“You have to fundamentally move the needle,” said Daniel Kaufman (UC San Diego B.A. ‘84) about the kinds of projects in Google’s Advanced Technology and Projects group (ATAP). ATAP is a small band of makers and believers that is mobile first, lean, agile and optimized for speed, and Kaufman is the Director of ATAP. As he explains it; ATAP pairs the speed, efficiency and execution of a startup with the scale and distribution of Google.
On June 2, Kaufman spoke about Google ATAP at the launch event for the UC San Diego Institute for the Global Entrepreneur, a collaboration between the Jacobs School of Engineering and Rady School of Management that aims at ensuring that Jacobs School engineers have the maximum positive impact possible as technology leaders, entrepreneurs and global citizens.
“We conduct fundamental scientific research, but we do it in a way that furthers a particular problem that we care about,” said Kaufman.
Electrical engineering professor Sujit Dey described a similar approach for the UC San Diego Institute for the Global Entrepreneur, which he directs. As a professor and entrepreneur, Dey wants to help faculty and students at the Jacobs School of Engineering and across campus pursue research that is both fundamental and relevant for the industry.
Bad Seinfeld routines
Kaufman hears a lot of pitches for potential new ATAP projects.
“Sometimes pitches sound like bad Seinfeld routines: ‘Don’t you hate it when…’ [insert awkward situation here].”
And when you step back and look at why things are the way they are, you often find fundamental scientific and technological problems that are in the way, explained Kaufman. Clearing these roadblocks sometimes requires developing new materials, or creating new processes, under tight time constraints. Google ATAP gives teams just two years to turn a specific idea into a compelling demonstration at a convincing scale.
“I don’t want to see slides, I want to see a thing, and I want it to work,” said Kaufman.
What is a convincing scale? “It has to be able to be built into a commercial product. So here’s an example: We created a wearable platform using conductive fibers. We wanted to integrate them into a garment, so we partnered with Levi’s. But the point wasn’t to just do it once, in a prototype. The point was to produce a large number of jackets, on the company’s mills—the same production lines as other Levi’s jackets—with industrial wattage. That’s convincing scale. And that’s what we’re doing; we’re now working on the release of that Levi’s jacket next year.”
Kaufman is looking for passionate, smart people with a big idea that solves a real problem.
“There is this cry for consensus, and I don’t like consensus. Most likely, if everyone agrees on it, that it’s a really good idea, then it’s probably not far thinking enough for us,” said Kaufman. “I like finding people that are brilliant and passionate about what they do. I want that one person who thinks they can make the impossible, possible.”
What will the future look like?
“I don’t believe in predicting the future,” quipped Kaufman during the Q&A after his talk at the launch event. (An attendee had asked what Kaufman thought the world look like in 10 years.) Kaufman went on to explain that he doesn’t think anyone knows what the future will look like. But he did highlight a trend. “I’m really interested in human-computer partnerships,” which, according to Kaufman, will be more useful than pure artificial Intelligence (AI) solutions.
“Today, we make people act like computers…and we try to make computers do human things, which they are abysmally bad at. When you do a search, we make you type because that’s how engineers think about things. But it’s a completely non-natural way to do anything. But I think we can put things together in really interesting ways,” said Kaufman.
After reminding the audience that the IBM supercomputer Deep Blue beat chess champion Garry Kasparov in 1997, Kaufman said: “A group of pretty good chess players have now partnered with computers that are not anywhere near as good as Deep Blue, just solid good computers, and those together went against Deep Blue and are beating it. That’s fascinating to me. I think that’s the sweet spot: when we think about machines as partners. Time will tell, but I believe that people are going to start focusing much more on Explainable AI.
“Anyone who has worked with AI has run into this: you ask it a question and invariably at some point it gives you a stupid answer. And you think, that’s a stupid answer, so I’m not going to trust any of your answers. So you throw away the whole system, and you build another one. But think about it for a moment: you have friends that say stupid things all the time, but you don’t throw them away. I mean if they say A LOT of stupid things you do, but mostly you don’t. So what’s the difference?
“The difference is this: when your friend says something stupid, you can ask them why they said the stupid thing, and it often turns out not to be stupid: they misunderstood or they didn’t quite hear what you said. You get calibrated back to ‘OK, they’re not stupid, we just had a miscommunication.’ You can’t do that with a computer. But imagine that you could. I think that’s a really interesting area of research. If your AI systems gives you a dummy answer, imagine that you don’t just throw it away, imagine that you ask ‘But why?’
“A human could lie to you. A human might not tell you, ‘I wasn’t listening to you because I was thinking about something else; I was just bored; or I was watching a movie.’ A human might not tell you that; a human might just make up something. But the computer will actually tell you. A computer has a bunch of weighted things. It will just go through the stack and report: ‘I was told this, and I was told this, and I was told this, so I concluded THAT.’
“And the human can identify what idea was wrong, and rebalance the weighted items the system is relying on.
“If I had to think of one trend in the future, I would think of human-computer partnerships.”
As it turns out, collaborations between humans and robotic systems are a focus area of the UC San Diego Contextual Robotics Institute, which the Jacobs School launched in October 2015 in conjunction with the UC San Diego Division of Social Sciences. Read about the new faculty director of the Contextual Robotics Institute, Henrik Christensen.