Slashdot Mirror
AI Experts In High Demand
An anonymous reader writes: The field of artificial intelligence is getting hotter by the moment as Google, Facebook, Amazon, Microsoft and other tech companies snap up experts and pour funding into university research. Commercial uses for AI are still limited. Predictive text and Siri, the iPhone's voice-recognition feature, are early manifestations. But AI's potential has exploded as the cost of computing power drops and as the ability to collect and process data soars. Big tech companies like Facebook and Google now vacuum up the huge amount of data that needs to be processed to help machines make "intelligent" decisions. The relationship between tech giants and academia can be difficult to navigate. Some faculty members complain tech companies aren't doing enough in the many collaborative efforts now under way. One big gripe: Companies aren't willing to share the vast data they are able to collect.
Getting a job with AI is still limited. Companies don't trust it. Spooky sounding tech scares managers and business decision makers. Better off calling it a statistic driven predictor
That's like throwing money away!
“He’s not deformed, he’s just drunk!”
... the companies are actually willing to look beyond H1B visa holders and low wages. If they're not yet ready to pay what a real expert costs, they're not really in high demand.
It's a series of complex rules with some pattern recognition, it's not fucking AI.
*explains how cognition in human brain works*
to help machines make "intelligent" decisions
IMHO, AI is more than just a way to profit off of individual consumers' weaknesses.
I, and probably a lot of other people out there, would be very interested in knowing how to create a 'vacuum' program that could know it all. This power would be abused within months after creation, but at least it would be in the hands of the people, not in some giant's arms, so abuse wouldn't span too far.
And what do you think you are? The more I learn about machine learning, the more impressed I am with natural neural networks and the incredible sophistication of the layered methods which are being applied to achieve complex behaviors.
Also:
http://www.artificial-intelligence.com/comic/7
Tech people don't understand it either. Just look at the article in Slashdot the other day about guessing age. It gets it wrong some times. Wow. That's kind of the point. Cars driven by an AI will crash sometimes too. And financial AI systems will make the same kind of mistakes humans make, in addition to the ordinary bugs. The technology has its place. But it isn't something that magically does the right thing.
It's a series of complex rules with some pattern recognition
That is also a pretty good description of what a brain does.
And what do you think you are? The more I learn about machine learning, the more impressed I am with natural neural networks and the incredible sophistication of the layered methods which are being applied to achieve complex behaviors.
Exactly.
Besides, the first entity that gets a functioning superhuman brain, if it has enough lead time over its competitors and is able to gain enough data and indoctrinate goals, is very likely to win. This applies in every field of human endeavor. Think of the enormous transaction and synchronization costs we have to deal with in the incredibly inefficient networking between your brain and mine, for example. Now network a hundred of those brains together with those costs a few million times smaller...
It won't happen all at once like that until it does, of course.
It's a series of complex rules with some pattern recognition
That is also a pretty good description of what a brain does.
That's a pretty-good description of what an *adult* brain does, but it's not a good description of intelligence - artificial or otherwise. Your adult brain learned the rules from its environment with no assumptions about what those rules were.
Try writing an algorithm that can learn to play either chess or checkers, depending on what game it sees.
Make that same algorithm be able to play asteroids, or drive a car, or OCR.
Make that same algorithm be able to recognize a tune ("row row row your boat") even if it's played in a different key, at a different speed, with variations in tempo, and even variations in key.
Any time you know beforehand what the rules are you are not simulating intelligence - you are simulating the *results* of intelligence. You are just writing down whatever it is that the intelligence in your head has decided.
The intelligence never makes it into the program - it stays in your head.
It only has to be slightly less stupid than typical humans, and/or cost less than humans. It may also need more trace-ability, such as knowing why it gave an answer it did. With humans you can ask and usually get an answer such as "we always did it this way", "that way usually works for me", or "because the alternative confuses the sales team", etc.
But career-wise AI has had multiple boom/bust cycles as the usual hype-masters overdo claims and damage AI's cred. Have a Plan B if you go into AI. (No, not a Plan 9.)
Table-ized A.I.
The term "AI" dropped out of favor a decade ago as a result of a lot of over-promising and under-delivering from the decade before that. Remember "expert systems"? Yeah, that was "AI" in a different guise. It looks like the term "AI" is making a bit of a comeback. I'm not sure that's a good thing, because it never really describes these systems adequately, as "intelligence" has very little to do with it.
Irony: Agile development has too much intertia to be abandoned now.
>> AI Experts In High Demand
What they really mean is business data-mining experts. Unfortunately AI expert just sounds cooler and is easier to say, no matter how far from the truth it is.
All those 80s era AI LISPers are rejoicing.
If you were me, you'd be good lookin'. - six string samurai
Anyone interested in a good theory on how human intelligence actually works should read Jeff Hawkin's (the guy who invented PalmPilot) On Intelligence. It proposes and describes a rather interesting theory on how our perceptions, reactions, and intelligence all work.
