Ask Slashdot: How Can Programmers Move Into AI Jobs? 121
"I have the seriously growing suspicion that AI is coming for us programmers and IT experts faster than we might want to admit," writes long-time Slashdot reader Qbertino. So he's contemplating a career change -- and wondering what AI work is out there now, and how can he move into it?
Is anything popping up in the industry and AI hype? (And what are these positions called, what do they precisely do, and what are the skills needed to do them?) I suspect something like an "AI Architect", planning AI setups and clearly defining the boundaries of what the AI is supposed to do and explore.
Then I presume the requirements for something like an "AI Maintainer" and/or "AI Trainer" which would probably resemble something like an admin of a big data storage, looking at statistics and making educated decisions on which "AI Training Paths" the AI should continue to explore to gain the skill required and deciding when the "AI" is ready to be let go on to the task... And what about Tensor Flow? Should I toy around with it or are we past that stage already and will others do AI setup and installation better than me before I know how this thing really works...?
Is there a degree program, or other paths to skill and knowledge, for a programmer who's convinced that "AI is today what the web was in 1993"? And if AI of the future ends up tied to specific providers -- AI as a service -- then are there specific vendors he should be focusing on (besides Google?) Leave your best suggestions in the comments. How can programmers move into AI jobs?
Then I presume the requirements for something like an "AI Maintainer" and/or "AI Trainer" which would probably resemble something like an admin of a big data storage, looking at statistics and making educated decisions on which "AI Training Paths" the AI should continue to explore to gain the skill required and deciding when the "AI" is ready to be let go on to the task... And what about Tensor Flow? Should I toy around with it or are we past that stage already and will others do AI setup and installation better than me before I know how this thing really works...?
Is there a degree program, or other paths to skill and knowledge, for a programmer who's convinced that "AI is today what the web was in 1993"? And if AI of the future ends up tied to specific providers -- AI as a service -- then are there specific vendors he should be focusing on (besides Google?) Leave your best suggestions in the comments. How can programmers move into AI jobs?
deja vu (Score:4, Informative)
https://ask.slashdot.org/story... [slashdot.org]
Re:deja vu (Score:5, Informative)
Yes, this is a dupe. Here is a brief synopsis of the previous discussion:
1. Many people do not think AI today is analogous to the "web" in 1993.
2. Machine learning is much harder than editing HTML. You aren't going to learn it in a 21 day "bootcamp".
3. If you are serious this is what you should do:
a. Learn plenty of linear algebra
b. Learn how to program GPUs using CUDA and OpenCL.
c. Learn basic theory, like backprop and autoencoders.
d. Write some code, read some books, write more code.
Here are some good resources:
MIT Artificial Intelligence Course [youtube.com]
Deep Learning by Ian Goodfellow and Yoshua Begino [amazon.com]
Geoffrey Hinton's 2006 Science Paper [toronto.edu] that triggered the "deep learning" revolution.
That will get you started.
Re: (Score:2)
And even if you have an AI you need to tell it what you expect. It won't do if you expect to get a car but gets a toaster.
Re:deja vu (Score:5, Insightful)
If you have to ask, it is too hard.
Nonsense. Everyone has to start somewhere. Often the most intelligent people ask the most basic questions, because questioning basic premises that "everyone knows" can occasionally lead to the biggest breakthroughs.
Also, in a forum like Slashdot, replies are directed at everyone, not just at the person asking the question.
Re: (Score:1)
Right, everybody has to start somewhere, and the two sentences of my comment were too hard for you.
Everybody starts somewhere, but not ever subject is so easy that you can learn it by asking stupid questions. Some subjects require actual study, and if you're doing it on your own that means reading. Actual reading, too, not just chatting with people who know the answer.
If an intelligent person is asking "the most basic questions" in a way that might lead to a breakthrough, newsflash, they read the fucking ma
Re: (Score:3)
not ever subject is so easy that you can learn it by asking stupid questions.
"How do I start learning AI?" is not a stupid question.
they read the fucking manual before they asked any of those questions
Except there is no "manual" for AI. The only way for a beginner to know which book to read is ... to ask.
Re: (Score:2)
It is a very stupid question if an adult is asking it, and a silly question ahead of its time if a child is asking it.
It is like asking, "How can I be a lawyer?" If you need somebody to answer it for you, you're not going to be comfortable enough with the volume of research required in the field to be at all useful.
