Last year, Clara Shih, the founder of Hearsay Systems, was taking part in a routine visit with one of her insurance industry clients in San Francisco. Hearsay works with more than 150,000 financial and insurance advisers, providing them with artificial intelligence-driven tools to improve client relationships and workflow processes. This particular visit was to a small firm with four employees--two of whom did nothing but follow up on delinquent payments and policy renewals. The approach, involving numerous phone calls that were never returned, was not only unproductive, but also laborious and tedious.
During the visit, Shih and her team demonstrated a new A.I.-driven tool that digitizes manual customer-outreach processes by sending a text to dozens of customers reminding them of overdue bills, instead of calling each one. As they explained the tool's uses, one of the advisers, a middle-aged man, started to cry. For a moment, Shih and her colleagues feared the adviser thought their A.I. product was going to put him out of a job. After all, that's the knee-jerk reaction many workers have when confronted with machine learning. But his tears were for a different reason. "This is amazing," Shih recounts him saying. "What have I been wasting my time doing for the past 20 years?"
Machine learning--whether it be robotic process automation, advanced data analytics, or A.I.--will undoubtedly reshape the workplace. The question of how many jobs will be lost and created is the subject of much speculation. According to the World Economic Forum's "The Future of Jobs 2018" report, by 2025 more than half of the total time spent on labor will be handled by machines. Nearly 50 percent of companies expect that by 2022, automation will lead to some reduction in their full-time staff, while 38 percent surveyed expect to grow their workforces to new productivity-enhancing roles. Another recent study by PwC estimates that in the United Kingdom, seven million existing jobs could be lost to machines over the next 20 years, but another 7.2 million could be created.
The transition to a workplace where humans and machines will need to productively co-exist could make or break a business. As company leaders plan for the future, they will have to consider machine learning's impact on everything from productivity to skills to morale and culture. And they will have to learn how to lead a business that may have as many intelligent machines as people.
"A.I. doesn't just offer to make the existing things we do better, more efficient, and cheaper. It also has the potential to help us do things that would have been inconceivable before," says Dave Coplin, author of The Rise of the Humans and CEO of the Envisioners, a futurist consultancy. "But unless humans understand how to make the best of it, we risk belittling the potential it offers."
Here's what we do know: The more robotic minds there are in the workplace, the more companies will want workers who don't think robotically. "We need to make sure that humans develop complementary, not competing, skills with the technology," says Coplin. "We wouldn't try to outcalculate Excel, and we don't try to remember more facts than Google. Instead, we need to consider: What are the fundamentally human skills that the computers will be unable to replicate for decades to come?"
Machine learning can do many tasks far better than humans, but it still takes humans to interpret its work, and apply the results in ways that are strategic, empathetic, and creative. The key, says Shih, is realizing that the machine is just one resource humans can call upon, and that they, not the machine, have the skill set that makes the relationship truly useful. "It's about being open-minded and having the ability to delegate the right task to the machine," Shih says.
The best way to ensure that approach is to establish what those in the industry call a "humans-in-the-loop" relationship. Let the algorithm do its thing, with people overseeing and refining it. "Machine learning is hard to get 100 percent right," says Shih, but with such a process in place, "you don't have to be perfect. The human intervenes in the process and the algorithm learns."
She points to a recent rollout of a new Hearsay service that provides automated quick text responses for advisers and insurance agents to send their clients. When the eight-year-old company first introduced the service, the algorithm came up with a few eyebrow-raising suggestions. In one case, it suggested an adviser wish his client happy birthday. When the client responded, "Thanks for the kind thoughts," the algorithm replied, "Sounds good to me!" leaving the client thinking the adviser wasn't paying attention or was slightly unhinged. (Google's new automated email reply service has suffered from even more bizarre response fails in recent months.) As Hearsay's human employees and machine learning refined the algorithm, they were able to smooth out the rough edges around the message prompts and create a set of more appropriate responses.
The only way to achieve this type of human-machine symbiosis, though, is if humans don't enter this new relationship with fear--"the worst decision-making sentiment to have," observes Kristian Hammond, the co-founder of Narrative Science, a company that uses A.I. to create natural-language reporting out of raw data and statistics. When interactions are driven by fear, the emphasis shifts to the technology, rather than the business need for using the technology. Hammond recommends assembling a team comprising both data architects and those in strategic business roles. "You want A.I. experts to be part of a broader initiative that speaks to who you want to be as a company and how A.I. can shape the business," he says.
Learning to Trust the Machine
If humans are going to regard machines as partners rather than as adversaries, they need to have faith in the work being produced. To win your employees' trust, Coplin recommends taking an incremental approach to A.I. "Apply the algorithm to a small portion of the overall workload to give humans time to see how the algorithm works, and to build trust that the outcome is what was expected," he says.
One example he cites is a new algorithm-based table-booking system a large restaurant chain implemented. Initially, individual restaurant managers were skeptical that an algorithm in the cloud could do a better job at managing the tables than they could. To alleviate their concerns, the company agreed to allocate just a small portion of available tables to the algorithm, and if the managers were happy with the results, more tables would be added. After starting with a pool of just 10 percent of available tables, the managers quickly realized that not only did the algorithm do a great job, but it also freed them up to do more useful tasks.
Stephen Ufford is the co-founder and CEO of Trulioo, an A.I.-powered global verification service to support financial services' anti-money-laundering monitoring. Traditionally, this important area of banking security was handled by human workers, but increased computing power and the sheer volume of digital data now being produced have left them outnumbered and outgunned by criminal gangs. Now, algorithms like Trulioo's scan millions of transactions at a scale no human could manage and are trained (by humans) to spot potential fraud or block suspicious individuals.
When dealing with something as sensitive as identifying fraudulent transactions, Trulioo's employees had to be certain the algorithm they had built wasn't showing any bias in its decision making or making rogue recommendations. "Learning to trust A.I. isn't that different from when you employ a new babysitter," Ufford says. "To start off, you watch what they are doing via the nanny cam, but after a while you start to relax as you gain confidence in how they work."
The End of Work as We Know It?
The growth of machine learning within companies ultimately raises the type of existential questions that executives don't like to confront: How many of us will actually work with machines in the future?
The reality is that change is inevitable, so companies need to work on ensuring a soft landing for those currently doing the type of functional or repetitive tasks that automation can do better. For some, that may mean retraining or upskilling staff to get the most out of their institutional knowledge and experience. Others, however, will inevitably find employees automated out of a job, just like those at Foxconn, which in 2016 replaced 60,000 workers with robots.
Yet, as we look to the long term, could it be that the fears of automation-induced mass unemployment have been overblown? After all, the next generation of workers--the Alexa generation, for want of a better term--will already be used to living with and learning from machines. And these new workers are showing signs that they are more motivated by the experiences, freedom, and creativity they can have in their work than by conventional incentives.
"Two of my grandparents worked in factories, yet my grandfather also loved to paint," says Ufford. "Shouldn't work harness that creativity rather than crush it?" It could turn out that machines, in the end, are the mass creativity catalyst we've all been waiting for.