In contrast, ML methods do work, observably and clearly, and they work in a ridiculously general way, to a degree larger than almost anyone thought (or even thinks) is reasonable for them to.
And it's not just my opinion; there's a reason AI conference attendance has shot up a factor of 10 or so in the last few years, why NeurIPS is the leading one (and even historically GOFAI conferences are majority NNs), why the big AI labs with big AI cash are all doing NNs, and why all of a sudden AI is a popular topic outside academia.
If this doesn't answer your question, perhaps answer the opposite; how do you know that it's wrong?
I know the literature. It's my job.
>> And it's not just my opinion; there's a reason AI conference attendance has shot up a factor of 10 or so in the last few years, why NeurIPS is the leading one (and even historically GOFAI conferences are majority NNs), why the big AI labs with big AI cash are all doing NNs, and why all of a sudden AI is a popular topic outside academia.
That's still an opinion- "it's not just my opinion, everyone says so". A.k.a. "It is known", in Dothraki. And of course it is of no consequence who's spending money on what and who's submitting papers where. The volume of research was never a criterion for its quality. Heed thee well the legend of our Lord Geoff Hinton's years in the academic wilderness and how he emerged victorious with the laws of deep learning in his hands.
I think what you've said so far has convinced me you're expressing a personal opinion that is strongly held without a good reason to do so. You make sweeping statements with great certainty, but you don't really seem to know how you know the things you know, so you end up "knowing" some things that you don't really know. For instance, you claimed that "GOFAI" successes just "aren't there" but I listed a few, like Deep Blue or MYCIN - and you didn't seem to have heard of these before (I'm more surprised about not knowning of Deep Blue than MYCIN).
You also claim that "these approaches are not AI". That's a "No True Scottsman" right there. Except there really is no True Scottsman (i.e. "AI" in the sense you use it)- ask Yoshua Bengio:
Bengio: In terms of how much progress we’ve made in this work over the last two decades: I don’t think we’re anywhere close today to the level of intelligence of a two-year-old child. But maybe we have algorithms that are equivalent to lower animals, for perception. And we’re gradually climbing this ladder in terms of tools that allow an entity to explore its environment.
Spectrum: Will any of these ideas be used in the real world anytime soon?
Bengio: No. This is all very basic research using toy problems. That’s fine, that’s where we’re at. We can debug these ideas, move on to new hypotheses. This is not ready for industry tomorrow morning.
https://spectrum.ieee.org/tech-talk/artificial-intelligence/...
Or, you know, ask any AI researcher :)
Edit: Which "GOFAI" conferences are majority NNs? What period are we talking about?
At this point I think we're just hopelessly talking past each other. Of course I know about Deep Blue. I didn't know about MYCIN, but, like, “MYCIN was never actually used in practice”, so I don't feel particularly bad about missing that one.
But neither of those challenge my point. If you want to go back in time 30 years, then sure, if you want to be an AI expert, then you have to know GOFAI. That's what the ‘OF’ stands for.
> I know the literature. It's my job.
Yah I read the literature too. (Albeit it seems a very different subset.) That's not an argument though.
> Or, you know, ask any AI researcher :)
OpenAI is explicitly about the path to AGI, https://openai.com/about/.
DeepMind was also founded to tackle AGI (no source, sorry).
Geoffrey Hinton thinks NNs will get to AGI https://www.technologyreview.com/2020/11/03/1011616/ai-godfa....
Even in your own link, Yoshua Bengio is saying that this is a path to AGI, it's just not there yet.
> Which "GOFAI" conferences are majority NNs?
I said “historically GOFAI conferences”, so eg. AAAI.
I too can quote Hinton -from memory and without a link. I remember him saying that the next big thing in AI will come from a grad student who distrusts everything he (Hinton) has ever said. Unfortunately, I won't be that grad student- I haven't heard everything that Hinton has ever said.
---
Like, my point is not about whether you can find the odd person trying to solve intelligence with grammars, or what were GOFAI conferences still harbour GOFAI research in the corners, my point is that a) these approaches don't work as a way to actually tackle AI, the problem, b) the vast majority of the field does not take them as seriously as a method of doing so, regardless of other uses, and c) therefore it's natural, not ‘impossible’, to gain AI expertise without having much care for those parts of the field.
>> Yes, but there's a reason I suffixed that comment with “(again, in an AI context)”. GOFAI is great if you ignore the last two letters of the name, and how it failed almost all its major promises.
>> These used to be considered AI because it was thought that you could build a useful reasoning agent out of a combination of these techniques, given appropriate developments. Now (almost) nobody does that; Google Map's pathfinding is just a pathfinder, not a general reasoner.
I keep asking- how do you know these things to be true? Are they just your opinion?
Can you please satisfy my curiousity on this?