Monday, February 26, 2024
HomeAi in GamingMeet The AI That Will Beat You At Physical Challenges As Well...

Meet The AI That Will Beat You At Physical Challenges As Well As Mind Games

Artificial intelligence machines that can defeat humans at complex strategy games such as chess and Go are becoming almost run-of-the mill, but until now, AI has not been able to compete with mankind in the physical domain. However, Raffaello D’Andrea, a professor at the Swiss university ETH Zürich and the founder and CEO of drone-focused start-up Verity, claims to have changed that.

Together with one of his ETH Zürich students, Thomas Bi, D’Andrea has built an AI system capable of learning how to play “the labyrinth”, a game in which players must steer a marble through a maze while avoiding traps, and then actually doing it. With six hours of training, his AI-powered robot, christened CyberRunner, managed to beat the game more quickly than human competitors who have been playing it for years. A film recording CyberRunner’s skills makes chastening viewing for human competitors.

“A year ago, we set ourselves a challenge: to build a robotic AI that could learn how to accomplish a physical task and do so more quickly than a human,” D’Andrea explains. “The big successes of AI to date have been virtual, but we’ve seen much less progress when it comes to physical challenges.”

CyberRunner uses a camera and physical controls to play the labyrinth, recording its observations of each of its efforts in order to understand how to master the game more effectively next time. The system proved so smart that D’Andrea even had to intervene to stop it “cheating” when it discovered possible short cuts enabling it to miss out parts of the maze.

D’Andrea’s experiment proves that AI can be used to do jobs that require very fine motor skills and spatial awareness – and to learn how to do this work more effectively than human operators in a short space of time. “The fact we can build something that beats a human with just six hours of training is eye-opening,” he says.

To develop CyberRunner, D’Andrea explains, he used the relatively new science of model-based reinforcement learning, with the AI trained simultaneously on the task in hand and on a virtual model of the work. “It’s much more akin to how humans learn to do something,” he explains. “We practice the task we’re trying to accomplish, but we also build an internal model of the task in our minds, which the brain works on in the background.”

D’Andrea plans to publish the coding for the machine on open-source platforms in order to encourage other innovators to put it to work on multiple use cases. “I see this as setting the benchmark,” he explains. “It’s a low-cost option for extending the applications of AI into the physical world.”

That generosity may raise eyebrows, given the potential commercial applications of the technology – including at D’Andrea’s very own Verity. The business, founded in 2014, has grown quickly, raising $80 million of funding to date; it serves clients such as Ikea and Maersk, which use its autonomous drones to collect data inside their vast warehouses. The drones can, for example, check inventories far more quickly than any human operative.

D’Andrea’s view is that the open-source approach will unlock widescale adoption of his technology much more quickly than keeping it to himself, to the benefit of all organisations, including Verity. “Our biggest competitors are not other companies in our space, but the entrenched technologies that dominate our domain,” he argues.

D’Andrea’s record as a founder suggests he understands the commercial realities – while working at Cornell University, he co-founded Kiva Systems, which was acquired by Amazon for $775 million in 2012. The company’s technology has subsequently been incorporated in Amazon’s fulfilment centres worldwide.

Nevertheless, D’Andrea is adamant that Verity should not retain sole rights to his latest innovation. “It’s just powerful to explore the boundaries from an academic perspective,” he says. “We’re really exploring what we can do with this technology.”

One obvious question is whether this work will add to the disruptive impact of AI on the labour market, where many analysts are already concerned about the prospect of large scale job losses as technology replaces humans.

D’Andrea is sympathetic to such concerns. “Most of the tasks that people do in places such as warehouses are menial and often unpleasant, so it is no bad thing to get rid of those tasks,” he says. “But the displacement problem is a real one, because this technology is moving much more quickly than previous waves of upheaval, so there is less time to explore redeployment opportunities.”

This will be the challenge for policymakers, he suggests. Interventions in the labour market such as education and training can help ensure people continue to hold down jobs, even as the world transitions to new technology.

Follow me on Twitter

Source link

Leah Sirama
Leah Sirama
Leah Sirama, a lifelong enthusiast of Artificial Intelligence, has been exploring technology and the digital realm since childhood. Known for his creative thinking, he's dedicated to improving AI experiences for all, making him a respected figure in the field. His passion, curiosity, and creativity drive advancements in the AI world.


Please enter your comment!
Please enter your name here

- Advertisment -

Most Popular