Way to go wired, an article with a bunch of references to things that are definitely not AI, would not have been considered AI (even in the dark gray past) and that are not considered AI today even by those that would like to believe that an 'AI revolution' is on.
According to this article AI is just another way of saying 'the leading edge in real world interaction', from inventory control to (apparently) crappy search results that still need further interpretation. (google 'what is the color of grass' for a nice example of AI in action, any three year old would tell you 'green' is the answer, not a bunch of links).
The 'AI winter' was not brought on by a lack of progress or interesting results, it was mostly brought on by it being used as buzz word hyping the notion of AI long before even baby steps had been made in the field, and this article fits right in with the over-selling of AI. An ABS system is AI? Please.
As of now we do not know if there will ever be something that we would call an AI but just like with pornography, I'll know it when I see it, and an inventory control system, no matter how clever it appears is not it, especially not if it took a bunch of programmers to write.
I think the one thing that will set a true AI off against the background of wanna-be AI systems is that a true AI can be taught by non programmers to reason about the world around it and draw meaningful conclusions without external input. We are still - in my opinion at least - very far away from anything that comes close to that.
Maybe their example sucked but Search engines fall into AI in my book.
I think the author is emphasizing the complexity behind very simple things. Obviously, it's not intelligent but it's doing way more than expected.
To me "intelligence" in software is about hiding complexity, making decisions and deducing things based on data, predicting and adapting based on input to deliver unforeseen results using a simple UI (where UI could be keywords in a search engine or a warehouse reorganizing itself to deliver and locate stuff faster based on new input/output).
The next generation of software is not about building actual intelligent systems. It's about accomplishing specific tasks and delivering more than expected.
Think of your IDE fixing your most common typos because it's constantly analyzing your input, or a file selection program that saves patterns/dependencies between your opened files and tries to guess which file you mean when you hit `gt` etc.
To me, this is smart behavior or _artificial_ intelligence. This is as far as software will ever get unless you start mixing it with biological stuff.
Diapers.com: Then, to fill an order, the first available robot simply finds the closest requested item. < That's AI.
Google: "what is the color of grass" is equated to "define:color of grass"
Web definitions for color of grass
a primary color between yellow and blue in the spectrum, like the color of grass
encarta.msn.com/dictionary_/green.html - Definition in context
So yea it's stupid AI, but they are both still AI.
PS: Decoding a natural language queries structure is considered AI. "1600 Pennsylvania Avenue" to a map location is more than just a search for web pages.
There is a joke that goes more or less that any sufficiently advanced AI is indistinguishable from 'just' statistics. But statistics (the unreasonable effectiveness of data) and the philosophy of probability are very profound topics that delve into the very fabric and nature of the universe. The same things keep appearing everywhere too. Bayesian networks and quantum mechanics. Entropy. Monte Carlo. So many.
I have a personal theory that we learn physical actions from statistical models more than modelling physical variables. Otherwise training how to do something as easy as a back flip (from a physics perspective) or flare would not be so hard.
Another problem is that nobody knows what the I part of AI is. So from the get go saying what AI is and isn't, is a failed experiment. I don't know what I is but i bet there are times you will know it when you see it and i won't. and vice versa. I also contend that modern systems do learn in a very basic sense. But I suspect we will also disagree on what learning is.
Yeah, I think this has to do more with systems and who controls them, than with AI per se. As far as classic writing on human/machine interaction goes, it's less about the stuff that AI pioneers like John McCarthy and Alan Newell were writing, and more about the stuff that Ted Nelson was writing in Computer Lib, about the dangers of people's lives being significantly structured by computers they didn't understand and didn't know how to control.
"The computers are in control, and we just live in their world"
No, we are ostensibly in control, or at least the humans who know how to program and control this type of system are. Like any machine or complex system, it may fly out of our control, but it is not in some other entity's control at that point, it's just out of control. We're not to the point of facing a true artificial intelligence, yet...
Agreed. The mystification of what computers do is a worrying trend.
For now, at least, computers can't do anything at all interesting or remarkable. They can, however, do boring things really really fast. And lots of boring things, done really really fast, can yield some pretty impressive results.
But it doesn't change the fact that what computers actually do is utterly trivial. Anything interesting they achieve is the result of algorithms programmed by very intelligent and sophisticated human beings.
That's a clear, qualitative difference from any real AI. When computers are creative, when they start doing things that haven't been done before, without the direct programming of any human, then a computer can truly be said to be "in control."
The mystification of what computers do is a worrying trend.
Trend? When have the masses ever had a clear grasp on computation? Quoth Babbage:
On two occasions I have been asked,—"Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?" In one case a member of the Upper, and in the other a member of the Lower, House put this question. I am not able rightly to apprehend the kind of confusion of ideas that could provoke such a question.
I agree mystification of computation is not something we should encourage, but I don't think that's really what is happening in this article. The things that computers do today are not utterly trivial, the algorithms are getting more complex every day and there is some evidence that some of the very complex probabilistic structures being built have corollaries in the human brain.
We have already passed the point where all computers are being directly programmed by a human being to perform an exact task. Things like google, and Wolfram Alpha and Netflix are following instructions designed by humans that describe complex behaviors and even reasoning, not the particular outcomes they yield. These instructions are far beyond listing steps for a robot to pick up a cup off a table.
Are the machines being creative? No. But, how would we know if they were, we don't even know what that means.
Are they 'in control'? No. No more than your bicycle is in control when you stop pedaling and take your hands off the handle bars. It has no will, it just keep going, just like google's servers. But, do we know enough about what intelligence (or sentience, which is what everyone is actually talking about here) is in order to answer that question if the answer ever changes?
How complex does something have to be in order for us to consider it 'intelligent' and what kind of complexity will it need to have?
