Artificial intelligence is coming to the masses.
It wasn’t long ago that AI was the exclusive domain of data scientists. But now, industries as diverse as retailing, manufacturing, finance and insurance are taking advantage of new products that make it much easier for businesses to create AI tools specific to their needs. By plugging data into standardized AI templates, auditors, analysts and actuaries who lack specialized AI training are able to identify sales prospects, spot risks and fraud, and boost organizational efficiency.
These templates still require tech-savvy users and can take weeks of model training, testing and validation. But the enabling of “citizen data scientists”—an industry term for professionals who have developed competence in AI using automated tools but who aren’t trained specialists in statistics or analytics—also translates to cost savings for businesses that can dispense with data scientists who would otherwise spend several weeks coding a custom program.
The growing adoption of these products follows a classic pattern in tech evolution, where “technology turns every employee into a programmer,” says Craig Le Clair, a principal analyst at Forrester Research. For example, tasks once viewed as too complicated for the layperson, such as sending an attachment or cropping a photo, have become effortless as digital-productivity tools kept improving and getting easier to use.
Thanks to easy-to-use interfaces, programs for these AI templates—known as automated machine learning, or automated ML—are even being used by data scientists themselves.
“This is the future,” says Chirag Dekate, a research director at research and advisory firm GartnerInc. who specializes in AI and quantum computing. “Automated ML eases the pain of data science.” Gartner recently predicted that by next year, citizen data scientists will surpass data-science pros in handling the majority of automated ML tasks in business and industry. By 2024, the research firm says, automated ML will become so commonplace that the adoption of data science and machine learning will no longer be held back by the chronic shortage of data scientists.
Because AI involves vast amounts of sensitive data and computing power, some experts fear that use of automated ML by less-well-trained users could lead to errors and bias that could prove embarrassing and costly. Indeed, skeptics caution that automated ML will require careful supervision and guidance from a data scientist, AI ethicist or other third party.
Those who use the technology developed by DataRobot Inc., a Boston-based company, are mostly data engineers, software engineers and business analysts, says Greg Michaelson, DataRobot’s chief success officer. According to Mr. Michaelson, success in automated ML requires only basic competence with data but a deep understanding of the business problems that AI is being harnessed to solve.
DataRobot’s customers are discouraged from operating their programs with complete independence. The company’s software license requires clients to work with a dedicated DataRobot field engineer and a data scientist in the first year, when the automated ML functions are set up and the algorithms are deployed.DataRobot workshops, training, site visits and virtual office hours are required for new customers.
The citizen data scientist is sometimes described as a technological quick study and an astute business strategist.
Jeff Dwyer, director of engineering at Boston-based ezCater Inc., an online marketplace that connects catering companies with businesses looking for catering services, implemented an automated ML program about two years ago. The company, with 600 employees and more than 60,000 caterers on its platform, was interested in using AI to segment business clients according to sales potential to prioritize marketing and retention efforts.
To prepare for the task, Mr. Dwyer read a book about statistical learning and watched an online Coursera educational video about machine learning.
He used an AI program developed by Boston-based RapidMiner Inc. and launched it after two months of testing and validation.
“It’s not pure magic,” Mr. Dwyer says. “You do need to spend some time with it.”
But he says the technology discovered intriguing patterns that had escaped human detection. For instance, one clue that a customer will be a heavy spender on catering services is a tendency to schedule the first catered meal of the week for a Tuesday event; what’s more, these businesses tend to place their Tuesday catering orders days ahead, not one day before the event.
“It’s hard to envision running the business without it,” Mr. Dwyer now says of automated ML. “It’s giving us a predictive value of our customers.”