Tech Giants Are Paying Huge Salaries For Scarce AI Talent (santafenewmexican.com)

jmcbain writes: Machine learning and artificial intelligence skills are in hot demand right now, and it's driving up the already-high salaries in Silicon Valley. "Tech's biggest companies are placing huge bets on artificial intelligence (Warning: may be paywalled; alternative source)," reports the New York Times, and "typical AI specialists, including both Ph.D.s fresh out of school and people with less education and just a few years of experience, can be paid from $300,000 to $500,000 a year or more in salary and company stock." The New York Times notes there are several catalysts for rocketing salaries that all come down to supply and demand. There is competition among the giant companies (e.g. Google, Facebook, and Uber) as well as the automative companies wanting help with self-driving cars. However, the biggest issue is the supply: "Most of all, there is a shortage of talent, and the big companies are trying to land as much of it as they can. Solving tough A.I. problems is not like building the flavor-of-the-month smartphone app. In the entire world, fewer than 10,000 people have the skills necessary to tackle serious artificial intelligence research, according to Element AI, an independent lab in Montreal."

Comments

That is true of all specialities....

[140Mandak262Jamuna]

In the entire world, fewer than 10,000 people have the skills necessary to tackle serious artificial intelligence research, according to Element AI, an independent lab in Montreal

In the entire world, fewer than 1000 people have the skills necessary to do unstructured tetrahedral finite element mesh generation. It is possible there are fewer than 1000 people who have the skills necessary to understand what exactly we mesh makers do. And, Surprise! there is demand for fewer than 1000 people to write unstructured tetrahedral finite element mesh generation. And far fewer than 1000 people are needed to manage them.

I am glad the periodical bubbles that infect Wall Street and venture capitalists benefits PhDs once in a while. Most of the time it benefits hedge fund monkeys or stock market cheats or lottery winners with delusions of grandeur or plain sociopaths. Happy for my grad school classmates. Enjoy the windfall while lasts, Ramachandran\s, Yang\s, Hsu\s, Gupta\s, Parpia\s and Wickramasinghe\s.

Re:More AI hype

[Anonymous Coward]

You should at least try to understand what you are talking about. Deep learning, aka neural networks, are not "algorithms and programs". They are part of the machine learning branch of AI. The computer is not programmed but learns by itself. People in computer science have been trying to do that for as long as computers have been around but were never quite successful until about 2012. Deep learning excels at tasks which are too complicated for humans to write code such as detecting objects in picture and analyzing recordings of voices or translating text. This is revolutionary. Even the primitive neural net technology we currently have will transform many applications in the next few years, in that they perform much better than what humans used to code and they require just a handful of AI specialists to train instead of team of 100 programmers. If the technology continues to improve, it could take over just about every field: driving, medicine, law, manufacturing, etc. But the current technology has limitations and it's not clear how much it can progress further.

Computer programmer could be one of the first job to be made obsolete by deep learning. Programmers will have retrain themselves as teachers to neural nets instead.

Pure BS

[guruevi]

Modern AI software isn't that complicated and not nearly as expensive to get people in. Look at job offers: $150k for AI research scientists in NYC. $65k in more rural areas. That's not well paid by definition at all. Sure, a pure AI scientist gets paid $500k just like a top neuroscience scientist gets paid $500k or a top biology researcher, but the majority of companies do not want to do the theoretical development of AI, any regular programmer can wrap their heads around the existing literature and build something.

Here in my area, there are a number of employers looking for AI engineers/scientists. They pay about what I make as a non-AI IT sysadmin, which is given my experience on the higher scale but by no means exceptional.

What Google and co wants is a glut of people 4-6 years from now that are "trained" in AI from college. You put out a report like this, you get massive amounts of people applying for the schools that offer programs and 5 years from now you have an over-abundance of people driving down overall wages. You also get to hire a bunch of people on H1B because the "US doesn't have the skillz" and you end up with a bunch of programmers on H1B under the guise of AI development.