> When computers are creative, when they start doing things that haven't been done before, without the direct programming of any human, then a computer can truly be said to be "in control."
It not under the control of a conscious entity which can flexibly formulate new goals, adapt to new tasks, etc. But under a loose definition you could say that quite a few large systems are under "an entity's control" in the sense that they have an internal set of procedures they're carrying out, criteria they use to make decisions, etc. It's a rather hardwired and unchanging entity, but in large (and especially old and incrementally grown) systems, what exactly it's doing and why might actually be somewhat opaque to the human operators.
It's basically a philosophical question what you want to call that, but I'm okay with at least metaphorically calling that a situation where "the machine is in control", because in a sense it "knows" what it's doing and carries it out, whereas the human operators have only a vague sense of what it's doing, and sometimes have to engage in software archaeology to figure it out and change anything.
(But yes, I wouldn't say that an "AI" is in charge of anything.)
A.I. is probably one of the most exciting concepts I know of, yet if you think about it the somewhat disappointing utilitarian "reality" of A.I. is what makes sense. For example why are no other species as intelligent as humans? Are they just slow evolvers, or suck at evolving? Are humans just blessed evolvers to achieve intelligence before everybody else? No, survival is a purely utilitarian exercise. You find a stable niche and fill it, or die. In other words humans and their brains have just stumbled upon a particular pattern of living (what with language and societies and "intelligence") that is currently "stable". What if in 500 years all humans are extinct, yet bacteria still happily do their thing. Would we still consider humans so 'intelligent'? The same rule applies to processes embodied in silicon as to processes embodied in carbon. And for as intelligent as we humans think we are, we are very bad at predicting where the next stable niche will be (otherwise we'd all be millionaire entrepreneurs). This is why when "A.I." finds a stable niche in areas like search engines and warehouse management, we are both surprised and a bit disappointed.
I disagree with your analogy; branches of math/stats/CS are not subject to natural selection. They have quite the pantheon of "designers" (academia) deciding for what to use them and to what to apply them.
AI is the sexiest, most romanticized and misunderstood fields in computer science. Certainly, it never ceases to generate boring article topics for Wired.
I read and enjoyed the article. Was not boring to me. I am extremely interested in machine learing and get excited about things like graphical models, MCMC, hybrid monte carlo, online learning, reinforcement learning, deep learners, sparse learners, co-evolution, ensembles etc.
I found this article interesting and actually very well informed as such things go. It does not focus on Human like AI but practical approaches currently being used that are able to do the most basic type of learning at a complex level - ability to infer albeit on restricted spaces. For example, its 9 AM someone is at the door, given my past experience on situations like this (data) who is it most likely to be? I have a best guess based on what I have learned about who tends to come knocking when.
They can also pick out basic patterns whose details are very intricate, sometimes nuanced. What is interesting is that in these small spaces they can account for and infer on and recommend basic actions on vast number of parameters so accurately that they blow away any human attempt to match them to smitherings. This is why they are used as they are. Creativity, abstraction, self aware introspection,emotions, exploration for exploration's sake etc. are nice to think of as AI but that would be a waste of the type of niche intelligence computers are particularly well suited to based on their architecture. They do need to handle nuanced ambiguity better though. As well as transfering learning and dealing with situations far outside their training. The logic overflow of the 1960s movies is quite apt sometimes.
I do believe though that when (if) AI arrives it will be sudden as a unifying framework, a Standard Model of ML if you will is created. But in the mean time there is a lot of very interesting things going on and the combined explotion of introspection in the field (of Statistical Inference/Learning), data and processing power means very interesting times are ahead.
If you're more into the algorithms side of AI, you should certainly read AI: A Modern Approach (http://www.amazon.com/Artificial-Intelligence-Modern-Approac...). It's a text book, don't get me wrong, but it clearly explains many relevant algorithms in AI today with accompanying pseudocode and theory. If you've got a CS background, it's a great reference/learning tool. I bought mine for a class, and won't be returning it!
"Today’s AI doesn’t try to re-create the brain. Instead, it uses machine learning, massive data sets, sophisticated sensors, and clever algorithms to master discrete tasks."
The second sentence is basically my model for how the human brain works.
I've seen a chat bot that's indistinguishable from a real human being. It reaches that goal by acting like a stupid and rude 12 year old kid who think he's cool and l33t.
hey man
wtf u talking about
lololol wtf man ur dumb
thats shit
whatever
You can't tell whether he's really an annoying 12 years old or whether he's a bot.
OTOH, this is wired we are talking about. Until the drones hunt me down and eliminate me with impossible, ruthless efficiency, I remain skeptical. And I'll die screaming "Yeah, but it was a SYSTEMSGUY who made you ACTUALLY work!"
According to this article AI is just another way of saying 'the leading edge in real world interaction', from inventory control to (apparently) crappy search results that still need further interpretation. (google 'what is the color of grass' for a nice example of AI in action, any three year old would tell you 'green' is the answer, not a bunch of links).
The 'AI winter' was not brought on by a lack of progress or interesting results, it was mostly brought on by it being used as buzz word hyping the notion of AI long before even baby steps had been made in the field, and this article fits right in with the over-selling of AI. An ABS system is AI? Please.
As of now we do not know if there will ever be something that we would call an AI but just like with pornography, I'll know it when I see it, and an inventory control system, no matter how clever it appears is not it, especially not if it took a bunch of programmers to write.
I think the one thing that will set a true AI off against the background of wanna-be AI systems is that a true AI can be taught by non programmers to reason about the world around it and draw meaningful conclusions without external input. We are still - in my opinion at least - very far away from anything that comes close to that.
Learning is the key to AI, not programming.