The human side of implementing AI

The issues holding back AI adoption aren’t technical. Here’s how to navigate unrealistic expectations, change management issues and employees leery of AI job loss.

The human side of implementing AI
Thinkstock

What's keeping the company where you work from implementing AI?

Larry Lefkowitz, chief scientist for AI practice at tech consultancy Sapient, says only 20 percent of businesses use AI at scale. “AI adoption is in its infancy,” he explains. “Clearly AI is hot ... but it's also clear that it's not being adapted and adopted as quickly as we'd like.” The issues that hold implementation back, though, often aren’t technical — they’re human: Unrealistic expectations from the C-suite, difficulties in change management, uncooperative employees. So how can IT make onboarding artificial intelligence easier for people?

Lefkowitz says some people “are confused by what [AI] means. They don't understand what AI they should use or how they can use it best.” Michael Jabbara, senior director of global product for Visa, agrees. AI is so hot right now that Jabbara’s biggest challenge is convincing management that it can't solve everything. He compares public perception to “the father from My Big Fat Greek Wedding [who] believes the solution to any malady you have is a little Windex.” From AI that fights speeding tickets to AI-driven sexbots, trendy applications of the tech have given laypeople the impression that it can do anything. “So from a business perspective,” Jabbara continues, management thinks, “‘Oh, you have a problem? Just spray a little AI on it.’”

Laser focus

Unfortunately, Windex can’t fix everything and neither can AI. That’s why Jabbara says “to start with a business objective that you're trying to accomplish, and then set up a holistic strategy around how you're going to get there: How are you going to align your systems, your processes, your people.”

At Visa, for example, this might mean a program to analyze large purchases that don’t fit individual buying patterns. Instead of flagging them as fraud, contextually-aware AI would realize the cardholder’s just Cyber Monday shopping. This type of project would directly benefit a focused area of business operations — fraud detection — while delivering value to customers and the company at-large.

Starting with a single, focused task also keeps total cost of ownership low. “Start small and prove some value in a small area so that way there's not a big investment from the business,” says Steve Meester, senior vice president at AIG. Partnering with vendors can also minimize management’s sticker shock, he continues: “Rather than invest in [in-house tech] upfront, not knowing whether it's going to prove out or not, we can go with an external vendor. Essentially, you're building a business case for bringing that capability in-house and developing the ability in-house.”

The network effect

Another way to gain support is through networking — the human kind. After pinpointing the best problem to solve, Jabbara says, “Develop a couple of [proof of concepts]. ... Then start to socialize that within the organization to increase the level of awareness and knowledge about what AI is and what it can do. ... Creat[e] a network for yourself so when you come forward you're surrounded.”

Socialization is major, Meester agrees. He recommends identifying an early adopter — advice that’s often given but rarely easy. Under Jabbara’s approach, though, that internal champion could be you. Just make sure you communicate in a way that less tech-savvy staff can understand. “Put everything in the language of the business,” Meester says. Since you’re in IT, he continues, “you'll think that you're putting it into the language of the business and you'll go talk to the business leaders and then you'll realize, no, this is still not in the language that they understand. But you have to get your point across in the language that they understand, the language of the business.”

Language, after all, is why some non-techies worry when they hear “AI”: The words we use to describe artificial intelligence — “innovative” or “revolutionary,” for example — may be positive by definition, but may still negatively impact employee perception. After all, revolutions are wars. And to those who lack confidence in their work, “innovation” can sound like a fancy way of telling them they’re not good enough.

Right now, says Lefkowitz, “Solutions, especially those in automation, are sold on the premise of ‘I'm going to reduce your headcount and save you money.’ At the same time, when we get out there in the field, we usually go to people and tell them, ‘Don't worry. You're not going to be replaced. Simply the old drudgery is going to go away and you'll be able to do fun things.’” It’s one message in, another message out — leaving skeptical employees unable to trust technology in between.

The fix, says Ward Eldred, solution architect at AI computing company NVIDIA, is to keep staff updated when projects you’re developing “don't eliminate headcount, [but] change what that headcount is doing.” This goes back to the need for colleagues to better understand what AI is and what it isn’t. Eldred emphasizes that while machine learning has made fantastic gains, ML still can’t “completely replace the human in all instances.” Acknowledging that you realize how difficult an employee’s job is may help him understand why he won’t be replaced by a computer any time soon. Eldred says, “We actually find that people — once they understand what the implications are going to be — they're actually very receptive.”

In this, change management for AI implementation isn’t really that different than for other initiatives. “It is incumbent to involve [users] early,” says Dave Parsin, vice president at natural language interaction provider Artificial Solutions. “Involve them as being part of the solution.”

Also remember that even if a project optimizes processes for only one department, other business units still need to be looped in. No developer, Eldred adds, wants to be in the middle of IT and architecture integration just to have “the security guy walk in the room and say, ‘What are you doing?’” Make sure no one involved with the project or its data gets left out.

This, Jabbara says, goes back to his initial suggested framework of aligning systems, processes, and people. When all three come together, he explains, you can finally “take a look at where AI can help advance that journey further along. And that's how you're going to be able to really successfully get some changeable results back — because, ultimately, AI is really more of a means to an end than an end itself.” The key to solving AI’s human problems, he adds, is making sure its adoption is “part of a business’s overall strategy, not just an AI strategy by itself.”

Related artificial intelligence articles:

NEW! Download the Spring 2018 digital edition of CIO magazine