Low-Code Machine Learning: Good for Workers, Bad for Companies?

Microsoft’s new tool, part of its Azure cloud platform, wants to make machine learning (ML) a drag-and-drop proposition.

In theory, this is a good thing for employees who aren’t mathematically inclined, or who have no background in programming or machine learning; while every organization would like its teams to become machine-learning experts, that’s simply not feasible. Microsoft’s platform allows employees to select a dataset, select a value to predict, and let the algorithms do their work in the background.

In a move that will no doubt make data scientists and analysts sigh with relief, the tool doesn’t obfuscate its processes behind a shiny-looking dashboard; experts will have the ability to study the algorithms producing the output. 

Microsoft is big on “visual coding,” which is no doubt vital to its broader ambitions as a software provider for corporations of all shapes and sizes. Last year, for instance, it unleashed an update to PowerApps, its low-code building platform, that allowed ordinary employees to put together mobile apps via drag-and-drop and some simple coding. At the time, the company liked to highlight stories of rank-and-file workers using PowerApps to radically change how their organizations worked—for instance, a security professional at Heathrow Airport used the platform to build apps that quickened his colleagues’ workflows (earning himself a promotion in the process). 

But low-code (and no code) has its limits. For one thing, sysadmins and other high-level tech pros often dislike it when employees begin churning out new apps and functions without much oversight and supervision. PowerApps came with an “App Checker” that acted as a front-line debugger, but ultimately it was Microsoft’s intent to create “controlled chaos” within organizations by allowing employees to take a lot of initiative in terms of designing and releasing apps. Not all sysadmins, CTOs, and product leads are onboard with that. 

With artificial intelligence (A.I.) and machine learning, the potential downsides of low-code platforms are higher. It’s not so much a question of an A.I. or ML algorithm unleashing tons of chaos—nobody’s going to create a low-code Skynet anytime soon. Rather, these algorithms are delicate and often temperamental things, and it’s easy to picture Tad from Accounting getting increasingly frustrated over why his machine-learning tool doesn’t work like he thinks it should, or why the output seems so messed up. Data scientists could end up spending as much time fixing their colleagues’ work as they could on their own tasks.

Although A.I. and ML experts remain much in demand (those with substantial experience and a track record of successful projects can easily earn six-figure salaries—if not higher), the rise of no- and low-code tools (combined with companies instituting at least rudimentary A.I. education for workers) could boost the supply of A.I.-knowledgeable workers, which in turn could make it difficult for newer graduates in the subtle arts of A.I. to land desirable gigs. In this scenario, those with specialized skills and experience would end up having the best chance on the market.

That’s a far-off hypothetical, though; for the moment, A.I. and ML remain complex and delicate disciplines, and it seems unlikely that any single no-code tool will disrupt that paradigm for the time being.