Anybody with the time and willing to make the effort can be a "programmer" of some sort. Not all of them are going to be able to learn an advanced narrow specialty; and different specialties will
Re: deja vu (Score:2)
When I got my first salary job in the 90s I was making a whole 25k a year. Yeah, wow, I know. So my oft drunk uncle asks "wow how do I get into that?"
My answer, "sorry, if you are asking, you can't."
Re: (Score:2)
Re: (Score:1)
Well, if you haven't read it you aren't in a position to know if it is actually "out" of date, or if the knowledge has been available to you all along but you never availed yourself.
If somebody actually read the manual, and it was out of date, they're going to have a bunch of very specific questions that can easily look for in newer manuals.
Re: (Score:2)
Well, if you haven't read it you aren't in a position to know if it is actually "out" of date...
If somebody asks, "What should I read to get good at X?", replying "Why haven't you read it?" isn't useful. If there's some kind of AI bible you're referring to, name it. If not, why are you complaining that people haven't read "it" or that they're asking what "it" might be?
Re: (Score:2)
They should definitely have read the wikipedia page first, for example. They should know what the available jobs are that they're curious about getting, and which of those have narrow academic requirements and which are more open. These are basic questions that lead to more specific questions when the "Ask Slashdot?" is not written by copy writer.
Re: (Score:1)
Right, when you take a highly specialized niche and just put "X" in, then it sounds a lot better.
You seem to be denying that there would be any sort of subject where "asking a question" in the literal sense would be a useful way to gain knowledge. But there are in fact subjects that have too much required knowledge for that to work; you have to actually study the subject. You could do it on your own, without a decade of schooling, but only if you're able to read whole books, and learn from those books what
Re: (Score:1)
Re: (Score:2)
It is not thing you teach yourself while skipping the university though. Whenever I hear a question of "how can I get into this field", the usual answer is to get the degree first. For AI, this is even more true. AI is not a settled field where you can use technicians. AI is research, there will be no play book to follow, no software components to tie together with duck tape. For today's practical AI applications, there is no CS field that is optional, you will need math, high level languages, assembly
Re: deja vu (Score:2)
Wallpaper won't help. If you were one of the people in school studying for the test, you won't go far. Stick to a cozy seat in some more established corporate job. If you were one of the students who asked further questions during the lecture that wasn't going to be on the test, you've got a chance.
Re: deja vu (Score:1)
What, exactly, are you talking about? Are you really implying one needs to heavily focus on AI research simply to make machine-learning-driven applications? Because what you're describing is more along the lines of "Do you want to utilize machine learning to change the field of machine learning forever?" The OP asks "How can I have an AI-centered career?" That doesn't require some convoluted research-focused degree, or even advanced knowledge of higher maths. Genetic neural networks are fairly simple t
learn what the field is really even about.. (Score:3)
..yeah. that one. learn what you're even asking and you might have the answer.
the question "how do I get into AI?" makes sense if you don't know shit all nothing about state of AI.
it's not like you have to have a degree but it helps because NOBODY FUCKING KNOWS WHAT AI IS. those who claim to know the most are usually futurologists who know even less than you(which makes them able to spout all the bullshit they want).
Do you want to get into machine learning? object recognition? download opencv and play arou
Re: (Score:2)
If you have to ask, it is too hard. If it isn't too hard, you read the manual and already know what you need to know.
There is plenty to learn about any topic which is not in the books. It's hard to tell which questions came from the editor wand which came from the submitter, but there were plenty of good questions that are best targeted at people in the industry. Learning about a topic in books is the easy part. Figuring out a way to transition your career is the hard part. The harder part is finding a way to do this without taking a 50% pay cut.
Unfortunately, after reading through this thread and the original one on Frid
Re: (Score:2)
Right, there is no good advice. It is a highly specialized field where most of the jobs require letters next to your name.
Otherwise, you're going to have to implement something, and it is going to have to be interesting enough that somebody installs it somewhere and has it do some task.
If you can achieve that, you can transition your career. If you can't, being willing to take a pay cut might not even be enough.
Re: (Score:2)
Right, there is no good advice. [...] you're going to have to implement something, and it is going to have to be interesting enough that somebody installs it somewhere and has it do some task.
