Demand for machine learning engineers (already in short supply) is high and only expected to grow as the complexity of, and access to, machine learning increases.
Machine learning has changed rapidly over the past few years. Within the tech world, the bounds of machine learning are constantly being pushed. The complexity of machine learning models and systems engineering has increased as more applications demand real-time or near real-time inferences. At the same time, so-called off-the-shelf machine learning tools have become more available in the workplace. Both of these dynamics are driving demand for machine learning engineers.
“There’s a tremendous demand,” said Marshall Choy, vice president of product at Samba Nova Systems, an AI innovation company. “[And] as far as I can see, there is no end to that demand spike that we’re seeing.”
Caio Soares, group manager for core AI at Intuit, described the demand for machine learning engineers as a steep ascent, and it’s a trend that’s been going on for a few years now.
For more mature machine learning companies — large tech or fintech companies, in particular — this demand, according to Soares, is due to the increase in complexity of machine learning solutions and the constraints being put on the processes, such as the speed of data that needs to get to a model. For example, Soares pointed out that security risk and fraud assessment tools that use machine learning today need to rapidly process inferences from massive, dynamic data sources.
“If you want something to be computed in the 10s of milliseconds, and it’s a very complex, very deep neural network, that’s going to require deeper engineering in order to be able to adhere to those constraints,” he said. Companies doing machine learning work of this sort need machine learning engineers who can fine-tune the hardware to meet the demands of their machine learning efforts.
“Chances are the machine learning engineers in that company are working more in advanced technology solutions of how to scale AI; how to do things like train neural networks that have billions or trillions of records, but also incredibly complex structures,” Soares said.
As the complexity of machine learning increases, the demand for data scientists also increases, according to Mike Roberts, vice president of AI and machine learning at Hypergiant, an enterprise AI company. This is indeed the case, according to the Bureau of Labor Statistics, which projects that demand for data scientists will increase much faster than average for other occupations, with 22 percent growth by 2030. And, in Robert’s experience, when data scientists are needed, so are machine learning engineers.
“ML engineers are really kind of a support squad for the data science team,” he said.
When a company is applying machine learning at a scale to the point where it is hiring data scientists, there is a division of labor that tends to happen, according to Roberts. While a data scientist generally focuses on the theory and practice of statistical data analysis and machine learning algorithms, the demands of the machine learning work involve “some pretty esoteric aspects of infrastructure,” he said.
Those infrastructure demands require someone with more of a software engineering and DevOps background with hands-on experience creating software and managing infrastructure in their daily lives — in other words, machine learning engineers.
While bigger or more innovative companies have been riding on — or close to — machine learning’s cutting edge, not every company has the resources for, or even needs to be, at that level. These companies tend to reach for as-a-service solutions to their machine learning needs.
But there is a spectrum of as-a-service solutions, Choy said. And where a company falls on that spectrum will impact their demand for machine learning engineers.
At one end of the spectrum is infrastructure-as-a-service, which offloads the need for on-site servers, storage and the networking know-how to string it all together to get it running in the operating system. Cloud offerings cover this. But trying to do machine learning at this point along the as-a-service spectrum leaves a great deal of work to the end user, said Choy.
“You still have to then build a platform on top of that — whether that be a database platform or a machine learning platform — and then you’ve got to build the applications on top of that,” he said. “The platform services take things one step higher in that they provide very broad and generic machine learning capabilities, but not specific to the applications, which is the ultimate goal of the business.”
So both of these points along the as-a-service spectrum still require the skills of a machine learning engineer or, failing that, a data analyst or data engineer with a lot of machine learning skills.
In addition to the growing complexity of machine learning in the tech world and its increasing availability to businesses at large — direct drivers of demand — experts Built In spoke to pointed to the number of people who quit their jobs during the pandemic as an indirect driver of demand, by way of the increased interest in automation.
“With the Great Resignation and demand for better quality jobs, automation will accelerate immensely to remove monotonous processes and ensure employees have high-quality work experiences with diverse and meaningful tasks,” said Eric Johnson, chief information officer at AI-driven experience management company Momentive, said.
Intelligent process automation is largely powered by machine learning and is expected to be a growing trend. In the 2020 McKinsey Global Survey, 66 percent of 1,179 companies polled said their organizations were at least piloting the automation of business processes in one or more business units or functions. This compares to 57 percent who said the same in a 2018 survey. Forrester similarly expects increased automation in 2022 as the changes forced on companies by the pandemic become part of business as usual.
More automation means more demand for machine learning skills, according to Scott Turman — CEO of BrightRay Publishing, who has a background in software engineering and has experience with companies’ early machine learning adoption efforts. He described the dynamic as a cycle. It starts with a company’s C-suite looking at process automation as a way to cut costs or address a lack of personnel. After a failed attempt at a custom solution, or after judging the costs associated would be too high, the company turns to an off-the-shelf solution. But applying a general solution to the company’s specific needs still requires someone with fairly advanced machine learning skills.
But those skills are hard to find. There aren’t enough machine learning engineers out there.
“There is a significant shortage of this type of talent in practically every metro area,” said Jay Denton, chief labor market analyst for LaborIQ by ThinkWhy, a compensation and labor market analytics software company.
Machine learning’s relative newness, rapid maturation and spread into business at large are causing what could be called pipeline issues, according to Choy. The academic qualifications for machine learning engineers — which usually involve at least a master’s degree if not a Ph.D. — can take at least five years. But academic programs focused on machine learning are hard to find, Choy said.
Luckily, bootcamps for machine learning have proliferated to help people acquire machine learning skills. But demand is still outpacing supply because it’s a “tremendously difficult combination of skill sets to find,” Soares said. Machine learning is just too new.
“We’re in the infancy stages of this machine learning build out,” said Choy. He compared the current situation in machine learning to what he experienced as a computer programmer and software developer in the early nineties; there were no courses to take and the people working in those spaces had to teach themselves the technologies in general and apply them specifically to their company’s needs.
“And that’s really what we see right now,” he said.
While the current demand for machine learning engineers is strong and growing, the rapid maturation of the field will mean the nature of the demand will change. Both Soares and Choy voiced similar perspectives that machine learning and AI skills will become more widespread and diffused in the future, though this will look different depending on where in the innovation spectrum one looks.
Choy sees machine learning tools becoming more accessible and easier for more people across different industries and roles to use.
“In the future, as the as-a-service type offerings mature and simplify, the end user may no longer be a machine learning engineer, but a business analyst — somebody in the line of business as opposed to somebody in a engineering and innovation research type of group, as we commonly see today.” he said.
This diffusion of machine learning skills would help streamline organizations in Choy’s view. It would allow individual businesses to create applications that suit their needs and goals faster. It would also “enable them to derive more velocity and competitive advantage relative to the field and in their chosen area of business,” he said.
The situation is similar within the tech world as far as Soares sees it.
“I think long term — especially with the advent of some of the off-the-shelf tools, the black box tools — machine learning will become an expected part of a software engineer,” Soares said. This will include knowing and understanding things like what training a model means, how to treat data, how to deploy and take advantage of an AI model and how to use techniques like natural language processing and so on.
“I think [AI and machine learning skills] will just become another tool in [a software engineer’s] tool belt,” he said. This shift won’t do away with machine learning engineers, but instead, move them “higher up in the stack of more advanced more complex types of modeling techniques and approaches.”