Incredibly, this was an even dumber post than I expected. For its length it's hard to pinpoint any wrong argument made, because AFAICT it's mostly just a semicoherent stream of unjustified assertions and associations, but it does sometimes try, as per
> Oh yeah, and the second problem with the AI doomsday theory is that it's ludicrous. AI is so smart it's gonna kill everyone by accident? Really really stupid superintelligence? That sounds like a contradiction.
Bill Gates must be feeling like a right idiot now, huh?
Anyway, there was one line I did find entertaining in its inanity (the rest, AFAICT, has zero value),
> Intelligence isn't a Platonic ideal that exists separately from humans, waiting to be discovered.
Very good article. AI is a branding exercise, and most people saying that we have general purpose intelligence are not experts in it. Call it for what it is, machine learning.
I now think of AI as basically deep learning based ML and nothing else. But it's not entirely right to say it's not "intelligent". What is "intelligent"? Is a mouse intelligent? It can't reason, speak, read, or write, yet I'd say it is intelligent because it can respond to the environment in a way which on average maximizes its survival. This is not that different to e.g. a RL system responding to its environment in a way that will maximize reward. Then why would a mouse be "intelligent", and RL system not "intelligent"?
I find it bizarre because deep learning is only one (arguably fairly small) subset of the entire field of AI. AI is a HUGE field, spanning an incredible myriad of techniques and approaches.
Deep Learning may be the one receiving the most "hype" at the moment, but there is - so far as I can tell - no sense in which anyone connected to AI considers "AI" and "deep learning" synonymous.
Which "other", non DL areas of AI are in use at the moment? Which haven't yet been completely blown out of the water by DL counterparts? NLU, translation, sentiment analysis, image classification/segmentation, object detection, instance segmentation, speech recognition, speech synthesis, sentence generation, NER, the list goes on and on. In all of these ANNs completely destroy previous approaches, it's not even close.
I am not saying that DL isn't very successful and very popular. I'm just saying that it isn't, all by itself, the entirety of the field of AI. I'm not arguing anything about popularity and I definitely don't have usage stats, since approximately nobody is going to publish exactly what stack they're using if they get a competitive advantage from it.
But from a research perspective, it's not hard to find papers on arXiv showing people continuing to research other aspects of AI.
It's also not hard to find papers from people suggesting that the DL train may be running out of steam in the near future. See, for example,
I guess if you're only looking at what the popular media report for the general public, that's sorta true. But dig any deeper at all, and that's way off the mark.
> Oh yeah, and the second problem with the AI doomsday theory is that it's ludicrous. AI is so smart it's gonna kill everyone by accident? Really really stupid superintelligence? That sounds like a contradiction.
Bill Gates must be feeling like a right idiot now, huh?
Anyway, there was one line I did find entertaining in its inanity (the rest, AFAICT, has zero value),
> Intelligence isn't a Platonic ideal that exists separately from humans, waiting to be discovered.