"Robots and AI models share one crucial characteristic," [1]writes the Washington Post . "Whether to move around, conduct conversations or solve problems, they function by constantly taking in and computing increasingly vast quantities of data. It's a brute-force approach to automation. Processing all that data makes them such energy guzzlers that their planet-warming pollution could outweigh any benefits they offer."

But then the article visits the robot soccer team of René van de Molengraft (chair of robotics at Eindhoven University of Technology in the Netherlands). "One solution, Molengraft thinks, might lie in 'lazy robotics,' a cheeky term to describe machines doing less and taking shortcuts..."

> There may be ceilings for laziness: limits to how much superfluous energy use can be stripped away before robots stop functioning as they should. Still, Molengraft said, "The truth is: Robots are still doing a lot of things that they shouldn't be doing." To waste less energy, robots need to do less of everything: move less, and think less, and sense less. They need to focus only on what's important at any particular moment. Which, after all, is what humans do, even if we don't always realize it....

>

> Lazy robotics is already percolating out of university labs and into the R&D wings of corporations.... On the outskirts of Eindhoven, engineers at health technology firm Philips have encoded lazy robotics into two porcelain-white machines. These robots, named FlexArm and Biplane, move around an operating theater with smooth hums, taking X-ray images to help surgeons install cardiac stents or work on the brain with greater precision.... The robots use proximity sensors, which use far less energy. Lazy robotics can also cut down on the number of X-rays during a procedure. Frequently, surgeons take multiple X-rays to make their work as precise as possible. But with the robots' help, they can track the exact coordinates on a patient's body they are operating on in real time...

>

> The theories behind lazy robotics make robots smart in a more practical way: by coding in an awareness of what they don't need to know. It may be a while before these solutions are deployed at scale out in the world, but their potential applications are already evident... Molengraft sees an extension of lazy robotics into the realm of generative AI, in which machines don't learn how to move but learn how to learn by processing veritable oceans of data... It's wiser to build versions that contain only the necessary information. A language model used by software engineers, for instance, shouldn't need to run through its training data about world history, sporting records or children's literature. "Not every AI model has to be able to tell us about the first Harry Potter book," Molengraft said.

>

> The less data an AI model crunches, the less energy it uses — a vital efficiency fillip given that ChatGPT now uses 500,000 kilowatt-hours of energy a day, responding to 200 million queries. A U.S. household would need more than 17,000 days on average to rack up the same electricity bill... Molengraft sees this work as indispensable if the forthcoming age of machines is to be a cleaner time as well.



[1] https://www.msn.com/en-us/news/technology/why-the-world-needs-lazier-robots/ar-AA1wKqcH