Ask HN: Is AI really a thing?

5 points by lbriner ↗ HN
Call me cynical but I see AI as the next hyped technology even though I am unaware that anything has changed in the last few years and that it seems to be nothing other than pattern matching or machine learning, which is software that is told how to react to inputs, just like normal software.

Am I wrong?

9 comments

[ 2.0 ms ] story [ 34.6 ms ] thread
No - and not really controversial to say so. But one can argue that the explosive growth in computerpower has made it more useful.
Skynet - probably not. Harder-to-debug, yet useful algorithms - that's already here. Then there's stuff in-between, and that is uncertain...I mean, I am typing this on a tiny supercomputer, possibly on the other side of the planet from you; I didn't even imagine this three decades ago.

TL;DR: wherever the future takes us, it will not be like we imagine it now.

Am I wrong?

If you insist that the only thing that merits the label "AI" is what the rest of us all "AGI", then yes. AI can do some pretty amazing things, but as the old saying goes "once it works, it isn't considered AI anymore". A lot of the stuff we have today would have been like black magic not very long ago. But we still don't have "human like AI" so you get pundits and critics saying "AI isn't real, AI is just pattern matching", etc.

unaware that anything has changed in the last few years

A computer program just beat a arguably the best human Go player in the world at Go. You don't see any significance in that?

which is software that is told how to react to inputs, just like normal software.

Great, so write me a program, without using any machine learning, that can recognize the MNIST images with an error rate of 0.21%. If you finish before you die of boredom, or the heat death of the universe, I'll be stunned.

Depends, is he allowed to have the same budget/resources?
Sounds similar to a classical "curse of the AI": when AI does anything well, it stops being AI.

Before AlphaGo: "Hah! Go requires strategy and deep thinking. A computer will never match the insight of masters!"

A few days(!) after AlphaGo beats Lee Sedol at the first game: "Ha! Of course it's just a game of counting everything really fast. Of course computers are good at it, this game is totally unfair!"

Watching the media and people flipping their stances in a matter of days was quite comical.

The technologies that currently exist are examples of ANI ("artificial narrow intelligence" or weak AI[1]). The way you are using the term AI is in line with what is known as AGI ("artificial general intelligence" or strong AI[2]). Strong AI doesn't currently exist and, as far as we can tell, is still a fair way away. Some people believe that strong AI isn't even possible.

So to answer your question directly, AI is a thing, but currently it is only the subset of AI called ANI. AGI is still over the horizon.

[1] https://en.wikipedia.org/wiki/Weak_AI

[2] https://en.wikipedia.org/wiki/Artificial_general_intelligenc...

Let's first see when autonomous vehicles come.

I understand what you mean, considering the vast amount of resources that A.I. requires.

Most software is just reacting to inputs, yes, and AI algorithms are no different there. But one should recognise the inherent reductionism in any such statements claiming that that's all AI is.

Even if/when we do have AGI, it will still, arguably, just be software (plus hardware) that is reacting to inputs — such a definition helps nobody in this instance.

What about these things:

- Speech recognition - Facial detection/recognition - Fingerprint recognition - Personal agents (Siri/Alexa/etc) - Self-driving cars - Recommendation engines - Route planning - Spam detection

All of these things have either improved significantly thanks to AI, or would not exist without AI (that is: many of these algorithms are in fact categorised as being AI). Of course, as others also point out: once it's been solved, most folk don't categorise it as AI any longer, it's merely "just another algorithm".

Furthermore, some manufacturers are now including dedicated hardware-acceleration for some of these algorithms, in mobile devices (e.g. Apple's A11 and A12).

Granted, yes, many recent advances are "nothing other than pattern matching or machine learning" but that's a very wide brush you are weilding. One day even AGI may be classed as "nothing other than machine learning".

No, we don't yet have AGI. But yes: AI is really a thing, and it's here now, hype or not.

[TBH, I'm uncertain if you're trolling or not, but hey]

I see HN doesn't do lists very well / at all? Hmm.