It’s been a banner week for artificial intelligence: Scientists have found or are looking for ways where AI can help diagnose depression, predict earthquakes and fly drones. They’ve also figured out ways to make microscopic robots swarm like bees. Getting a little too futuristic for you down here on the blue planet? Here, just step into this elevator …
What is it? Following in the grand tradition of the builders of the Tower of Babel, researchers at Japan’s Shizuoka University are taking steps toward building an earth-bound structure that could reach the heavens: a space elevator. This month they’re testing a tiny mock-up of their idea in orbit, sending up two tiny satellites, just 4 inches per side, that are connected to one another by a 33-foot steel cable. A motorized box — the elevator — will run between the two points, with engineers keeping close track of how it performs in space.
Why does it matter? Fuel is expensive. Rockets are mostly not reusable. And launches are complex, time-consuming endeavors that create a lot of pollution. A space elevator, long a sci-fi dream, sidesteps all those problems, potentially giving humans and their payload easy access to low earth orbit.
How does it work? Not very easily, that’s for sure. A paramount technical challenge is how to construct a cable that could run from earth to orbit — a distance of 60,000 miles — without collapsing under its own weight. The elevator will also have to contend with gravitational pull from the earth, sun, and moon; weather conditions down here on the ground; and “the centrifugal force acting on the elevator’s counterweight, way beyond Earth’s atmosphere, to keep the elevator upright,” according to Science Alert. Still, researchers think they’ll be able to pull it off: Shizuoka partner Obayashi Corp. has previously said it wants to get a space elevator up and running by 2050.
What is it? At Harvard, a team of scientists is using artificial intelligence to better predict where earthquake aftershocks are likely to occur. And as reported in Nature, the algorithms they developed could forecast aftershocks significantly better than previous models.
Why does it matter? After powerful earthquakes, recovery efforts are often hindered by aftershocks occurring at unpredictable times and locations. An idea of where they might take place would help survivors and recovery personnel avoid further harm. That’s the kind of prediction well suited to AI, which is adept at processing huge amounts of information — such as records of prior earthquakes — and sussing out underlying patterns.
How does it work? By feeding a whole bunch of data into the computer and asking the right questions. The team relied on a database that collated observations following 199 major earthquakes — tremors big enough that scientists afterward recorded what part of the fault slipped, and by how much. They paired this with a physics model of how quakes stress the earth, asking the neural network they developed “to look for correlations between locations where aftershocks occurred and the stresses generated by the main earthquake,” according to a release from Harvard. “We’re still a long way from actually being able to forecast them,” said Phoebe DeVries, a post-doctoral fellow who worked on the project. “But I think machine learning has huge potential here.”
What is it? At the Interspeech conference in Hyderabad, India, scientists from MIT announced they’ve developed a machine-learning model that enables a neural network to search human speech patterns and detect symptoms of depression.
Why does it matter? Chalk up another diagnostic win for artificial intelligence: Lately we’ve heard about AI being used to catch everything from eye disease to pneumonia. Though the work on mental health is in the early stages, researchers imagine that one day it can be paired with mobile apps, for instance, “that monitor a user’s text and voice for mental distress and send alerts,” according to MIT. “This could be especially useful for those who can’t get to a clinician for an initial diagnosis, due to distance, cost, or a lack of awareness that something may be wrong.”
How does it work? What’s special about this model is that it’s “context-free” — unlike other AI models, which “are provided with a specific set of questions, and then given examples of how a person without depression responds and examples of how a person with depression responds.” Here, by contrast, researchers fed bits of everyday speech from depressed and nondepressed people into the computer program, which learned to pick up patterns — words like “sad,” for example, paired with flat vocal tones or monotones. Lead author Tuka Alhani, a researcher at MIT’s Computer Science and Artificial Intelligence Laboratory, or CSAIL, explained, “The model sees sequences of words or speaking style, and determines that these patterns are more likely to be seen in people who are depressed or not depressed. Then, if it sees the same sequences in new subjects, it can predict if they’re depressed too.”
What is it? I bet you’ve been sitting around for years with an amazing idea for an artificial intelligence technology that would enable autonomous drones to race against — and vanquish — unmanned aerial vehicles piloted remotely by human beings. Well, now’s your chance: Lockheed Martin, the Drone Racing League and NVIDIA want to give you a pile of money for an AI drone that can do just that.
Why does it matter? Lockheed Martin is looking at the competition, an open-innovation challenge called AlphaPilot (PDF), as a way to push its technology forward. Drones are relied on increasingly in all sorts of industrial, military and personal applications and, as has been manifestly demonstrated in today’s 5 Coolest Things, AI is everywhere.
How does it work? The challenge, which opens later this year, culminates in 2019 in a contest in which human drone pilots will go wing-to-wing against contestant entries — AI programs, as Lockheed put it, “that can navigate a fully autonomous drone through complex, multi-dimensional racing courses — without any pre-programming or human intervention.” The first team whose drone beats a human will take home $250,000, with the grand prize winners getting a cool million.
What is it? Annals of “Didn’t I see this in a ‘Black Mirror’ episode?”: Mechanical engineers at the Chinese University of Hong Kong have created nanobots — tiny machines — that engage in swarming behavior, such as is found in bees, bacterial colonies and schools of fish.
Why does it matter? Unlike how they perform in sci-fi dystopias, swarming nanobots in real life could have valuable applications: The researchers, who published their findings in Nature Communications, envision them used to deliver drugs in the body, treat blood clots or assist surgeons in performing complex tasks. “The nanobot swarm can be operated in a controlled fashion with a high speed, which has never been reported before,” said study leader Li Zhang.
How does it work? Researchers were inspired by the way animals swarm in nature: Ants, for instance, can use their bodies to build chains or bridges to get the colony across rough terrain. Zhang’s team used oscillating magnetic fields to influence the movement of the nanobots, which are each less than one micron wide, or a fifth of the size of a red blood cell; they found they could manipulate a bunch of tiny machines into elongating, splitting apart and merging back together.