Three successful AI pros on how they built a career in artificial intelligence
Artificial intelligence is central to the ongoing tech revolution, and it's getting smarter all the time. The driving force behind computer vision, speech analysis and natural language processing, AI impacts industry and society in numerous ways — and will continue to do so far into the future.
It's no surprise, then, that the AI field is rife with career opportunities — so many of them, in fact, that the sector now faces a unique challenge: There are too many jobs and too few qualified candidates. On the up side, that means it offers virtually guaranteed (and well-paying) employment for those who've got the goods.
So how does one get into AI, and what does an artificial intelligence career path look like? We asked some of the field's top experts to share insights from their journey to help guide the way. They include Satya Mallick, founder of Big Vision LLC and interim CEO of OpenCV.org; Jana Eggers, CEO of Nara Logistics; and Joanna Bryson, an associate professor of computer science at the University of Bath in England.
How did you get into the AI field?
Founder, Big Vision LLC/Interim CEO, OpenCV.org
I stumbled upon computer vision (a branch of AI) as an undergrad at the Indian Institute of Technology, Kharagpur (India) around 1999-2000. I saw one of my seniors doing a robotics project where he used cameras to help the robots "see." I found the idea so fascinating that I decided to apply to PhD programs in computer vision and machine learning after my undergrad.
CEO, Nara Logics
I was researching conductivity in plastics at Los Alamos National Laboratory in the early '90s, and neural networks and genetic algorithms were tools I used for some of my work. When I left research and went into business, I landed at a startup doing expert systems for logistics. After that, I went to one of the original search engines, Lycos. So basically, because I was always in cutting-edge technology, some form of AI was typically the right tool for the job I was working on.
Associate Professor, Dept. of Computer Science
University of Bath (U.K.)
My first degree was in non-clinical psychology from the University of Chicago. And I also was good at programming, so I worked in the computer science department there. I was interested in AI, but the first time that anyone taught it was the year after I graduated. But they let me tutor it anyway. So then I spent five years in Chicago paying off my undergraduate debt. And it was right at the period where we saw computing decentralized. PCs were starting to become a thing, and we were figuring out that not every computer needs to know everything. And I had decided that I never wanted to be a professional programmer, so I went back to graduate school for artificial intelligence. I wanted to go to MIT, but I chose Edinburgh because I wanted to go abroad but I couldn't speak any languages. I was lucky at that point, because it was the only place that in 1991 that would have given me a master's degree in AI. From there, I found out about Rod Brooks [at MIT] and behavior-based AI — the idea that rather than trying to understand everything, you had specialist sub-parts of a robot that understood different problems. And then a few of us decided we would try to get into MIT and work with Brooks.
What is the scope of your work in AI?
Mallick: I work in the field of computer vision, probably the most important sub-field of AI. In computer vision, our goal is to make machines make sense of the world by image and video analysis. I run a computer vision and machine learning consulting company called Big Vision LLC and a popular blog called LearnOpenCV.com.
At my consulting company, we tackle a variety of computer vision problems ranging from detecting parasites in horse feces to identifying the model of high-end fashion bags in user uploaded images. Our biggest project involved OCR and fraud detection in ID documents. In addition, we have worked on security applications involving intrusion detection, vision applications for urban warfare, sports analytics and biomedical devices. I am also the Interim CEO of OpenCV.org which maintains the largest computer vision library in the world (OpenCV). We just launched a Kickstarter campaign for three AI courses. It is doing very well.
Eggers: Nara Logics provides an AI platform for large enterprises and the federal government that focuses on decision support. Much of deep learning is focused on perception — recognizing images, understanding language — we focus on the next stage decisions from what's perceived. Our platform helps bring together the perception streams of vision, language, sensors, etc. to support better decisions, from what to show to a customer (personalization) to how to prioritize operations (decision support).
