Ever since mathematician Alan Turing first asked, “Can machines think?” in 1950, technologists the world over have been working to answer that question in the affirmative. The result: artificial intelligence. With AI, machines use data to mimic, and even improve upon, what was previously only possible with the human mind, and are capable of performing human-like tasks.
These days, the artificial intelligence industry is booming, promising to make our lives easier and our worlds more personalized. From chatbots and virtual assistants to automated threat detection and stock market predictions, AI has become a massive industry, and everyone seems to want a piece of it.
What Is An AI Engineer?
But the development, implementation and maintenance of this technology requires experts with quite a bit of experience and know-how. This is where AI engineers come in.
What Is an AI Engineer?
Artificial intelligence engineers are in charge of developing intelligent algorithms capable of learning, analyzing and predicting future events, turning those algorithms into AI models and systems, and then testing and maintaining them.
We encounter the work of AI engineers every time we use Netflix, Spotify or YouTube, when machine learning customizes suggestions based on past behavior. Or when we’re able to have productive conversations with a chatbot or AI voice assistant. AI engineers develop a lot of company-facing products as well, helping them increase their efficiency and profits, cut costs and make more informed business decisions.
Tariq Haque works as an AI engineer at CVS Health, where he helps the company make more informed decisions about their customers and interact with them more efficiently.
“For example, I want to identify what my customer would like to buy. I cannot do it manually,” Haque told Built In. “That’s where AI comes in. It will run all those complex calculations and do it for me and identify what the customer wants, and I can deliver that to the customer.”
Meanwhile, Gayatri Shandar, a software engineer focusing on AI at LinkedIn, mainly works on the site’s advertising AI, which essentially allows LinkedIn to continue showing users ads that are relevant to them despite the privacy and transparency constraints imposed by Apple, Google and others. Before LinkedIn, she was a machine learning engineer at Peloton, where she worked on the company’s recommender system, providing suggestions for which classes users should sign up for based on their instructor preferences, fitness level and so on.
“We do a bit of data science, we do a bit of engineering, and — depending on the company you work for and the size of the company — we also do a little bit of product, doing stack for new features and talking to product managers and other stakeholders as well in getting that feature up and running,” Shandar told Built In. “[AI engineers] wear a lot of hats.”
The Responsibilities of an AI Engineer
AI engineers play an important role in organizations that rely on artificial intelligence. They are responsible for not only identifying problems that could be solved using AI, but they’re also in charge of the development and production of AI systems, as well as implementing them.
Some specific tasks include creating and managing the development and production of an AI infrastructure, conducting statistical analyses and interpreting the results in order to guide future decisions, building AI models from scratch and helping product managers and other teams implement and analyze them, and transforming machine learning models into specific applications.
AI Engineer Responsibilities
- Creating and managing the development and production of AI infrastructure
- Conducting statistical analyses and interpreting the results
- Building AI models from scratch
- Transforming machine learning models into specific applications
- Data wrangling
Day to day, Shandar said AI engineers are responsible for doing a lot of data wrangling, or making sure that the data being fed to the models is correct and what is expected, as well as finding ways to better store, prepare, extract, transform and load that data. They also are tasked with figuring out which modeling techniques work for the given problem they want to solve, whether that be machine learning, deep learning, natural language processing, computer vision and so on.
“We do a lot of thinking,” Shandar said. “Thinking about how we can improve a model, how maybe our models can be built in a way that reflects whatever problem we’re trying to solve for at that moment.”
“We do a lot of thinking. Thinking about how we can improve a model, how maybe our models can be built in a way that reflects whatever problem we’re trying to solve for that moment.”
Haque said most of his time working as an AI engineer at CVS Health involves just good, old-fashioned programming. It’s a lot of building software, testing, deploying, testing it again and going back to refine it. In the end, he and his team come out with products that have proven to be massively beneficial to the company. In fact, one of the products he helped build for CVS Health is saving the company upwards of $2.5 million annually, Haque said. “It’s a huge impact.”
Not all AI engineers are responsible for improving a company’s efficiency, though. Kulsoom Abdullah, an AI engineer at Duke University Health Systems’ Bashir Lab is more interested in furthering medical research and diagnostics — specifically by applying deep learning technology to the analysis of medical images. Because Abdullah works in academia, her role as an AI engineer looks different from those working in the corporate world. She’s not a professor or postdoctoral fellow, so the work she’s doing has an application as opposed to just pure research. But the audience for the work she does is smaller than what it likely would be were she working at an industry level.
“You have more time in academia to work on something,” she told Built In. “In industry, because of it being more on a quarter kind of system, and you have your stakeholders and business units, you do have to spend a certain amount of your time on things that are going to provide, say, immediate value.”
AI Engineers vs. Data Scientists
Before Duke University, Abdullah worked as a data scientist at companies like Anthem and ADP, and she says there is “a lot of overlap” between AI engineers and positions like data engineers and data scientists. In some cases, the work of a data scientist at one company could be the work of an AI engineer at another.