Deep nets

[DrYak]

From my investigation of the matter it looks to be some sort of multi-variate analysis in drag. Uninteresting. Basically you get guys sitting around twiddling knobs. Finding the right parameters which works for a little bit and then you start knob twiddling again to find the next ones.

Except for 1 key difference. With Deep-Neural-Nets, the knobs twiddle themselves alone.

DNN get inspiration of how some neural network work in the nature (e.g.: a column in the primary visual cortex of the brain) to design thing that you can throw at problems, and which will autonomously train themselves.

Some years back I wrote a day trading program for a friend. It dynamically changed its behavior depending on the market signals and the rules he gave it (stops, buys, shift to a different stock etc.) which he found useful. Now that was fun.

These older program require you to have precise criteria in advance.

That works perfectly well with clearly codified problem - the friend has a clear set of rules that need implementation.

That completely fails for more vague problem ("detect a face") - it would be possible in theory to design a set of rules that can detect a face - a Haar Cascade. But designing such set of rules is extremely complex and cumbersome. And each time you need something new ("detect if there's a bird"), you would need to repeat all the hard work to invent yet another set of rules.
At that point, better take an advice from how mother nature solved the problem (by using stacks of neural network in a columns) and simply throw a DNN (e.g: a Convolution Neural Net - a ConvNet) at the problem, and watch it self organize and come up with a solution to your problem.

It's the modern-day equivalent of training pigeons to peck a city images to steer a missile.

curvature as captive starch

[epine]

In the entire world, fewer than 1000 people have the skills necessary to do unstructured tetrahedral finite element mesh generation.

Nice rhetoric—factual statement masquerading as metaphor, for any reader dumb enough to go along for the ride.

The Evolution of the Flour Mill from Prehistoric Ages to Modern Times — 1905

Before the first actual grinding mill came into existence, grain was merely shelled or husked by pounding. This simple kind of a "first break" was effected by spreading the grain upon a slab or block of stone and beating it with a hand stone; a subsequent development of this rude apparatus being a hollow mortar and an improved hand stone. The original hand pounder was used on a flat block... Such relics are found throughout both hemispheres, having been used by all primitive nations throughout the world; but eventually they were universally discarded for more perfect apparatus, which really ground the grain into meal.

That's about the present state of machine learning, the hand-crafting of "features" playing the role of the recently discarded flat blocks.

Wheat is an incredible dietary resource, with the starch being light enough to transport over long distances, if only one can find a way to remove it (contrast potatoes, only ever transported downhill, if at all, until the invention of steam power). Once upon a time, all food was local, as, too, was starvation (fear the blight).

A better method to mill the world's vast stores of accumulated data is a big deal, even if we remain in the relatively crude era of water-powered stone grinding wheels.

Data is a bit like wheat, it doesn't give up its curvature easily. Too much applied force creates heat and destroys the end product. The applied force must have exactly the right ratio of compressive to shear stress, which only an expert miller can judge. Deep learning is nothing more than a slightly better mill than the one we had before, and it ranks right up there beside becoming slightly better at milling wheat.

The economic value of the curvature we can now hope to unlock is quite large. And probably there's a lot of curvature yet to find that remains inaccessible to current methodology.

Data is oil. Data is also wheat.

By way of contrast, unstructured tetrahedral finite element mesh generation shaves 5% of the metal mass off a milling apparatus that already worked just fine, being just one of ten thousand noisy specializations in the great roil of small improvements where a penny shaved is a penny earned.

In the entire world, fewer than 1000 people have the skills necessary to do unstructured tetrahedral finite element mesh generation.

Nevertheless, apparently a great career option for the metaphorically challenged.

Re:curvature as captive starch

[SlaveToTheGrind]

Data is a bit like wheat, it doesn't give up its curvature easily. Too much applied force creates heat and destroys the end product. The applied force must have exactly the right ratio of compressive to shear stress, which only an expert miller can judge.

Deep, dude... deep. Have you considered writing Slashdot summaries?