There’s a big temptation for businesses to use artificial intelligence to shave off time and money wherever they can. According to experts, that’s not the smartest use of the technology.
A few months into my time as an insurance claims adjuster, a customer called and said he thought his house was breaking apart. He’d heard what sounded like wood beams snapping in the basement, so he went downstairs and crept into the crawl space to investigate.
While he was lying in the dark surrounded by cement, he told me, he started to panic. It brought him back to years earlier when, as a firefighter with the New York City Fire Department, he served as a first responder on September 11, crawling through the giant blocks of brick and mortar that had collapsed hours earlier.
This revelation came 10 or 15 minutes into our conversation, after I’d gathered his basic information and logged the details of the case. He’d been friendly, even lighthearted (if a little excitable) for the first part of it. But recalling his moment of anxiety, his voice quivered.
To an extent, I was able to relate. My father was a captain in the FDNY on September 11. He, too, spent the hours, days, and weeks that followed at Ground Zero, digging around in the rubble and coming home covered in dust. I know that he, too, has painful memories stored away in the deep parts of his mind.
I relayed this to the man on the other line. We talked for a few minutes about that day and its aftermath, and then I brought the conversation back to what was going to happen with his claim. He asked if it was OK if he called back just to chat if he needed to.
A few hours later, he did just that. I don’t remember exactly what we talked about, but I remember he was calm, he used my name a lot, and he thanked me.
These are the experiences I’d argue that artificial intelligence cannot fully replace. Or can it?
Case in point: Fukoku Mutual, an insurance firm in Japan, is replacing 34 claims adjusters with A.I. According to a press release from the company, the system, which uses the technology found in IBM’s Watson, will be able to study factors including the length of a hospital stay, procedures performed, and the patient’s medical history to determine a payout.
Benefits of using the system, according to the release, include improving operating efficiency by 30 percent. Japanese publication The Mainichi reports that it will save Fukoku Mutual $1.2 million (140 million yen) in wages annually.
Fukoku is just the latest example of a company testing the promises of artificial intelligence. It’s clear that A.I. is getting increasingly sophisticated at doing what humans do–but more efficiently and cheaply. What’s less clear is whether those gains trump the huge implications it would have for the future of work.
A.I. gets smarter
There’s a big temptation for businesses to use artificial intelligence to shave off time and money wherever they can, but experts say that’s not the smartest use of the technology.
“I actually think that’s the worst reason to automate things,” says Josh Sutton, global head of data and A.I. at marketing giant Publicis-Sapient. “Over time, it’s a losing proposition. It’s a race to the bottom.” Improving the customer experience and creating new revenue sources are much better applications of A.I., he says.
In the case of Fukoku Mutual, the more noble objective is getting customers a decision–and a payout–faster, which the company claims the system will accomplish. The company I worked for, which was one of the largest insurance providers in the U.S., made this the No. 1 priority, citing evidence that customer satisfaction is most closely related to the speed with which a payout is made.
But, as adjusters, we were also trained incessantly on our bedside manners. Someone who calls a home insurance firm is usually in a vulnerable place–and sometimes in full-blown crisis mode–so it was essential that a touch of humanity was included.
The question, then, is whether computers will ever have an emotional intelligence interchangeable with that of humans. “There’s no question in my mind that the technology is moving in that direction,” says David Schatsky, managing director and emerging technology analyst at Deloitte. “A.I. can already perceive and understand emotional cues, based on factors like tone of voice and word choice. And it’s getting better at projecting them back at users.”
So what does this mean for the work force? Until now, many of the jobs that have been displaced by machines are of the manual-labor kind: bots that fulfill orders in Amazon warehouses, for example, and machines that move products along an assembly line. The assumption has been that those jobs that require more training would be safe for some time. Arguably, that’s no longer the case.
“We’re starting to see the effects of technology automating cognitive work–things we used to think only people could do,” Schatsky says.
A 2015 Forrester analysis predicted automation would replace 25 percent of all job tasks by 2019. Losing some specific roles, like the country’s 1.6 million truck-driving jobs, seems like a foregone conclusion with the rise of self-driving cars. (Recently, an autonomous tractor trailer from Otto completed a 120-mile beer delivery.) Other occupations are using A.I. in tandem with people: Lawyers use software that can analyze cases and search for relevant past rulings; pharmaceutical firms use algorithms to aid in drug discovery.
Some experts argue that even humans in some of the highest-paying roles could become redundant. Last year, a team of British and American researchers fed an A.I. system the details of a series of court cases. The computer reached the same verdict as the judge 79 percent of the time. Other jobs, like those in the medical field, could eventually be replaced by faster, more accurate machines.
“Many of us in the A.I. field believe that physicians will be replaced long before nurses are replaced,” says Andrew Moore, dean of Carnegie Mellon’s School of Computer Science. “The parts of medical care to do more with interacting with the patients, making them comfortable, and communicating clearly with them are going to turn out to be the things that only humans can do well. The diagnostics, coming up with a theory as to what’s going on, or what a good subsequent test would be–those are the things that look very promising for automation. Compared with human experts, computers are doing a very good job.”
What happens next?
If it’s cost effective, it’s hard to imagine companies passing up the opportunity to take advantage of A.I. That’s especially true for positions like customer service reps, which require problem solving and a strong knowledge of the company’s operating standards. “The cost of both finding and retaining people with that combination of skills is pretty high,” Moore says.
At the insurance company where I worked, for example, training took anywhere from six to 12 weeks, and the employees turned over at a very high rate–sometimes just months after the program, which included a weeklong trip to company headquarters, was complete.
The ideal application for A.I., Moore says, is for handling routine interactions, while the more involved cases can get patched through to a human. It’s the ideal result for the customer, who generally doesn’t want a 20-minute phone call to determine a payout for a straightforward carpet replacement.
“If you can provide the experience to the consumer to get the information in their hands in a fast and seamless way, they don’t care about that loss of contact,” says Monte Zweben, founder of software firm Splice Machine and former head of NASA’s artificial intelligence branch. “Banks are a good example of an industry that has found that balance. For the very fast, transactional approaches, you can use the ATM, but they still have tellers for when you need some interaction.”
In customer-facing fields, automation might take over lower roles, so the position becomes more of an overseer than an actual doer. The key, then, will be for the A.I. system to be able to quickly recognize the situations in which humans should intervene. Until, that is, maybe even that won’t be necessary.
For some, the question is less about whether technology will advance to the point where it can react just like humans (it will, experts say), and more so about whether the benefits of using the technology outweigh the societal consequences. Losing the human touch would change the nature of work–and just about every transactional interaction. Widespread job losses would result. Anticipating such a future, Finland, France, and Canada have already approved pilot tests for government-provided universal basic income, something Elon Musk has said will be an inevitable necessity as A.I. spreads. These are real dilemmas companies and governments will need to consider.
“Some people believe that anything that can be automated should be automated,” Schatsky says. “I think it’s fair to say that’s not always the case.”
When I think back to that shaken firefighter, I can’t help but see his point.