Artificial intelligence machine learning getting hard to keep up with

News about how and where artificial intelligence is being put to work is authentically dumbfounding. It looks like there's going to be no stopping AI anytime soon.

Here's just a sample of some recent headlines.

AI has just helped spot 72 previously undetected "fast radio bursts" coming from a galaxy 3 billion light years away.

Scientists aren't sure what "the mysterious repeating source" is. A star that somehow got magnetized? E.T. trying to phone home? But the tracks of the bursts were hiding in plain sight all along, overlooked or missed by conventional analysis of existing data. It took AI to sift through the mountain of information and reveal the telltale tracks.

"This work is only the beginning of using these methods to find radio transients," said UC Berkeley PhD student Gerry Zhang, who led the team of analysts. "We hope our success may inspire other serious endeavors in applying machine learning to radio astronomy."

Undergraduates at Worcester Polytechnic Institute (WPI) are working on a way to use AI to save coral reefs in the US and Caribbean from a devastating invasion.

The invaders are lionfish, colorful aquarium fish with elaborate spiny fins that are native to the South Pacific and Indian oceans, where they have natural predators to keep their numbers in check.

Released into the wild in ocean waters on this side of the globe (probably by aquarium owners whose pets outgrew their homes), where there are no natural predators, they have taken off, their population expanding at an alarming rate.

With prodigiously expanding stomachs, they gobble up native juvenile fish, reducing populations by as much as 90 percent in a matter of weeks, and threaten the whole chain of life on a reef. NOAA calls lionfish the "poster child for invasive species".

The crisis has an upside－lionfish are delicious to eat and sell for as much as $20 a pound.

Students at WPI's robotics department have brought AI to the rescue. They've designed an autonomous, self-propelled lionfish-hunting robot. Dropped into the water, it can recognize a lionfish, intercept it and ram it with a buoyant spear that detaches and floats the fish to the surface for collection. Each robot has a revolving cylinder that holds eight spears.

The team showed the robot's AI program thousands of images of lionfish of different colors, from different angles and in different lighting conditions, until its recognition rate reached 95 percent. (They also trained the gizmo what definitely not to shoot, as in scuba divers or snorkelers.)

So, keep an eye out for lionfish on the menu.

AI-brained shark-detecting drones have been deployed off Australia's beaches for a year now. The so-called "Little Ripper Life Saver" can distinguish between sharks and dolphins, surfers, boats and whales, and a variety of other marine life, in real time. If a shark is spotted, Little Ripper can warn swimmers with a megaphone.

Another fascinating AI application was just unveiled by a team from MIT. Call it filling in the blanks.

Given, say, three frames of a video－a stack of cans, a human finger touching it and a pile of tumbled cans－people have no trouble envisioning what happens in between the frames.

Computers, for some reason, struggle with the concept. MIT researchers this week presented a paper at the European Conference on Computer Vision describing an add-on called Temporal Relation Network that "learns" how objects change in videos at different times.

"We built an artificial intelligence system to recognize the transformation of objects, rather than the appearance of objects," said Bolei Zhou, assistant professor of computer science at the Chinese University of Hong Kong, a co-author of the study.

"That's important for robotics applications," Zhou said. "You want [your robot] to anticipate and forecast what will happen early on, when you do a specific action."

Stay tuned. What happens next with AI is anyone's guess.

Contact the writer at chrisdavis@chinadailyusa.com