Slashdot Mirror
Tencent Says There Are Only 300,000 AI Engineers Worldwide, But Millions Are Needed (theverge.com)
An anonymous reader quotes a report from The Verge: It's well-established that talent is in short supply in the AI industry, but a new report from Chinese tech giant Tencent underscores how great the need might be. According to the study, compiled by the Tencent Research Institute, there are just 300,000 "AI researchers and practitioners" worldwide, but the "market demand" is for millions of roles. These are unavoidably speculative figures, and the study does not offer much detail on how they were reached, but as a general trend they fit with other, more anecdotal reports. Around the world, tech giants regularly complain about the difficulty hiring AI engineers, and the demand has pushed salaries to absurd heights. Individuals with just a few year's experience can expect base pay of between $300,000 and $500,000 a year, says The New York Times, while the very best will collect millions. One independent AI lab told the publication that there were only 10,000 individuals worldwide with the right skills to spearhead serious new AI projects.
Tencent's new "2017 Global AI Talent White Paper" suggests the bottleneck here is education. It estimates that 200,000 of the 300,000 active researchers are already employed in various industries (not just tech), while the remaining 100,000 are still studying. Attendance in machine learning and AI courses has skyrocketed in recent years, as has enrollment in online courses, but there is obviously a lag as individuals complete their education.
Tencent's new "2017 Global AI Talent White Paper" suggests the bottleneck here is education. It estimates that 200,000 of the 300,000 active researchers are already employed in various industries (not just tech), while the remaining 100,000 are still studying. Attendance in machine learning and AI courses has skyrocketed in recent years, as has enrollment in online courses, but there is obviously a lag as individuals complete their education.
Tecent... isn't that 50 Cent's little brother? What is a rapper doing telling us what we need for AI?
Besides it seems the AI's are better at building themselves than we are, so I say just give them unlimited compute power and internet access and have at it.
"There is more worth loving than we have strength to love." - Brian Jay Stanley
If the salary isn't enough to get you interested it's very likely the very last job that will be taken over by AI. It would also be a great opportunity to be a part of the next major transformation in human civilization.
Glassdoor says:
How much does a Machine Learning Engineer make? The national average salary for a Machine Learning Engineer is $128,549 in United States.
Yikes so even though the opportunity for profit is limitless. The available workers are a fraction of the demand and this is a sufficiently difficult subject that nobody will obtain credentials without hard work.
It's still not as valuable as a Masters in English Literature from an Ivy, or even a law degree from a mediocre school. Playing with math that is currently almost magic and practicing a craft that approaches playing god. You're still not worth as much as even the most lowly of the elites you engineering scum and you can bet that we'll be shoving your wages way down as soon as someone shows us how to replace you with an H1B
Attendance in machine learning and AI courses has skyrocketed in recent years, as has enrollment in online courses, but there is obviously a lag as individuals complete their education.
No direct experience but an acquaintance of mine quit his job in ASIC layout to pursue a career in machine learning. He took a bunch of classes outside of a formal degree program and found that breaking in the field wasn't nearly as easy as he expected. I haven't talked to him in about six months but he was still looking the last I knew.
This might explain the "shortage". If most of the students are in bootstrap style programs but employers deem those programs unsuitable, it is going to be a while before the gap is closed.
They said the same thing in the 90's about programming, and look how that turned out. A few hot spots if you wanted to work for a lot of money at the expense of quality of life, and competing with foreigners from third world countries. I've known a few people who left for a real good position in my lifetime, but that only lasted for so long and a lot came back.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
More "AI" hype. Show me the job listings.
...Google made an "AI" that created an "AI" that's better than itself. Seems like the direction to go?
way I see it last thing I need (as a member of the working class) is more automation. I see lots of folks railing against socialism and nobody giving any answers about what to do when there's suddenly millions of jobs just gone. I hear the same tired crap about new jobs in a new economy that I heard when the outsourcing began in the 90s and carried through into the 2000s. Anybody else remember biotech? Turns out you don't need that many biotech engineers. Not at the level of work I can do. If I was a genius maybe, but if everybody was a genius we wouldn't be in this mess, would we.