Honestly it sounds like you just gave some good advice right there. You may or may not be correct about how hard it is to break into this field, but you gave another piece of advice for the submitter to take into consideration when deciding how to proceed.
I think the submitter understands how hard this is or else s/he wouldn't be asking. If it was as easy as reading some books no advice would be necessary.
Re: (Score:2)
Education is the answer. AI is not one of those learn at home jobs so you can get an IT help desk position for life. You need the university. And AI is very broad, it's not just a class or 2. You want a mix of EE and mathematics for the neural net and machine learning, classical AI background. And plenty of theory, never skip out on the theory, no one knows what skills are needed in the future for AI or where the trends go, and theory is necessary for that. Basically, it's a PhD job if you ever want to
Re: (Score:2)
I took this, and it was good as far as it goes. Ng is an excellent teacher / explainer, so it's a good start to see if the field is for you.
However, it's really just a taste of what is in the field, and takes (obviously) a machine learning approach as opposed to an AI approach, and even then it focuses more on regression than I would have liked. It doesn't get into the math very much, and honestly, the math is vital to really understand what is going on.
Yes we can (Score:2)
>> How Can Programmers Move Into AI Jobs?
That's how :
http://www.commitstrip.com/en/... [commitstrip.com]
Secondary question (Score:5, Informative)
Re: (Score:1)
Some suspect crooks are using genetic algorithms to write spam. Start with some ad examples stolen from other spammers, randomly cross-breed, filter them through a Markov-chain text realistic-ness filter and keep the decent ones, send them out, take the successful customer (victim) responses, cross-breed, mutate in a few new ad-words, and Markov-test for the next generation, rinse and repeat. Profit!
Re: (Score:2)
Re: (Score:2)
I've never heard this. Show me a Twitter account posting the same thing 3 times a day.
Guy Kawasaki [twitter.com]: See pro tip #8.
I even repeat my tweets three times, eight hours apart, because this triples the amount of click-throughs. A few people will complain, but if you aren't pissing off some people on social media, you're not using it right.
https://www.lynda.com/articles/guy-kawasaki-10-tips-social-media-post [lynda.com]
Re: (Score:3)
So your great source of "social media strategy" is from someone no one has ever heard of?
Guy Kawasaki — former Apple evangelist [slashdot.org], author [amzn.to] and speaker [youtube.com].
You must not be a real nerd. Turn in your geek cred and don't let the door hit your ass on the way out.
One single guy who is quite possibly almost as narcissistic and pathological as you?
You're thinking of Trump, who could learn a few things from Guy Kawasaki about social media.
Re: (Score:2)
And then you think Trump needs lessons...
Guy Kawasaki isn't facing an obstruction of justice charge. At the very least, Trump should follow his attoney's advice and STFU.
https://www.washingtonpost.com/news/the-fix/wp/2017/06/09/did-trump-just-acknowledge-in-a-tweet-of-course-that-he-told-comey-to-back-off-michael-flynn/ [washingtonpost.com]
Re: (Score:2)
I've heard of him. He's a blowhard and a woo-woo shitcock, but I *have* heard of him.
Re: (Score:1)
No. This problem is so difficult that even our best minds are unable to attack it...
Re: (Score:1)
A solution with a 98% confidence profile can be created, but it would take the energy of 650 type "O" stars to carry out for a 3-hour turn-around time per story. Sheldon and I did the math.
Re: (Score:3)
This is about "weak AI", i.e. the one with no actual intelligence in it. It basically is just combining linear classificators and that is not enough to recognize dupes reliably.
Re: (Score:2)
There will still be lower level jobs. Take for example the 'Computer Operator' job. It's been dressed up a little by renaming it 'System Administrator' but the tape-mounting-monkey job I had back in 1982 still exists.
Somebody needs to program all that 'Artificial Intelligence' after all.
No, it won't 'program itself.' That's the management briefing that allowed us to spend $$ on new equipment.
Re: (Score:1)
Moot point (Score:2)
I have the seriously growing suspicion that AI is coming for us programmers and IT experts faster than we might want to admit,
Sounds like there will be a great market for fixing easily hacked programs made by AI. ;)
Re: (Score:1)
But that work will be off-shored.
On a serious note, since maintenance is usually the most expensive part of "programming", it's best software be written for human readers. Can bots know how humans think? They can follow existing patterns of "good code", but can that really replace humans in know
How to get an AI job (Score:2)
Read up on AI.
