As a practitioner in this field, I kind of agree with the author. Historically “intelligence” was synonymous with “cognition” as a necessary ingredient for the dictionary definition of the word: “the ability to acquire and apply knowledge and skills”. To call the ML techniques we have right now “intelligence” requires one to stretch the truth pretty significantly.
Historically, all kinds of words meant all kinds of different things than what they do now. A word means what people think it means. It is a symbol used to communicate an idea. We can wish that prescriptivist interpretations applied, but one thing we can’t seem to do is make that the reality.
Normally I'd agree with this sentiment: language changes constantly, usage is king. However we're in an era where people (often in the course of marketing themselves or a product) intentionally move the goalposts of word meanings in a way that's generally unhelpful to users of that language. So we describe there is something called X that sounds very impressive and powerful. And there is something else kind of like the existing idea we called X, but it's more mundane ... it's close enough that salespeople can get away with calling it that, so they do, and the definition of X is now hopelessly muddled. I don't know what to do about it, but this does not strike me as being the same as the natural evolution of word meanings over time. I guess really it is still one form of that. But maybe it's one that as a society we should be more careful of - you can't, for example, call one medication by the name of another for marketing reasons.
Maybe AI is not the worst example. But with most buzzwords that are not legally protected, they just get marketed into meaninglessness in a way that harms non-expert consumers.
Nevertheless, I maintain that you can’t, in all honesty, call the ML techniques we employ today “intelligence”. They are little more than function approximation. Qualitatively this is a technique we had for decades, it’s just that now we can approximate functions that are far more complex.
A qualitatively different thing that I’d have no problem calling “intelligence” would also supplement perceptual and transducer stuff we have now with higher level cognition and knowledge representation. But at the moment all of that is so far out of scope that the unit of bogosity is named after the last guy who tried to do something about those higher level issues (Doug Lenat) and everybody else is afraid to touch them with a 10 foot pole.
I'd tend to agree with "AI is not Intelligence", my problem with the article is that it seems to say "what's called AI today is not Artificial Intelligence", and IMHO the term "Artificial intelligence" has always included system that only do some functions or the mere appearance of intelligence, which is why distinguished cases like "strong AI" exist. I guess the intended audience for the article might not be aware of that as such, but then the article IMHO should cover that appropriately.
As another practitioner, I wholeheartedly agree - most of what is called AI today is ML, and thats basically pattern recognition via boundary relaxation solutions - it is necessary for "AI" but not sufficient - BUT - my problem with the authors position is that he keeps not noticing A stands for artificial - we've had other solutions in other techs, Mycin and Dendral come to mind - no, they weren't AI either, however I think when future work begins to incorporate more than the PDP groups seed stock, we may see AI's begin to appear - and then the arguments will start all over with 'too narrow to be intelligent' - first we'll have to figure out 'it looks like cognition' before we move on to anything remotely resembling actual cognition - at present, not only can we not do it, we have never ending arguments about its very definition.
I'm not sure that's true. I remember reading Alex Martelli (famous in the Python world) write about how people now consider methods based on Bayes Theorem (ie, statistical learning) to be part of AI. His view in 2003 was:
> considering Bayes' theorem to be part
of AI makes just about as much sense as considering addition in the
same light, if "expert systems" had been the first context in which
you had ever seen numbers being summed. In the '80s, when at IBM
Research we developed the first large-vocabulary real-time dictation
taking systems, I remember continuous attacks coming from the Artificial
Intelligentsia due to the fact that we were using NO "AI" techniques --
rather, stuff named after Bayes, Markov and Viterbi, all dead white
mathematicians (it sure didn't help that our languages were PL/I, Rexx,
Fortran, and the like -- no, particularly, that our system _worked_,
the most unforgivable of sins:-). I recall T-shirts boldly emblazoned
with "P(A|B) = P(B|A) P(A) / P(B)" worn at computational linguistics
conferences as a deliberately inflammatory gesture, too:-).
So we have an algorithm which wasn't considered part of AI by the then-AI practitioners, because it didn't aim for semantic understanding, become part of an expanded definition of AI.
This article is seriously undervoted. I know it's against the interest of several stakeholders here, but admitting that what we're working on now has little to do with AI could actually help us to direct our efforts in workings towards true AI, if it's possible to reach. Instead, we call simple classification "AI", even though it has much more to do with statistics. Let's just leave it at "machine learning" and stop talking about AI as something that already exists, especially in your products.
17 comments
[ 2.6 ms ] story [ 63.2 ms ] threadMaybe AI is not the worst example. But with most buzzwords that are not legally protected, they just get marketed into meaninglessness in a way that harms non-expert consumers.
A qualitatively different thing that I’d have no problem calling “intelligence” would also supplement perceptual and transducer stuff we have now with higher level cognition and knowledge representation. But at the moment all of that is so far out of scope that the unit of bogosity is named after the last guy who tried to do something about those higher level issues (Doug Lenat) and everybody else is afraid to touch them with a 10 foot pole.
https://news.ycombinator.com/item?id=14178405
> considering Bayes' theorem to be part of AI makes just about as much sense as considering addition in the same light, if "expert systems" had been the first context in which you had ever seen numbers being summed. In the '80s, when at IBM Research we developed the first large-vocabulary real-time dictation taking systems, I remember continuous attacks coming from the Artificial Intelligentsia due to the fact that we were using NO "AI" techniques -- rather, stuff named after Bayes, Markov and Viterbi, all dead white mathematicians (it sure didn't help that our languages were PL/I, Rexx, Fortran, and the like -- no, particularly, that our system _worked_, the most unforgivable of sins:-). I recall T-shirts boldly emblazoned with "P(A|B) = P(B|A) P(A) / P(B)" worn at computational linguistics conferences as a deliberately inflammatory gesture, too:-).
So we have an algorithm which wasn't considered part of AI by the then-AI practitioners, because it didn't aim for semantic understanding, become part of an expanded definition of AI.
Same problem with the term "crypto" which now apparently means anything from Encryption to Digital money.
Its super annoying.