So You Want To Be An AI Developer?
by April Smallwood
Since 2014, AI startups have experienced an explosion of funding, giving even more incentive to the keen computer scientist. In essence, programming computers to think for themselves may just be the ultimate goal. With a massive $976 million invested since 2010 into AI startups — peaking at $201 million in Q4 2014 — AI developers may be torn between joining a tech giant with R&D budgets in the millions, turning around an existing organisation, or co-founding a startup.
What was once the domain of science fiction, AI is cropping up everywhere. It’s taking notes every time you use Google for search; is an essential component of Facebook algorithms, creating a smarter news feed of things you actually want to see; it’s choosing recommended albums on Spotify, based on your past activity. In short, AI development is the science of machines that can solve problems as a human would. For Wired magazine in October 2014, Kevin Kelly wrote: “The business plans of the next 10,000 startups are easy to forecast: Take X and add AI. This is a big deal, and now it’s here.”
Enrolment in CS courses are steadily climbing. According to the 2015 Taulbee Survey — an annual survey conducted by the Computing Research Association — US enrollments in Computer Science were up 13.6% on the previous year. In addition, software engineering, artificial intelligence, databases and networks are the most popular areas of specialization for doctoral graduates, in that order.
Meanwhile, use of Github’s awesome machine learning has 16K stars. Data scientists and AI developers have the best pay in entry-level careers and are supposedly the smartest people in the world! So, how do you become an AI developer?
Angela Bassa, a data science manager at EnerNoc, and also a WTB speaker, has some advice for you budding AI devs. “The pace at which our field keeps growing is wonderful and amazing and a testament to how much can be achieved, but it is also seductively distracting,” she says. “It is easy to become enamoured with the latest trend or tool, and the more complex models may not be the most appropriate for every application.” Angela urges you to start simple:
“break a problem down to its atomic components, then build it back up towards a solution, and only ever add complexity if a simpler model doesn’t get the job done”, she says.
We also asked research scientist at OpenAI Ian Goodfellow for counsel.
“Focus on learning the fundamentals: good linear algebra, probability, and software engineering skills. The state of the art in machine learning changes from one year or even month to the next, but the fundamentals stay the same for decades.”
Naturally, you’d want to earn your Bachelor’s degree in CS, Master’s degree in Computer Science or Computer Engineering, and/or a PhD in either of those. That said, we recently asked academic and machine learning expert Tom Portegys if those interested in AI should get a PhD, as he did. His response?
“No. If you’re in academia, it’s good to have that. However, I know so many people who are super highly skilled who never graduated from college,” he says. “They started out on the computer, learned from other people and just kept practicing. Especially in the healthcare startup that I’m in, they probably want to see some PhDs. In most situations, you don’t need that.”
Here are some good starting points:
Foundation
- Education — a Bachelor’s degree in CS or DS; Master’s degree in computer science or computer engineering, and/or a PhD in either field, or in machine learning.
- Labs provide cutting-edge research you get trained to design, UX, UI, Implementation and testing, etc. Some lab projects end up as startups in their own right
Languages
Become well-versed in C++, Java, Python and R.
Practice
- There are many opportunities to practice and meet like-minded peers at meetups, hackathons, slack groups, conferences and other course
- Check out Open Source libraries, for example Awesome Machine Learning Github
- Recommended Python libraries: http://www.learnunbound.com/articles/recommended-python-libraries-for-budding-data-scientists-
- Best is just to practise on an Open source projects to familiarise yourself with corrections. etc (e.g. http://numenta.org/)
- As you probably know, YouTube have awesome tutorials and slideshares
- Be helpful in Github, and answer Qs in Stackoverflow. Quora is also useful in the community.
Start a project
Come up with an idea and put it to the test. Find others you want to work with and code away. Tom Portegys says,
“I recommend you just start coding. A lot of people do this. And they get it wrong and they code again. They get it wrong and code again. Spending a lot of time designing is a lot of overhead and it doesn’t pay off.” Instead, get your prototype ready.
SDK (software development kit)
Research the industry you find most attractive and try out their relevant SDKs.
A gentle plug, if we may?
In the more immediate future, join us for AI WithTheBest. It’s an online conference with 100 high-profile speakers who’ll be dispensing invaluable knowledge of the field. Visit AI.withthebest.com to see the full agenda.
A hint of the speaker line-up?
Frédéric Bastien, Lead Theano Developer
Matthew Taylor, Open Source Community Flag-Bearer at Numenta
Alison Lowndes, Developer Relations NVIDIA
Ian Goodfellow, Research Scientist at OpenAI
Adam Gibson, Founder, Skymind
Angela Bassa, Data Science Manager, EnerNOC
Dennis Mortensen, Founder X.ai
Peter Norvig, Director of Google Research
Kamelia Aryafar, Senior Data Scientist, Etsy
Martin Bedard, Lead AI / Gameplay Developer Ubisoft
Originally posted on BMA Media
