueenne]

You were too pessimistic! Automatic programming, a.k.a. program synthesis, has been a thing since the early days of AI and computer science, for example the idea of deductive program synthesis, where a program is generated wholesale from a complete specification in a formal language that is not a computer language goes at least as back as Alonzo Church himself, in 1957:

https://en.wikipedia.org/wiki/Program_synthesis#Origin

Much work has been done since then and inductive program synthesis from incomplete specifications consisting of input/output examples (a form of machine learning) has been possible for quite some time. This is the second time this week I point to the report by Gulwani et al for a modern overview (of both kinds):

Program Synthesis

https://www.semanticscholar.org/paper/Program-Synthesis-Gulw...

Rather than Copilot being the "tip of the spear", it is really a step back, a system that can only generate code but not check its correctness, unlike pretty much every program synthesis system since forever. Although this may actually be an advantage for its commercial application (since it makes it easier to match expectations of the system's performance) it is not really any indication of progress, in any way, shape or form. In truth, Copilot is famous today because it is supported by a large company like Microsoft and because earlier work is not well-known by most people who get their AI knews from blogs and podcasts, who are also not very well aware of the history of the field. But, "tip of the spear"? Oh, no. Unless it's a very blunt, toy spear, more like a tech demo of a spear.

And this is by no means restricted to program synthesis, and Copilot. For example, the first self-driving car to drive on a stretch of real road, with real traffic, without human intervention, was Ernst Dickmann's 1995 robot car:

https://people.idsia.ch/~juergen/robotcars.html

Again: 1995. And yet, fully autonomous self-driving cars are still not here. The revolution hasn't happened and progress has only crawled marginally forward.

[tialaramex]

Dickmann's car does Autobahn (~US freeway) traffic under controlled conditions. It doesn't need to understand junctions or traffic signals, there aren't any. It doesn't need to understand pedestrians, there aren't any. It doesn't need to understand bicycles, there aren't any. What happens if things go wrong? In practice a human takes the wheel, this is not a system capable of safely leaving traffic when it can't cope, so it was never demonstrated without what we'd today call a "safety driver". That's not autonomous self-driving cars minus a little bit more research, it's improved Cruise Control.

Waymo has taxi service. It's losing money doing that, and it's only in a few select places, but it's doing autonomous journeys, with no "safety driver" on city streets. It understands junctions, and traffic signals, pedestrians and cyclists, because on the city streets those things are commonplace. It's doing the hard part.

I don't think that constitutes "crawled marginally forward", it's a considerable advance.

[carapace]

That's right, this is your field isn't it? Cheers!

Gosh, the robot car work was incredible, I can't believe I've never heard of it before!

- - - -

Copilot isn't cutting edge but it's significance is just that it's a mass market tool. It will be interesting to see how well it does, and whether users find it worthwhile overall after a couple of years. Will it be improved to the point where it starts to compete with its users?

I've asked other programmers before, if you could write a program that replaced you would you do it?

(One of the reasons I like Schmidhuber is that his goal, since early on, is to "Create an automatic scientist, and then retire.")

To me the ultimate effect of programming is to obviate the programmer.

[YeGoblynQueenne]

Ah, my field is Inductive Logic Programming - a sub-sub-sub-sub field of program synthesis. But I need to know the basics!

I think Copilot can be used very effectively, as long as its capabilities and their limitations are communicated clearly. For instance, I think it can make a great boilerplate generator, as long as users stick to short code snippets.

Well, I don't know about replacing programmers. I think that's Sci-Fi, for the time being, and for a while longer still. What I'm more interested in is creating tools to help programmers do their job. Copilot does that already, btw, I'm not dissing it. I'm just pointing out it doesn't represent a sudden shift in capabilities, to be clear.

>> (One of the reasons I like Schmidhuber is that his goal, since early on, is to "Create an automatic scientist, and then retire.")

I didn't know Schmidhuber had said that. My thesis advisor, Stephen Muggleton, was part of an interdisciplinary team who created a robot scientist that can develop its own theories and then choose, and run, the experiments to prove them:

https://en.wikipedia.org/wiki/Robot_Scientist

Another one of those things that are not well-known, I guess. I wasn't involved with that, btw, but I think recent advances could make for a much more powerful system. I am considering something similar as a research project, post-doc.

[carapace]

Good lord! Sometimes I think the only thing that has prevented the techno-singularity happening already is the tendency of humanity to strenuously ignore prior art.

Cheers!

[dougabug]

A wise old inventor told me, early in my career, that “people always significantly overestimate the rate of progress in the short run, and greatly underestimate it in the long run.”

Text conditional image generation is still a fairly new thing, we don’t know what its upper limits will prove to be. Also, it seems likely that, whatever level of capability they reach, humans will learn how to use them effectively as tools for their own purposes. Some people might use them for quickly preflighting ideas, others to create boilerplate, some people may use them to speed up their workflow or scale up productivity. People didn’t know what all the use cases for personal computers would be when they were introduced in the mid-70’s, or what capabilities they would possess decades later.

[ajafari1]

Nice! You were very close. With any exponential curve, it takes forever to get to point A, but once there, it's an immediate job to point Z.

[carapace]

Cheers!

I'm going to try to work up a decent reply to your article, but it might take me a day or two.

Have you read Wendell Berry's "What are People For?" (that essay not the whole eponymous book)?

- - - -

The (open) secret at the heart of AI is that no AI can answer: "What is good?" ( https://news.ycombinator.com/item?id=31720621 )

AKA "Lucky for whom?" Teela Brown, fictional character in Larry Niven novels

Spoiler alert!

This link gives details of the character that are spoilers: https://larryniven.fandom.com/wiki/Teela_Brown

Unfortunately those details are the point I'm trying to make, and I'm a huge Niven fan so I'm not going to spoil Ringworld, but it ties in with the (ultimately metaphysical) question, "What is good?"

[krageon]

The average is so bad it barely even has to produce code that works, so that's not really a bar I'm interested in seeing cleared.

[mr_toad]

AI vs average programmer, maybe 10 years.

Commercially viable AI vs skilled engineers, that’s going to take longer.

I wouldn’t be surprised if we could replace the average BA with an AI right now.