Irony: Agile development has too much intertia to be abandoned now.
Comment removed based on user account deletion
= ??? What?
Yes, there is some evidence of abstraction layers in biological neural networks/compartmentalization of functionality.
You have a number of degrees of freedom associated with your leg. Your brain does the inverse kinematics and motor control functions in the cerebellum to handle the lower level tasks involved with translating those desired outcomes in to muscle controls. Meanwhile your visual cortex is closing the loop between the feedback from the muscles and nerves and the visual feedback zeroing in on the completion of the task.
Muscle memory is the process of taking this cognitive load and converting it to what is essentially a look up table of recorded motions. The more times the motion has been trained, the more extensive and robust the look up tables are able to reference with less and less higher level processing. "If this then this" combinatoric decision trees where the process of repetition trains away error and improves the approximations until fluid motion can be achieved at high speed with almost zero overhead.
Deep learning is exciting because it scales well. We can get computers to benefit from the same brute force approach without an army of interns doing laborious preprocessing of data. Unsupervised learning = capital investments in processing hardware and Mobile App Startups and the meaning can be extracted from the data without the assistance of experts.
Imagine a blackjack dealer who doesn't have to hit on 15 and is allowed to count cards. With enough video footage of professional poker players, you may someday be able to train a machine to call Doyle Brunson's bluffs.
What does that mean for retail investors on Wall St.?
Quite the opposite, strong AI was the previous boom/bust.
This time we're using "sheer brute force" to train NNs in a bid to get emergent intelligence-like patterns, instead of using formal logic constraint solvers.
People are not tabla rasa. Evolution has baked in all kinds of assumptions.
Absolutely true. Kids learn to not touch the stove after burning themselves once. They were born with some kind of sense that pain is bad and to be avoided.
I think that if you could figure out the motivation bit, that would be half the battle.
And people aren't universally good at the motivation bit either. They buy lottery tickets, smoke, fail to estimate risk in a sane manner, and do all kinds of dumb stuff where the instincts they were born with basically reduce their chances of having surviving progeny.
The trend is real. Its not called AI, but "deep learning". Basically Google and Facebook bought few datamining companies and ball got rolling.
Google torch7 for specific stuff.
Many modern AI methods take advantage of unsupervised learning. Not only do they not need to know what the rules are, they don't even need to know the right answer most of the time. There are successful demonstrations of such algorithms learning to play Nintendo from watching people play, and Google's deep learning network learning the concept of "cat" from watching YouTube videos. I also remember a paper looking at recognizing melodies played in different keys.
Your knowledge of AI is a couple of decades out of date.
Uh, the term 'expert systems' predates 'AI'.
I'm not sure where you got this information, but as someone that studied Artificial Intelligence in college (in the 80's), I can assure you that Artificial Intelligence has been around far longer than Expert Systems.
Expert Systems are usually considered one of the first successful forms of AI, becoming of practical use in the 70's, and then proliferating in the 80's.
But Artificial Intelligence as a field has been around since the 1950's.
Here's some links about both of these terms:
http://en.wikipedia.org/wiki/Expert_system
http://en.wikipedia.org/wiki/Artificial_intelligence
Uh, the term 'expert systems' predates 'AI'.
Nope. The term "artificial intelligence" was coined in 1955 by John McCarthy. Edward Feigenbaum is considered to be the father of expert systems, and first published a paper about them in 1977.
Irony: Agile development has too much intertia to be abandoned now.
If you show a very young child (less than a year old, I think) something 'impossible' happening, they will pay attention to it for longer and find it more interesting. So if you hold a ball in the air and let go, but it doesn't fall, or you throw a ball and it goes through a wall, a baby can recognise that those are weird events, and will stare at them for a long time.
If you then give the baby a choice of toys, amongst which is the ball that did an impossible thing, they will spend more time playing with it, rather than equally spreading their attention around. Moreover, they will conduct small experiments that are related to the impossible thing they saw. They will pick up the ball and drop it repeatedly to make sure gravity works. They will hold the ball and bang it on a surface to make sure that the ball does not arbitrarily pass through things.
The brain has a lot of stuff built into it. There are whole sections of the brain devoted to image processing, or understanding smells and taste. These are not inconsequential starting points.
I find it best to just call everything I do Ai.
that way all the idea thieves have no idea how I do what I do.
Just remember that iPhone's voice recognition comes from Nuance, not Apple, and it's been developed over several decades.
http://en.wikipedia.org/wiki/N...
Awwww, man! I should have said 'yes' to that Mizzou grad school acceptance and, after ten patient years, pounce and corner it all!
Your adult brain learned the rules from its environment with no assumptions about what those rules were.
So very, very wrong. Well over 90% of what the brain does, even consciously, is either instinctive or a behaviour learned almost entirely from instinctive triggers.
Study someone with autism to see what the brain has to deal with when even a few of the built-in assumptions are missing.