There are some key differences, however. Data scientists handle the data collection, analysis and visualization, then sometimes build models off of what they find. AI engineers design and build AI systems and products, among other things.
For instance, a data scientist may be able to figure out what perfume a 25-year-old woman living in New York City and making $70,000 a year would be more likely to buy. And an AI engineer will use that information to create an automated product or a tool to put those insights into action without the need for human intervention, like a shopping suggestion tool or a targeted email marketing campaign.
Andrew Seligson, an AI engineer at Entanglement, a quantum computing and AI startup focused on cyber threat detection, describes the work of AI engineers as a kind of “outgrowth” of what data scientists have been doing for the last five or 10 years.
“There’s been kind of a seachange in the growth of data science and data engineering roles,” he told Built In. “What that means is that AI engineers sort of have to have a pretty broad skill set. … Someone who is able to work across a lot of the different domains of the data science, engineering, AI and ML space.”
Important Skills for AI Engineers
While a comprehensive and firm knowledge of the various facets of artificial intelligence is important as an AI engineer, software engineering skills are also essential. Python, R, Java and C++ are among the most used languages in this space.
Probability and statistics are also important components of AI engineering, since machine learning models are based on mathematical principles. Plus, a firm grasp on concepts like statistical significance helps if an AI engineer needs to determine the validity and accuracy of a given model.
AI Engineer Required Skills
- Artificial intelligence knowledge
- Math, statistics and probability
- Critical thinking
“AI is much more than algorithm development,” Seligson said. “It’s everything that goes into the entire data lifecycle, from taking it from a system all the way through the output — whether it’s training a robot to move, or training a system to detect anomalies, or a drone to fly over some area.”
Being an AI engineer also requires some soft skills, particularly as it relates to problem solving, communication and critical thinking. Seligson, whose educational background is in musical composition and religious studies, says he often has to lean on his non-technical background as an AI engineer at Entanglement, particularly when it comes to communication.
“Engineers respect good technical work, and being able to articulate yourself as a technician in ways that are clear, concise [and] logical,” Seligson said. “I’ve found that a lot of my old skills didn’t necessarily have to get thrown out the window just because I was working in STEM.”
How to Become an AI Engineer
Indeed, while AI engineers commonly have backgrounds in computer science and software engineering, that isn’t necessarily a prerequisite for landing a job in this field. Rather, becoming a successful AI engineer really comes down to an individual person’s willingness to learn, their passion for the industry and the opportunities they create for themselves.
Tips for Becoming an AI Engineer
- Learn the technical skills required
- Find a mentor to help narrow your interests
- Learn through experience
Learn the Technical Skills Required
Working as an AI engineer requires quite a bit of technical know-how, particularly when it comes to programming and mathematics, as well as AI algorithms and how to implement them with frameworks. Common machine learning algorithms include linear regression and decision trees, while common deep learning algorithms include recurrent neural networks and generative adversarial networks. Some common AI frameworks include Theano, TensorFlow and PyTorch.
“I think the most important thing is self-learning because nobody is going to teach you all of this,” CVS Health’s Haque said. “Anybody who has done high school and has good programming [skills] can start.”
There are tons of educational resources available online on sites like Codecademy, Simplilearn and even here on Built In. Of course, keeping up to date with the ways AI is evolving is vital, but knowing the fundamentals is just as important.
“It’s always good to make sure you know the fundamentals of all of this because those things aren’t going to change. If you know the fundamentals as everything keeps changing, you’ll still be able to understand what’s going on,” Abdullah, of Duke University, said. “Just remember that you don’t have to know everything, because it’s not possible.”
Find a Mentor to Help Narrow Your Interests
Learning the ins and outs of AI on one’s own can get “overwhelming,” Abdullah said, especially if you’re in the really early stages of career development and want to narrow your interests down. She suggests finding a mentor who actively works in the industry, so they can give you a clear idea of what working in the space is actually like, and even help narrow down an area of focus.
“You might or might not narrow it down to one. But let’s say you narrow it down to three areas of work, then that’s better than before,” she said. Once your area of interest is narrowed a bit more, prioritize learning the specific tools or technology required. That way, you can avoid getting too overwhelmed by the sheer magnitude of the artificial intelligence space.
Learn Through Experience
After establishing a good foundation of industry knowledge, and narrowing down your interests, the next step toward establishing a career in AI engineering is striking a balance between “the learning part and the doing part,” Abdullah said. It’s great to continue learning, but a textbook or class can only teach you so much. Sometimes, the best way to learn and get comfortable with this technology is to jump into it and experiment.
Seligson calls this “repetitions.”
“It’s one thing to have talent or problem-solving ability. Even if you have all that, you still need to get a lot of repetitions as an engineer in order to notice and kind of evaluate,” he said, likening it to playing a sport. “Even if you’re really fast or really tall or really strong, if you don’t know how to play and if you haven’t seen the field a lot, you might go out there and fumble the ball the first few times. But the more you’re out there, the more you realize, ‘OK, this is kind of the flow of the game.’”
