Wondering why a human CSR can’t just handle this conversation? “People interpret tones of voice differently, so they respond differently to customers,” explains IBM consultant Aman Kochhar. By contrast, he adds, “A.I. is not subjective. So it’s much more consistent.”

Kochhar has been learning to apply artificial intelligence to business problems since last December, when he started taking A.I. courses as part of the first phase of a gigantic new training push inside IBM. Called AI Skills Academy (AISA), the program is designed to do two things. First, it teaches employees about integrating A.I. into their own jobs within the company, from creating marketing apps to improving supply chain efficiency. At the same time, AISA educates IBMers in consulting, sales, operations, and elsewhere how to collaborate with clients to use A.I. in their businesses, too. Divided into two tracks—one for techies (software developers, engineers, research scientists) and one for everybody else—the curriculum has four levels, from basic to expert.

More than 2,200 IBM staffers have started the training since it launched last year, and IBM expects at least 4,000 graduates of all four levels in 2019. But, says IBM vice president for talent Obed Louissant, that’s just for openers: “All of our employees will eventually be trained in A.I.” Moreover, AISA continually adds new content. In the works right now: New courses on making use of A.I. in project management and general management roles.

In one sense, it’s only logical that IBM is investing big chunks of its $500 million annual training budget in AISA. After all, “we build these A.I. technologies,” notes Louissant. “So we have a responsibility to teach people how to use them, both inside and outside the company.”

Okay, but AISA also clearly does something else — to wit, it makes IBM’s 350,000 employees worldwide a lot more desirable to other employers. As more companies rely more heavily on data analytics, and more jobs call for a working knowledge of A.I., Gartner predicts 2.3 million new roles worldwide that will require these skills by the end of next year.

For IBM, AISA is a calculated risk. On the one hand, the company has no real choice but to train its workforce in A.I. But on the other hand, helping employees develop precisely the skills most in demand in the outside world right now seems dicey. “We did think a lot about this as we developed the program,” Louissant says, adding wryly, “We were concerned from the outset about whether we’d be creating a public service.”

It may work out that way, but for now, Louissant thinks most graduates of IBM’s program will want to stick around. He points to the fact that, among the roughly 800 people who have already completed AISA training—and who are therefore even more marketable than they were a year ago—attrition, so far, is lower than for IBM’s workforce overall.

It’s early days yet, of course, but that tiny attrition rate may be a reflection of what employees said, in detailed surveys, about what motivates and engages them. Even more than money, which of course competitors can offer too, IBMers say they’re “most interested in keeping up with the cutting edge in technology and continually learning new skills,” says Louissant. “So offering them new training is a retention strategy.” In this era of persistent (and, it seems, multiplying) skills gaps, that’s a notion worth pondering.

Anne Fisher is a career expert and advice columnist who writes “Work It Out,” Fortune’s guide to working and living in the 21st century. Each week, she’ll answer your most challenging career questions. Have one? Ask her on Twitter or email her at workitout@fortune.com.