WAN Show AI vs Machine Learning

[Thomas A. Fine]

While I generally agree with what Linus was getting at on the WAN show about how overblown "AI" is, Linus' attempts to define newer better terms is inaccurate and inadequate.

 

"An Artificial Intelligence should be able to reason" - this shows that even Linus doesn't really understand how overblown "AI" is as a term, because nobody has invented an AI that can reason.  At all.  Not even close.

And what Linus describes as "machine learning" is ... simultaneously too broad and too narrow.  The vast majority of algorithms that are considered AI algorithms use the method he describes in one form or another, under a long list of names (genetic algorithms, evolutionary computation, stochastic gradient descent, etc.).  Even a technique like neural networks is trained by a some sort of random thrashing about like that. There are however other approaches also used in machine learning that are not this thrashing about method.

The term "AI", as used by professionals, is a (ridiculously) large umbrella term, and they absolutely use "AI" to cover what Linus says is "machine learning".

 

While using "machine learning" instead of "AI" may be a (very small) step in the right direction, it isn't consistent with how professionals use the terms (unfortunately).  And machine learning itself still comes across as a loaded term, in which they imagine that some sort of sentient machine is learning things.

 

Again though, I agree it's frustrating.  Especially since the field of AI really hasn't evolved very much at all in the decades since I got my software engineering degree (except in the field of hardware acceleration, making applications that were once impossible now relatively easy).  But somehow in the past decade, a lot of people are convinced that thinking machines are right around the corner.

[Sarra]

Machine Undumbing.

 

Just call it 'rapid maths'. That's what it ends up being, yes?

[Holmes108]

Yeah, I was thinking about this topic too. I'm not particularly knowledgeable about the subject at all, but based on my limited understanding, if we wanted to use the term AI completely properly, we'd basically never be able to use it at all, outside of future potential/sci fi scenarios.

 

I generally think it's okay for language to evolve, and whatnot. But to do so in this case leaves the issue that if/when we start getting into the real thing, like general AI, what term are we going to use then?