Friday, August 23, 2019

And now, a bicycle built for none

  • >>Cade Metz, The New York Times
    Published: 2019-08-01 12:21:50 BdST

bdnews24
In an undated image provided by Pei et al, Nature, a frame from a video by researchers in China showing a self-driving bicycle, with a neurotrophic computer chip that helps it understand certain commands. It's not the first self-driving bike, but it may be the nearest to thinking for itself. The New York Times

As corporate giants like Ford, GM and Waymo struggle to get their self-driving cars on the road, a team of researchers in China is rethinking autonomous transportation using a souped-up bicycle.

This bike can roll over a bump on its own, staying perfectly upright. When the man walking just behind it says “left,” it turns left, angling back in the direction from which it came.

It also has eyes: It can follow someone jogging several yards ahead, turning each time the person turns. And if it encounters an obstacle, it can swerve to the side, keeping its balance and continuing its pursuit.

It is not the first-ever autonomous bicycle (Cornell University has a project underway) or, probably, the future of transportation, although it could find a niche in a future world swarming with package-delivery vehicles, drones and robots. (There are even weirder ideas out there.) Nonetheless, the Chinese researchers who built the bike believe it demonstrates the future of computer hardware. It navigates the world with help from what is called a neuromorphic chip, modelled after the human brain.

In a paper published Wednesday in Nature, the researchers described how such a chip could help machines respond to voice commands, recognise the surrounding world, avoid obstacles and maintain balance. The researchers also provided a video showing these skills at work on a motorised bicycle.

The short video did not show the limitations of the bicycle (which presumably tips over occasionally), and even the researchers who built the bike admitted in an email to The Times that the skills on display could be duplicated with existing computer hardware. But in handling all these skills with a neuromorphic processor, the project highlighted the wider effort to achieve new levels of artificial intelligence with novel kinds of chips.

This effort spans myriad startup companies and academic labs, as well as big-name tech companies like Google, Intel and IBM. And as the Nature paper demonstrates, the movement is gaining significant momentum in China, a country with little experience designing its own computer processors but which has invested heavily in the idea of an “AI chip.”

The hope is that such chips will eventually allow machines to navigate the world with an autonomy not possible today. Existing robots can learn to open a door or toss a Ping-Pong ball into a plastic bin, but the training takes hours to days of trial and error. Even then, the skills are viable only in very particular situations. With help from neuromorphic chips and other new processors, machines could learn more complex tasks more efficiently and be more adaptable in executing them.

“That is where we see the big promise,” said Mike Davies, who oversees Intel’s efforts to build neuromorphic chips.

Over the past decade, the development of artificial intelligence has accelerated thanks to what are called neural networks: complex mathematical systems that can learn tasks by analysing vast amounts of data. By metabolising thousands of cat photos, for instance, a neural network can learn to recognise a cat.

This is the technology that recognises faces in the photos you post to Facebook, identifies the commands you bark into your smartphone and translates between languages on internet services like Microsoft Skype. It is also hastening the advance of autonomous robots, including self-driving cars. But it faces significant limitations.

A neural network doesn’t really learn on the fly. Engineers train a neural network for a particular task before sending it out into the real world, and it can’t learn without enormous numbers of examples. OpenAI, a San Francisco artificial intelligence lab, recently built a system that could beat the world’s best players at a complex video game called “Dota 2.” But the system first spent months playing the game against itself, burning through millions of dollars in computing power.

Researchers aim to build systems that can learn skills in a manner similar to the way people do. And that could require new kinds of computer hardware. Dozens of companies and academic labs are developing chips specifically for training and operating AI systems. The most ambitious projects are the neuromorphic processors, including the Tianjic chip under development at Tsinghua University in China.

Such chips are designed to imitate the network of neurons in the brain, not unlike a neural network but with even greater fidelity, at least in theory.

Neuromorphic chips typically include hundreds of thousands of faux neurons, and rather than just processing 1s and 0s, these neurons operate by trading tiny bursts of electrical signals, “firing” or “spiking” only when input signals reach critical thresholds, as biological neurons do.

“This is about trying to bridge and unify computer science and neuroscience,” said Gordon Wilson, the chief executive of Rain Neuromorphics, a startup company that is developing a neuromorphic chip.

Neuromorphic chips are by no means a re-creation of the brain. In so many respects, the workings of the brain remain a mystery. But the hope for such chips is that, by operating a bit more like the brain, they can help AI systems learn skills and execute tasks more efficiently.

Because each faux neuron fires only on demand rather than continuously, neuromorphic chips consume less energy than traditional processors. And because they are designed to process information in short bursts, some researchers believe they could lead to systems that learn on the fly, from much smaller amounts of data.

In the video, the bicycle is not learning; it is merely executing software that had been trained to handle specific tasks, including recognising spoken words and avoiding obstacles. But it is executing the software in an efficient way, which is important to vehicles that run on battery power. Researchers believe they can eventually merge the training process and the in-the-moment execution, so that a bicycle could learn as it goes, from just a few moments of experience.

The rub is that building the right hardware may require at least several more years of research. “We are still in the trial and error stage,” said Georgios Dimou, who previously worked on Intel’s neuromorphic project.

The Chinese researchers believe that time will bring far more than just autonomous bicycles. Their paper paints the Tianjic chip as a step toward “artificial general intelligence,” a machine that can do anything you and your brain can do. But that is merely the promise du jour. Maybe start with helping it learn to ride a bike.

 

 

© 2019 New York Times News Service