November 19, 2019
COMPUTERS THINK IT OUT
Growing up in Toronto, Peter Tu surrounded himself with companionable creatures: the dog he ran around the neighborhood with, the reptiles living in a pit he built in the backyard. These days, rather than training dogs, Tu’s job is to train computers to learn — but Tu himself is still up to some of his old tricks. As the chief scientist for artificial intelligence at GE Research, he’s currently working with researchers from Siena College under a grant funded by the U.S. Defense Department’s Grounded Artificial Intelligence Language Acquisition program, or GAILA. The idea is to teach computers to learn language through visual and contextual cues — just like kids and animals do.
Sit, stay, compute: AI is great at processing massive amounts of data and, say, spotting hidden patterns. It’s less good in situations when the AI agent has what Tu describes as a “poverty of stimulus” — simply not enough information to go on. That’s something kids and animals are great at: “As children, we do a large number of inductions,” Tu said. “A child with limited examples is very good at drawing out general rules.” If he can figure out how AI-enabled machines can master such nuances in communicating with humans, and eventually with one another, he thinks it’ll be a natural step toward helping all sorts of machines comprehend all sorts of problems. Compared to humans, of course, they’ll still have a ways to go — millions of years of evolution have given our brains a bit of a head start.
Learn more here about Peter Tu’s highly intelligent work at GE Research.
CURIOUS BORG
Another thing computers could take their cues on from human beings? Knowing their limitations. Any person who’s just completed a 5K race for the first time, for instance, can’t expect to run a marathon the following weekend. Artificial intelligence can perform marathon feats of computing, but it runs into problems when asked to perform tasks outside its programming. Take the software that continuously adjusts the rotation of the turbines on a wind farm, making sure they’re cranking out as much energy as possible. Colin Parris and his team at GE Research have been working to teach their AI humility — so that, when faced with an unexpected wind it hasn’t been programmed to deal with, it can take a step back and switch the turbine to a safe mode. But Parris et al are wondering if they can’t take their AI a step past humble. Could they make it curious, too?
Confidence booster: Curious AI kicks in when the confidence level in a digital model of the turbine falls below a certain range. Rather than merely accept the limitations of the model, the computer asks why the certainty is below a given threshold and pushes for more data that might help round out the picture. Parris explained, “It’s like a person saying, ‘I know my weakness is punctuation in English — let me go practice that.’” Parris’ models got another kind of practice when he brought them to Argonne National Laboratory, where their predictive skills were sharpened as they were run through high-powered simulators. Now the refurbished models are being tested at a wind farm in the southwestern U.S. If engineers see improved performance there — and early signs are promising — Parris and his team will expand the curious AI model to wind farms around the globe.
Curious to learn more about this intriguing frontier in AI? Click here.
INTERSTELLAR VOYAGERS
A year ago, the spacecraft Voyager 2 escaped the heliosphere — the immediate realm of the sun’s influence, encompassing the eight planets (as well as Pluto) — and entered an area of the solar system called the interstellar medium. It was only the second probe to have made it that far: The first was its sibling craft, the Voyager 1, which crossed the threshold in 2012. Astronomers have compared the heliosphere to a kind of bubble, which acts in part to protect the planetary system from harmful galactic radiation — and they’ve wanted to know more about it. NASA researchers have just published five new scientific papers that crunch some of the data sent back by Voyager 2, and which help shed light on the outer solar system. National Geographic characterized the findings as “even weirder than expected.”
Stars in their eyes: “Taken together” — as NASA put it in a recent press release — “the findings help paint a picture of this cosmic shoreline, where the environment created by our Sun ends and the vast ocean of interstellar space begins.” Launched in 1977, the Voyager missions were only supposed to reach as far as Jupiter and Saturn, so researchers consider everything they’ve sent back since to be a bonus. And all the extra insight has been enabled, in part, by the remarkable power sources that keep the probes’ instruments and other onboard systems going: radioisotope thermoelectric generators (RTG), created by GE for both Voyagers. It’s not just the Voyagers that benefit from such tech. Earlier this year, another RTG-powered NASA probe, New Horizons, snapped photos of comet Ultima Thule — the most distant object ever visited by a human craft.
To infinity ... and beyond? Learn more here about the funky findings of Voyager 2.
COOLEST THINGS ON EARTH ?
1. Bridge Builders
Engineers at Switzerland’s Ecole Polytechnique Fédérale de Lausanne have created the “next generation” of eco-friendly concrete. The material is strong enough to reinforce bridges, but its manufacture involves fewer carbon emissions than typical concrete.
2. So Real You Can Touch It
At the University of Sussex, researchers have created holographic technology that doesn’t just project visual images: You can hear and touch these holograms too.
3. Accepting The Charges
Los Angeles has begun installing charging stations on streetlights, aiming to address one roadblock to the widespread adoption of electric vehicles: Where do apartment dwellers recharge their cars?
Learn more about this week’s Coolest Things on Earth here.
— QUOTE OF THE DAY —
“I’ve always been of the opinion that we need more companions to walk down this path with us.”
— Peter Tu, chief scientist for artificial intelligence at GE Research
Quote: GE Reports. Image: Getty Images.
ENJOY THIS NEWSLETTER?
Please send it to your friends and let them know they can subscribe here.