Bryson: For my PhD, I wanted to do psychology, but I wound up [in AI] at MIT. So I focused on something called systems AI, which is basically about how to make it easier to build AI. There are lot of people that focus on one magic algorithm that'll beat the other algorithms, but I noticed that we already had a lot of capabilities and that the real problem was putting them together — the engineering side of it. So that was what I focused on. And now it has become a really big deal. Systems engineering is actually a discipline that makes thing safe, going back well before AI.
I'm also still trying to understand human cooperation and the evolution of cognition — when we use it and when we don't — so I'm focusing a lot on understanding the different [ways] in which people cooperate and help each other out, and when they don't. We're working on a paper right now about why political polarization is well correlated with wealth inequality. How do they interact together? Then, of course, there's AI ethics. I'm spending all my time running around doing policy work, so just talking to people and trying to connect the dots. I think of it as just taking the best stuff I've heard and putting it all in one place, but other people see it as very innovative.
What do you find most rewarding about your work?
Mallick: One of the luminaries in our field, Dr. Andrew Ng, has compared AI to electricity. It is transforming multiple industries, and it is hugely rewarding to be part of this revolution. Some of the applications we have built remove drudgery from common tasks, some keep us safe, and still others in the medical domain improve health and even save lives. As a practitioner of this craft, it is deeply satisfying to see your work make a difference.
The other part that provides a lot of meaning to my life is my work as a teacher of online courses. At OpenCV.org, we are on a mission to train the global workforce in AI. This is an emerging field, and we are totally aware AI will cause the loss of many jobs through automation. So it is our responsibility to help people upgrade their skills and teach them what we learn through the practice of our craft.
Eggers: Enabling our customers to use their data to see forward versus analyze backwards.
Bryson: It's just intrinsically interesting, understanding how intelligence works. I guess that's my biggest internal reward, though I do also like to be helpful.
Which aspects of your work are the most challenging?
Mallick: Challenges in AI come in many different forms. First, we have technical challenges. Sometimes we do not have enough data. At other times, the data is available but it's extremely noisy, or not labeled in a way that can be easily used. In a few cases, people vastly underestimate the irreducible uncertainty of a problem — predicting elections is one such problem domain. Second, we have ethical challenges with no good answers. For example, would you use an AI that helps 99 percent of the people, but is heavily biased against one percent? Removing bias from your data is extremely challenging.
Eggers: Demystifying AI for customers. They are both intrigued and frustrated by the hype and mystique around AI. Our focus is quickly getting our customers from proof of concept to production.
Bryson: It's weird things, like [schedule] coordination. Because everything is interesting and everything is challenging and you never know when you're going to have a conversation that will help make a big difference.
What are the most crucial first steps for those who want to pursue an AI career?
Mallick: To have a technical career in AI, you first need to have good programming skills. The programming language Python has become the default choice for AI researchers and engineers. Although depending on the domain, knowledge of C++ may be required. If you are already a good programmer, the next step is to find an online course that can teach you the basics and gradually move you to mastery. There are several good choices like courses by OpenCV.org, Coursera, Udacity and DeepLearning.ai.
Eggers: Understanding that the tech is the easiest part of AI. The data and the results are both more critical. And those are both driven by the organization.
Bryson: Doing post-grad education, even if it's only a master's, to get insights from another discipline. When I was an undergraduate, as people graduated they were more likely to go into the careers that they had [experienced] in summer jobs than in their majors. So when you're looking for a job, it's really important that you bring in at least some of the skills [the company] needs and that you're [also] giving them something. But look for [experience] that's going to take you at least part of the way to where you want to go. Get your hands dirty doing any amount of work in programming or data analytics or whatever. If you're at a university, there's a billion little jobs that you can get just so you have some experience. That helps a lot.
How valuable is post-secondary education? What about graduate degrees?
Mallick: There are two huge benefits of a college education. First, it provides structure to the learning process. Without this structure in place, it is easy to get lost in confusion or lose motivation. Second, completing a college degree signals a level of competence in the job market. That said, increasingly people care more about the portfolio of work you have done and less about your resume. When a student releases their code on GitHub on a regular basis, we see first hand what they are learning and how good they are. If some of their work is unique and creative, I personally would not care about their college degree.
