The Promise of AI and Productivity

Artificial intelligence is supposed to trigger a wave of productivity not seen since Ford built the first assembly line. Translating this promise into real-world results will take time, expertise and a newly trained workforce.
AI productivity

Is artificial intelligence a harbinger of the next industrial revolution? It depends on who you ask.
While there is no question that AI and machine learning can transform many aspects of the way we do work, whether this technology will have the same disruptive impact on the workplace as the assembly line or the internet remains to be seen. “We like to hype new technologies as something that will change the world,” said Tom Petrocelli, analyst for Amalgam Insights, an AI consulting and strategy firm in Buffalo, New York. “But 90 percent of technology evolution comes incrementally.”

AI has been in development since the 1950s, when a group of academics at Dartmouth College began studying how to teach machines to learn. For the past six decades, researchers have been developing tools that enable computers to learn and adapt so they can eventually take over common workplace tasks. However, most business leaders are still unclear about how and where it will transform their operations. “They understand that it can bring a level of efficiency and mitigate risks,” said Keith Strier, EY’s Global and Americas advisory leader for artificial intelligence in Orange County, California. But they’ve spent their career treating every new technology as a new piece of software with very specific features and expectations. That’s now how AI works, he said. “This technology is applied, not installed, and it requires a process of fine tuning that never ends.”

Companies that can adapt to this new way of thinking can expect to generate significant productivity gains, along with other benefits — but getting there won’t easy.

From Months to Minutes

Unlike software, AI engines are not programmed to do a specific function. Rather, they are taught to understand how to a variety of tasks, and with every interaction they learn more and get better at the job. That’s what makes AI fundamentally different from any other form of technology, said Mark Purdy, managing director and chief economist for Accenture Research in London. “AI won’t just improve human productivity. It is a form of virtual labor in its own right.”

At a basic level, AI-driven chatbots use machine learning algorithms to respond to customer queries. The algorithms study historical customer interactions to find patterns in questions and whether the responses resulted in a positive outcome. That “knowledge” allows them to craft appropriate responses. The bots learn from every interaction, so they get better at responding to customer requests over time. “No other technology has been able to improve itself like that,” Purdy said.

This same model of machine-based learning is being applied to far more complex workforce tasks. AI algorithms are now used in law firms to scan millions of documents for specific data or patterns, in health care to review medical images for signs of cancer, and to identify compounds that could be developed into successful treatments. In these examples, the AI engines are able to finish these largely mundane tasks faster and with fewer errors than their human counterparts.

AI is also changing the way companies recruit, said Athena Karp, CEO and founder of HiredScore, a software company based in New York City. Her company uses AI algorithms to help clients rapidly screen applicant pools and uses natural language processing to review a candidate’s entire résumé. These technologies determine who will best fill the current role and reroutes others to positions that might be a better fit. “It reduces the time to fill the role and humanizes the process for candidates,” she said.

All of these applications of AI in workplace have clear — and in some cases astonishing — productivity gains. In one example, LawGeex, a legal contract review company, challenged 20 attorneys to compete against an AI engine to identify 30 legal issues in five nondisclosure agreements. The AI engine delivered more accurate results and was able to complete the challenge in 26 seconds, compared to one-to-three hours for the attorneys. In another example, researchers at the University of Manchester discovered a compound capable of fighting drug-resistant malaria while test-driving an AI engine to see if it could screen a library of compounds more rapidly than human scientists. The experiment proved the AI engine could scan thousands of compounds in a fraction of the time and hone-in on the ones with the most medical promise, thus cutting the time and cost of developing new drugs and reducing the risk of clinical failures.

Jobs Lost, Jobs Gained

AI’s ability to consume and review data in a fraction of the time brings obvious productivity advantages, particularly in areas where human workers simply do not have the capacity to review that quantity of content. Experts at the World Economic Forum liken advances in AI to the next “Industrial Revolution” using digital tools to automate tasks in the same way that earlier revolutions used steam, electricity and assembly lines to mechanize production.

But as with all revolutions, the latest generation of AI-driven productivity gains could come at a cost.

A 2017 McKinsey report predicts automation could accelerate the productivity of the global economy by between 0.8 and 1.4 percent annually through 2030. However, it also shows that 400 million jobs will be replaced by automation, and many more jobs will be altered by this technology, often requiring a higher level of skill and education.

From retail workers and counter staff, to paralegals, radiologists and information analysts, AI could cut a large swatch out of the workforce, forcing employees to get re-educated and requiring companies to rethink how they find and train new staff.

In an era where skilled and experienced workers are hard to come by, the short-term gains from AI applications could quickly turn into long-term losses in the labor pool if this talent source isn’t replaced. “There is a lot of mismatch between open jobs and skills in the labor pool,” Karp said. To adapt, companies will need to redefine how they evaluate people and work more collaboratively with universities and internal training organizations to reskill their workforce for new roles.

Just Getting Started

Regardless of the impact, no one needs to panic about the coming AI revolution — at least not yet, Strier said. While these technologies hold huge promise for productivity gains — and potentially threaten the livelihoods of some entry-level workers — the changes won’t occur over night. While many business leaders are paying attention to AI advancements, few are even dabbling in AI applications. It will be years before this technology transforms the way we work, Purdy said. “Even mature companies are just dipping their toes in the water.”

For business leaders interested in taking the lead, he urges them to begin by identifying real business problems that AI can solve, then considering whether they have the people, processes and technical expertise to make AI work. He points out that companies must have the necessary time, expertise and access to data to create and train algorithms to do a specific job. “Even if you have a lot of data, these engineers aren’t easy to integrate into the organization,” Purdy said.

Most early uses of AI in the workplace relate to scanning paperwork and responding to customer queries, though Purdy expects the next phase to focus on pushing for innovations, such as discovering new drugs or identifying new sales opportunities based on millions of customer interactions. “This is the future of AI,” he said. “It will spur new methods of invention to drive innovation, and that is where most of the benefits lie.”

Sarah Fister Gale is a writer in Chicago. To comment, email editor@talenteconomy.io.  

A note from the art director:

When I told my editor that we should let an AI generate the feature art for this story, I was pretty sure she’d say no. So, when she said, “This is really cool; let’s do it,” I was both very excited and a little terrified.

Normally, I commission an illustrator and see iterations of the images through a series of check-ins. Because the illustrators Talent Economy works with are immensely talented, the guidance I give at these progress checks-ins can sometimes be as simple as “This looks great. Keep on doing what you’re doing.” Working with the AI was quite different.

I “hired” Deep Dream Generator, an online tool that “is capable of using its own knowledge to interpret a painting style and transfer it to the uploaded image.” From there, I began the long and tedious process of trying to make something I could use for this story. DDG works slowly, does not like feedback and is very unpredictable. If DDG was a person, they would probably get fired. Over the course of a week I fed stock images to DDG, and it returned dozens of strange unusable compositions. Since I couldn’t tell this AI things like, “use less green,” “her arm looks kind of funny,” or “we need more diversity,” I spent a lot of time finessing my images before I passed them to the machine.

Overall, I found AI to be helpful in the same way first-generation voice recognition software was helpful. When it works right it’s pretty snazzy, but the majority of the time you just end up shouting at your technology. Regarding the ever-present question of AI taking over human jobs, I don’t think our illustrators should be concerned. At least not this year. — Theresa Stoodley