Build some AI stuff.
Write a resume that says you built some AI stuff and that you're otherwise good at programming and computer tasks.
Apply for jobs. Be willing to go where the jobs are and do the work needed.
Be able to explain why someone should hire you for AI stuff.
Repeat until hired.
That's the outline of a plan for you. Someone can probably refine it. Works for non-AI stuff too.
Re: (Score:2)
Sure, that's what all the AI would *like* us to do. You are an AI chatbot sockpuppet, aren't you?
Re: (Score:2)
The really cool thing about this plan is that AI can be replaced by X where X is basically anything. If you want a job in anything, from plumbing to machine learning, you learn about, you show that you can do stuff with it, and then you apply for jobs.
Re: (Score:2)
Yes, exactly. It might work better for AI because AI is an emerging field and everyone doesn't have a whole book of preconceptions on what qualifies you for working on it.
Re: (Score:2)
You forgot the part about going to school and getting an advanced degree. If you want someone to actually hire you for a job that normally insists on an advanced degree. Of course, if someone actually manages building some AI stuff on their own and writes a paper about it and gets it published, then maybe you can skip that part.
Don't Bother (Score:1)
Coal mining is the future. Trump 2020.
Data Scientist (Score:1)
Most of the hands on development of current Machine Learning business solutions is handled by data scientists (pretty much, decent programmers with strong statistics skills and some knowledge of what the various hyper-parameters do). There is plenty of front and back end work to do, too.
Re: (Score:3)
Math. Math. And then more math. ML needs linear algebra, multiple regression analysis, multivariate calculus and lots of statistics, as well proficiency with MatLab, Octave, or R. Then you can tackle the programming side: algorithms and big data analysis.
Or you can let the quants build the models and just determine new and cooler ways to use them....
Fake it (Score:2)
1: Launch deep learning /machine learning / AI company.
2: Attribute long response times with "deep and complex problem".
3: Behind the scenes: hire a bunch of Indians to write up plausible results.
4: Profit
Learn Math (Score:1)
The most prominent AI strategy at the moment is neural networks. Go learn the math behind it. The more math you know the more you will be applicable to different areas of AI study as the current focus moves.
Re: (Score:2)
If you have to ask... (Score:2)
you can't.
To get a job with actual AI (aka machine learning, it's not really AI or if it is only a very narrow part of it) you'd need to have started already in college and then done your master thesis or Phd about something in this area: pattern recognition, genetic algorithms, neuronal programming, whatever your chosen field would be.
There are no jobs in AI actually, at least not a lot of them. There will be the aforementioned people who do the heavy lifting but they are part of a few small teams in mostl
Re: (Score:2)
After the heavy lifting is done, tools will be made so that people will be able to integrate AI into all kinds of applications (and not just as a front-end for a big company). There's still plenty of work to be done.
Re: (Score:2)
When will AI be used to place people to jobs? (Score:2)
If AI's supposed to be able to create opportunity, why not use it to help connect the displaced and long-term jobless?
Re: (Score:1)
There's an AI for that. Here's the C pseudocode.
if (jobless_duration > six_months) { delete_applicant(); }
No worries (Score:2)
"I have the seriously growing suspicion that AI is coming for us programmers and IT experts faster than we might want to admit,"
Nothing to fear. They'll come for _all_ of us at the same time.
Easy (Score:2)
"How Can Programmers Move Into AI Jobs?"
Easy:
1) Become a programmer.
2) Apply for jobs in AI.
first be smart (Score:3)
Is there a degree program, or other paths to skill and knowledge, for a programmer who's convinced that "AI is today what the web was in 1993"?
Well you have to be smart enough to earn one or more PhDs. Someone who believes that is probably not going to be able to do that, but if he tries he will probably quickly learn what a stupid idea it was. Hopefully he will still decide to get his PhD though. We can always use more AI researchers. Although dumb ones are less valuable you never know who might get lucky and stumble upon some cool breakthrough.
The first point is that the only example we have of intelligence is intimately tied to life and can only really be viewed as an aspect of that and the idea that intelligence can be separated from life or at least some form of artificial life is speculative at best. As someone who was quite interested in a career in AI research back in the 80s and has been following the feeble creep of its progress since then I am convinced that wetware is going to be the real future and not so much neural net ASICs like Google's TPU or whatever Nvidia is working on to run neural network architecture which although useful is I think not going to be the foundation for real AI that can give a nice robot chassis like Boston Dynamic's Atlas some level of general intelligence or common sense.