Oh? And how do you have knowledge of how that works? Because, you know, the scientific world does not at this time.
Statistics cannot lead to intelligence. If you knew anything about either you would not claim such nonsense.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
People want to believe that nonsense because people are, on average, deeply stupid. And no, it is not AI at all. The human race is not capable at this time to produce any AI that deserves that name.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
Ah, a physicalist quasi-religious fanatic. Nice. Your delusion does not cause the world to work like you believe it works.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
At this time it is still completely unknown what the brain does to manifest intelligence. Have a look at some research before claiming utter nonsense.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
Excellent explanation. The brain certainly has some automated components that are not intelligent and they are configured by actual intelligence as you describe. What that actual intelligence is, how it works, and why it is only observable together with consciousness is completely unknown at this time.
Unfortunately, this explanation will fly right over the heads of most people here.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
The formal logic approach is still the only one that has a theoretical possibility of creating some aspects of true intelligence. Unfortunately, in this universe, it fails even at small problems due to limited computing power available. The brute-force approach is good enough to simulate very limited intelligence when an incredible lot of data is available. That has nothing to do with intelligence though, and statistical approaches cannot simulate more complex chains of reasoning. They are always shallow. On the commercial side, even shallow models can work, but calling them AI is just misleading.
The sad fact of the matter is that most people are morons and hence even a shallow, statistics driven system can now be a moron that is just a bit smarter than them (or rather knows more). Hence for many mundane tasks, humans are becoming redundant.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
Only problem is that buying companies that already do it does not create any jobs.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
Obviously... if it's AI, and it has some real world utility, then we rename it so that it's not AI any more. Then we complain loudly that "AI has never produced anything useful".
Well I'm glad we have you here to arbitrarily define intelligence. It's about time, the human race has been struggling to define it precisely for hundreds of years.
Where have you been all this time of self-declared definer of terms?
But just to clarify, basically, what you're saying, is that intelligence is a form of magic that we have inside us? Where inside us does this magic exist? Where does it come from? Are you saying it's a special undetectable thing? Your comment seems to imply it disappears in adults, but we can't detect it in babies either.
We can't do any of the things you suggest because we don't have computers even remotely as powerful and capable of rapid complex processing as the human brain. I also can't teach a dog to do any of the things you suggest, so are you declaring dogs as being unintelligent? Your examples seem to suggest that intelligence is unique to human beings and nothing else has the capacity for intelligence.
I doubt you'll ever see a job posting for an AI position. All products have a high failure rate of reaching the market, and I imagine the AI product failure rate is astronomical compared even to "regular" products. Large companies tend to buy out small companies that have developed slightly successful products, and they make money by enhancing and marketing them. Very rarely do large companies like Google, MS, etc actually put money into research developing new things. So, I think if you want a job in AI, you're going to have to invent something yourself.
Sort of. We actually do have fairly robust theorem solvers written in prolog, but thats not enough. Intuitively, "true AI" works like extracting formal logic theorems out of huge set of before/after data fed to a blackbox.
Just like humans do something intuitively at first, with some degree of success, but when they find rational backgrounds (with help of formal logic rigor) behind that intuition, it gives significant accuracy boost. The two work in tandem - formal rigor is toothless when facing the totally unknown, but can explain it after intuitive models are trained and it can feed its hypotheses into them in lieu of farmed data.
Trouble is that layered NN camp ("intuitive") and formal logic camps are still too separated. But corporate interests will force merger to a degree.
This is most visible in speech recognition, and more recently, vision where formal grammar models sit above low level intuition NN, or better said, directs training of layers so it can work with less data and reason about unknown inputs because it actually "understands" what's going on on a formal level.
Sadly that's absolutely true. I know because I have been developing a Strong AI project for over 20 years. Strong AI is one area where I would definitely NOT bet on university research, and am pretty sure it will only ever be an outsider like me who succeeds..
Below the speed of light Special Relativity is one of the most accurate theories in physics - above the speed of light..
Not sure what they were naming it but AI research dates back at least to the 1930's - for instance in the work of Alan Turing. And a lot earlier if you count the logicians. It was only as the field developed that the search for mechanical intelligence became separated from the field of general computing...
Below the speed of light Special Relativity is one of the most accurate theories in physics - above the speed of light..
Not entirely nothing. Computers themselves emerged at least partly from AI research. The real problem with the field is that for decades computers simply didn't have the power needed for Strong AI, even today they only barely do and still hit several near vertical walls - especially with reliability. I'm working on a Strong AI project and the heart of this work today is developing a custom hardware & software platform..
BTW : Nuclear fusion - working reactors do now exist, and they are now getting close to break even, and for power gen the technology is now really 20 or 30 years away. The real crime with fusion is lack of funding, poor planning, and inefficient organisation - otherwise we could already have commercial fusion today..
Below the speed of light Special Relativity is one of the most accurate theories in physics - above the speed of light..