Being an AI Engineer Has Many Rewards…
Working as an AI engineer can be quite a rewarding career. And there are lots of reasons to enjoy the work it entails. Artificial intelligence is at the forefront of virtually every company’s growth strategy, so the folks working behind the scenes of this technology can have a real impact — both technologically and financially.
“The gratifying part about the job is being able to say, ‘Hey, I made a change to this model and it had a tangible revenue impact’,” Shandar said, adding that, in the work she does with LinkedIn’s ads, the changes she makes have the potential to make the company hundreds of thousands of dollars. “It’s one of those things that you can’t immediately conceptualize. But it’s still really gratifying because you can say, ‘I made a model, or I made a change to a model, that actually made the company a lot of money.’”
… But It Also Comes With Challenges
That’s not to say that this technology is any walk in the park, though. AI technology is constantly evolving and gaining complexity. And while working with something so experimental and new can be exciting and rewarding, it can also be challenging.
AI Engineering Challenges
- Building an AI model can be time-consuming and tedious, and can end in failure.
- Sometimes there’s a disconnect between the lofty goals of managers and executives, and what their AI engineers are actually capable of doing given the time constraints and resources available.
- Artificial intelligence tends to be misunderstood, and its capabilities can be overestimated.
- If an organization has low-quality data, AI engineers have to do a lot more work to massage the data to make it compatible with a model.
“You will spend a lot of time running experiments, poring over models, and sometimes these experiments fail. Or sometimes it feels like you have nothing to show for it because you tried a bunch of things and they didn’t work the way you wanted them to,” Shandar continued. “Sometimes it gets frustrating.”
And because artificial intelligence requires quite a bit of expertise and know-how, there can sometimes be a disconnect between AI engineers and the people in charge. Generally, engineers rely on guidelines for the work they’re doing so that they have a concrete goal or mission for the otherwise very technical work that they do. For instance, if a system must be able to sort through 50 million data points in a certain amount of seconds with a certain amount of accuracy.
Managers and executives focus on what the goal is and what a given product will provide to the larger market. Meanwhile, engineers focus more on “OK, how do we actually do that thing?” Seligson explained. “Managers and executives can expect kind of crazy things from what they call and think of as AI, wherein it can’t really do those things with the resources that they allocate.”
Not every business function or problem requires artificial intelligence. Shandar says she’s been asked “many times” over the course of her career to implement machine learning solutions for things that don’t necessarily need machine learning. Instead, she suggests starting with a “solid statistical foundation” — something simple, and rules-based — and then going from there. This provides more of an understanding of the data, which is crucial if you eventually want to build a machine learning model that actually serves its purpose.
“[People] think that, if they have data, they can just throw it into a machine learning model, and it’s going to work, and you’re going to have millions of dollars. But you need to really understand your data,” Shandar said. “A lot of people try to build production machine learning models with, pardon my French, shitty data. Or data that comes in every six months, or is poorly annotated, or is filled with zeros or a lot of missing values. Or just straight up in formats that are not amenable to machine learning.”
This creates more work for the AI engineers, who then have to massage the data in order to get it compatible with a machine learning model.
The Future of AI Engineering
It’s important to remember the space that this job exists in. Just as artificial intelligence is rapidly evolving and expanding, so too are the job descriptions of the people who work with this technology.
Consider all of the advancements made in AI within just the decade: Amazon released its now-ubiquitous virtual home smart device Alexa; Google made the first self-driving car to pass a state driver’s test; Hanson Robotics created the first robot capable of facial recognition, verbal communication and facial expression; and Open AI released a natural language processing model called GPT-3, one of the closest things we have to artificial general intelligence. In just a few short years, artificial intelligence has revolutionized everything from healthcare to manufacturing to manufacturing to art.
“I definitely don’t think it’ll look the same, because it didn’t look the same five years ago.”
Looking ahead, the tech industry will continue to push the boundaries of what artificial intelligence is capable of, which will inevitably redefine the daily work of AI engineers — particularly when it comes to matters of bias and privacy, which are growing concerns with this technology.
“AI is as good as we make it. And if we are not thoughtful or conscientious about how we make our models, then our AI will be just as unfair and unjust as we are,” Shandar said. “I definitely don’t think it’ll look the same, because it didn’t look the same five years ago.”
As for exactly how much change occurs, Seligson says it’s a question of “revolutionary versus evolutionary change.”
“I don’t think we’re going to necessarily see a huge revolutionary change in the way people do AI,” he said. But he does think the realities of what it means to be an AI engineer will change in the next few years, particularly as things like automated AI and machine learning become more prevalent. “Evolutionary changes end up kind of revolutionizing in the long term.”
But no matter what direction AI takes us in the next five years, 10 years and beyond, AI engineers are going to be right at the center of it.
“It’s cutting edge, it’s impactful, it’s so cool,” Haque said. “I feel extremely powerful, in a positive way, about how I can change my and others’ lives with it.”