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
He said "learn to code", not "learn to cope".
Ezekiel 23:20
the bottleneck here is education
Indeed it is, and it will remain, since tech giants hired university staff that could teach AI
Change the way AI is done.
It doesn't have to be so esoteric: make it "visible" as layered voting machines where each factor "votes". Use data layouts similar to spreadsheets and relational database reports so that "regular" office workers can study, arrange, relate to, and adjust factor weightings, mask weightings, and routing paths (similar to "hidden layers") as needed.
Color coding, similar to Excel's conditional formatting can make high-match and low-match factors stand out for test cases or trouble-shooting.
Staff can be divided similar to the processing tree. For example, in vision recognition, one group can focus on people identification, another on furniture and building identification, another on outdoor patterns, etc. The idea of one giant do-it-all monolithic neural-network is not practical if we want rank-and-file AI and dissect-able AI. Bring in modularity and divide-and-conquer techniques.
You may need an experienced AI domain specialist to help divide up tasks and provide factor (test) guidelines or drafts, but once staff have their basic assignments they can focus and tune without being caught up in the big picture and way-out theory.
Table-ized A.I.
I was thinking... 300k? Are you serious? What actually justifies calling yourself an AI developer?
I haven't really put that much effort into AI and while I've done a lot of coding I imagine probably counts as AI from what I understand about it, to be a real AI developer would require a lot more than just writing code which makes decisions based on statistical analysis and thresholds.
For example, I wrote a signal decoder years ago which couldn't be handled using traditional DSP theory. High pass and low pass filters couldn't work. There was a signal that took a digital signal transmitted over an analog satellite broadcast link and then sampled at 2.7 times the original signal frequency. The phase was erratic, the amplitude was erratic, the white noise was crazy.... even human visual inspection of the signal was extremely difficult. I managed to write code that would progressively reconstruct the data from the signal given surrounding data. As it was reproducing formatted screens of text, I would perform pyramid scans surrounding the character and identify the formatting of the text to guess the approximate phase and amplitude and noise types of the current block to be decoded. As it decoded more text, it learned more and had an increased success rate. Then when phase, amplitude or noise types shifted, it would decrease its certainty regarding the quality of it's learned knowledge and go back to basics.
This I assume was AI, but I have no idea. There was a problem that needed to be solved. It wouldn't work using normal algorithms. So, I made a new algorithm that could solve the problem similar to how I would solve it manually using my eyes and intuition while also compensating for a limited data set by defining a simplistic series of rules that defined something that could considered a thought process.
Now that being said, for code to be AI, I would expect it to be trying to do something more interesting. I saw the research posted by Google where an engineer taught a robotic arm to open a door when it encountered one, showed it how to use a door knob and then let it figure out how to use a different door knob. The same technology could be used for example to say "If you encounter a screw and you encounter a bolt, put the two together and tighten it but not too much". With enough rules like that, it could easily replace humans in most manufacturing roles.
Use the same ideas and build a single type of robot that can lift, fold, manipulate and sew different types of fabrics. This sounds a lot easier than it is. Try as a human to sew two pieces of equal sized cotton together using a machine. Then try slinky silk or nylon. The texture of the fabric on the silk will constantly shift and slip, it's not a stable grid. The dog feed pulls the bottom piece but not necessarily the top. The last piece of fabric you sewed may have left a residue behind that effects whether the presser is sticky during the first bunch of stitches on the new fabric, etc... someone who sews a lot will have subconsciously learned to hold and manipulate fabric just the right way... which they can't really explain. Someone who doesn't will try sewing with silk and just never try again. It's a task that simply can't be solved by traditional robotics because as with humans, the machine driving the robot needs to make a lot of assumptions with incomplete data to achieve the workflow.
So... if there are 300,000 people in the world with the knowledge and studies for things like writing AI that can solve problems like the fabric and sewing problem... I'd be shocked.
Of course there are probably a bunch of people making software to high-frequency trade or play poker online.