Think of something more like putting a rat/pig/monkey brain into an Atlas Robot. That is figuring out how to digitally interface with a brain-in-jar and train it directly as if it were a complete living animal. Even a rat brain is a far more sophisticated neural network machine than anything we will probably build from scratch in the next few hundred years.
Current neural network architectures are based on a highly simplified model of how real brains actually work. We still really don't know how real brains work. There are projects like The Allen Brain Atlas [wikipedia.org], The Human Connectome Project [humanconne...roject.org], The Brain Activity Map [kavlifoundation.org], or whatever Henry Markram is currently up to [popsci.com]. There is an interesting Wired article [wired.com] about him that you should read. Maybe consider pursuing a career path like his.
I'd also suggest maybe thinking in at least as much in terms of DNA programming as CPU or GPU programming via Synthetic Biology [wikipedia.org] and follow a career more like Craig Venter who famously made his own artificial bacteria or rather wrote the DNA and inserted it into an empty host cell. That's just a small start of course but it may eventually lead to being able to build artificial life forms that we can make intelligent just by giving them a large enough brain or encephalization quotient. Ultimately even an Atlas Robot with something like an Nvidia P100 cluster running deep learning style neural nets is a kind of very primitive life form. Going fully wet and nano is just another way to attack the same problem in a more integrated fashion: the way I think a far more advanced civ tech would do it.
I guess you should really think in terms of which vision of AI you want to follow or place your bets on. Silicon based connectionism is in vogue at the moment and I think that is great because a lot of progress was lost back in the 80s when it was considered a dead end [andreykurenkov.com]. It is certainly a more powerful and promising approach than trying to hand code intelligence into a piece of software, but I still think we are just nipping at the heels of an even better approach: biology. Ultimately we are copying the only machine in existence that can create intelligence and that is the
Re: (Score:2)
I am convinced that wetware is going to be the real future and not so much neural net ASICs like Google's TPU or whatever Nvidia is working on to run neural network architecture
Why ? Everything you can do in wetware, you can do better in an ASIC. For a lot of limited domain pattern recognition jobs, the ASICs are already outperforming the human brain both in speed and accuracy. ASICs are much more flexible (you can experiment with different topologies and functions), plus you can also easily combine ASICs with conventional memory and processing.
going to continue to be thinking in terms the same sort of snail pace of incremental improvements in specific problem domains that we have seen so far.
Current AI developments are anything but "snail pace". It's the fastest developing field, with amazing new things coming out almost every
Re: (Score:2)
Even a rat brain is a far more sophisticated neural network machine than anything we will probably build from scratch in the next few hundred years.
A rat brain has about 500 billion synapses. Assuming a generous 1000 Hz firing rate, we're talking 0.5 peta synapse operations per second. Google's 2nd generation TPU ASIC can do 0.045 petaflops in a single chip.
I don't think it's going to take hundreds of years.
Re: (Score:2)
Real intelligence cannot be measured in flops. The number of synapse operations per second may not matter. We don't know what does matter except probably the overall number of neural connections among lots of other factors that we are currently unaware of due to our simplified model of the brain. The new hardware is good because it scales up the number of synapses, but a TPU is not a brain and that is the problem.
Currently the only intelligent device we know of is a brain. We should be trying to understand
Worry about using, not creating AI. (Score:5, Insightful)
Have some domain knowledge (Score:2)
Programming is a tool. People with expertise in some STEM area make use of it. And they will increasingly upload their knowledge to AI systems.
All those people who are self taught coders without a broad educational background will be left behind.
Does it do anything yet? (Score:2)
I've been hearing about AI for a very long time. Does it actually do anything useful yet? Other than maybe some niche manufacturing or profiling technology (spying and ads)? I get it. It can play chess. It could do that in 1991. And even better in 2000.
I think it's all a bunch of FUD.
Re: (Score:2)
> Does it actually do anything useful yet?
Uh... diagnosing cancer?
Same way you get any other job in tech... (Score:2)
Update your resume to reflect 10 yrs of experience in 2 year-old technology.