Eggers: As a PhD drop-out, my perspective is that a graduate degree grants you respect — and there are other ways of earning that respect. So it really depends on you and what you value. You can succeed either way.
Bryson: Not every Ph.D. is worth doing. Especially in America right now, there are a lot of people who will sell you a degree that isn't necessarily going to help you. Getting a degree now is partly about getting an education and partly about being in a cohort of people that are really trying to understand things, like I was at Edinburgh and MIT. They used to say a master's degree is the best thing to have if you're going to go out into industry, but you want a PhD if you're going to work in government labs or the top drawer of industrial labs — or if you want to be an academic yourself.
Are certain college degrees better than others?
Mallick: Absolutely. A degree from Stanford in AI is worth a lot more than many other universities because you get to work with top researchers who are at the cutting edge of AI research. The choice of your major also makes a huge difference.
Eggers: AI needs all degrees involved. However, basic tech understanding is crucial to participate.
Bryson: We have great students coming to do our master's degree [at Bath], and often the ones who wound up being the best were those coming from a psychology base. So psychology is a great degree. Math and physics will help you get into machine learning, but they don't necessarily give you the whole perspective on what's going on. Thinking about the dynamics of change is really important for understanding society. I'm valuable because I actually build AI and understand how it works, but I also understand how society works.
What resources — written or otherwise — should people tap in order to gain more insight about a career in AI?
Mallick: Unfortunately, good answers are spread across many different pages on the internet. There are three large application areas in AI: computer vision, speech analysis and natural language processing. Each one requires a different learning path. So, I would suggest people Google those terms and see what interests them. Here is a list of some blogs that people may find interesting (biased toward computer vision, because that is my area of expertise): LearnOpenCV.com; TowardsDataScience.com; MachineLearningMastery.com; PyImageSearch.com.
Eggers: This depends on how you learn and where you want to focus. Examples: Critical is learning how to develop tech products — my go to book for this is Marty Cagan's Inspired. I also recommend O'Reilly's AI conference because of the breath of material covering tech, research, business, ethics, etc. You can get access to the conference materials via their learning platform. If you want to be more on the R&D side of AI, you have resources like arxiv to keep up with the latest research.
Bryson: If they have specific questions, I usually point them at particular blog posts or papers. It's so easy to do research now that everybody has access to Google Scholar. Even if you hit a paywall, you're likely going to find that someone has put the PDF online.
Once someone gets a job in AI, what are the prospects for advancement early on and down the line?
Mallick: AI is a rocket ship that is taking off. Even entry-level jobs are insanely lucrative, paying two times or more compared to regular programming jobs. The reason is a huge demand for AI talent and not enough people with the right expertise. In the long run, these salary levels may not be sustainable, but people who get on this rocket ship in the next five years or so will have amazing careers financially as well as in terms of the quality of work.
Eggers: I don't see AI as being different from any other leading-edge tech. You have an ability to jump around from company to company (which can be good and bad) because companies are always looking for people with experience. You have the challenge that your own company might be slower to move forward than you want. Basically, there is a fine line between leading edge and bleeding edge.
How will the AI field change in the short- and long-term?
Mallick: With our current level of understanding of AI, we can solve many problems in several industries. The short term game is all about data. The organizations that acquire the most data will always have a huge advantage over those that don't. One does not need a team of elite researchers to solve these problems — we only need a large quantity of high-quality data and good engineers. In the long term, though, we need elite researchers for technical breakthroughs. Currently, the most successful AI algorithms do not learn by themselves. They are taught by humans by the painstaking process of collecting and labeling data. The next breakthrough in AI will occur when machines learn by themselves by observing the world, much like humans do today.
Eggers: Companies are now getting more serious about AI delivering results versus "tests." The longer term change in the field will be the widening of the aperture of AI — a longer depth of field into business.