How AI Can Amplify Human Competencies

Though artificial intelligence systems are already becoming a part of daily life, recent debates about AI and the future of work have gained a sense of urgency. The late Stephen Hawking worried that humans “couldn’t compete, and would be superseded” by machines, while Tesla founder Elon Musk has suggested that competition in AI could lead to World War III. The Economist reported earlier this year that nearly half of the jobs in 32 developed countries surveyed by the Organisation for Economic Co-operation and Development (OECD) were vulnerable to automation, declaring, “a wave of automation anxiety has hit the West.”

Ken Goldberg, professor and department chair of industrial engineering and operations research at UC Berkeley, is pushing back on all of that. Instead of embracing the notion that robots will surpass humans and replace us in the workforce (a concept referred to as “singularity”), he argues for “multiplicity” — a hybrid view of how new technologies and people might work in partnership toward human goals. To an extent, he says, this is how AI is already starting to function.

MIT Sloan Management Review correspondent Frieda Klotz spoke with Goldberg about a future in which AI is a complement, not a threat, to workers. What follows is an edited and condensed version of their conversation.

MIT Sloan Management Review: What areas of robotic technology is your lab currently working on?

Goldberg: We’re developing robot software for tasks as wide-ranging as warehouse order fulfillment, home decluttering and robot-assisted surgery. What’s common to all the work we’re doing is the idea of algorithms and learning for robots, improving our ability to analyze data and examples and then use that to build control policies — or models — for how robots can move.

The area I’ve been working on for 35 years is robot grasping — how to reliably pick up objects. It’s easy for humans, but it’s a problem for robots. Basically, every robot is still a klutz, and that’s a big challenge if you want to develop one that will declutter a home or pack boxes in a warehouse.

Can you talk about your concept of multiplicity?

Goldberg: People keep saying we’re on the verge of a transition, the singularity, when computers will take over. There’s a sense that AI is a magical technology that’s going to transform industries and replace humans, putting people out of work. But we’re not anywhere near that point.

There are really good technologies and many interesting developments, and in some domains machines can be better than humans. Machines are very good at precision; they’re very good at calculating numbers and pattern recognition. But there are several domains in which machines, and especially robots, don’t excel. The most advanced robotic grasping technique isn’t as deft as a 3-year-old! I’m concerned that people have expectations that are out of line with the current reality — and that these will distract us from what we should be worrying about and planning for. That’s what led me to multiplicity, the idea that we’ll see new partnerships between teams of humans and machines. Most of the systems that we use actually arise from human interaction. And this is already happening every day — for example, when by clicking on results, we give Google’s search algorithm feedback that it then uses to refine future results.

Multiplicity requires diversity. If you look at a body of thinking called ensemble theory, you can prove that diversity is helpful for a machine learning system. The relationship is something you can formulate mathematically. That’s really exciting, because it’s consistent with what we’re starting to find about groups of humans: that if you have a diverse group of people, you get better, more creative ideas, more insights, and better outcomes.

We’ll see different kinds of diversity, then — not just between people, but with people and robots putting their efforts together.

Goldberg: Exactly. Qualities like intuition, empathy, creativity are all very human — we’re very good at looking at holistic situations, generalizations — and we can blend those qualities with the precision that machines provide.

We should be celebrating this, because it literally leads to better decisions and better processes.

In the next few years, how might robotics not be as useful as people expect?

Goldberg: People claim that we’re going have autonomous trucks, which would eliminate truck driver jobs. They say this about Uber drivers or Lyft drivers too, but this is not going to come to pass.

We will make some progress; you can drive for good stretches on the freeway today with a robotic system. But there are so many complexities about driving in a city or a suburban environment that make it so much harder, especially if you’re in a truck, because there are narrow and winding streets to navigate. We’re going to need human truck drivers for the foreseeable future — for the rest of my lifetime and my kids’ lifetimes.

Another example is that some claim there’s no future for journalists. Computer systems take data about sporting events and then generate stories, which read reasonably well. That’s because they can identify patterns and put numbers and results into those patterns, and it may work to an extent. But machines don’t have the ability to pick up what is really interesting about a sporting event, the particular nuances of what’s going on, or make analogies about what the teams are doing.

Aren’t machine learning teams working on these kinds of distinctions?

Goldberg: They are, but realistically they are years away from making it happen. What robots are great at are jobs that no one else wants to do — the dirty, dull, and dangerous jobs. I do think we’ll have our decluttering robot that can tidy up around our homes in the next 10 years, at a price we can afford. Robots will also excel at tasks like washing windows on skyscrapers.

When it comes to more specialized fields like medicine, some of my work uses data from human surgeons and inferred models to develop robots that can perform suturing or remove fragments — tasks considered tedious by most surgeons. This gives physicians the ability to be focused and present and have more attention for the things that matter most.

What could business leaders be doing to allow these sorts of partnerships to flourish in their organizations?

Goldberg: CEOs should appreciate the value of the people who work for them and reassure employees that AI systems can actually help them do their jobs better, instead of replacing them.

AI will be able to perform many of the duller office tasks. Think of the pain points that hinder workers from getting on with the more important parts of their jobs — scheduling meetings, transcribing, taking notes, summarizing and indexing documents. What CEOs should be thinking about is how these tools can enhance the performance of employees.

Is there any risk that you are underestimating machines and their abilities?

Goldberg: I could be wrong, of course. But I have not seen any evidence that a computer is capable of innovation and creativity. Robots can be programmed to behave in a way that mimics human inventiveness, but they’re unable to innovate spontaneously, to exchange ideas the way people do, to forge truly new insights or designs, and to recognize them as such. Doing this requires a vast understanding of what is normal and what isn’t, which we don’t know how to formalize.

It’s one element of the Turing test, which examines whether a machine can keep up its end of an interesting conversation in a way that’s indistinguishable from human intelligence. We’re not even close; by that measure, we don’t have intelligent machines, and we haven’t made any progress, really, in 60 years. All the developments in AI are exciting, but that human-level frontier is still as hard to breach as it was decades ago.

Why do you think people have latched onto the idea of singularity when it may not accurately represent technological advances?

Goldberg: Even in the beginning of the 20th century, when automation came out, there was talk about robots taking over. It’s cyclical.

People say this time is different — the technology is different. Yes and no. The fact is, we do have faster computers, we have a lot more data to work with, and we have made some progress. But in the most important ways, machines are nowhere near surpassing humans.