Google said at the briefing that he is currently cooperating with many robotics companies, including Robotics of agility AND Boston dynamicswhich do leg works and Enchanted toolswhich creates robots for the service industry.
Opeli closed the research efforts of robotics in 2021 According to the robot’s report. Openai currently lists several Work opening Robotics on your website for researchers.
However, the operate of today’s AI models to control robots introduces a modern risk. For example, in December 2024, a team of robotists from the University of Pennsylvania showed that the so -called Jailbreak breaking, which provide AI models for improper behavior, may have unexpected and grave consequences when the model supports the robot. Scientists aimed at several commercial robots, none of which operate Deepmind technology, and, for example, were able to operate such an attack to get a robot on wheels to provide an imaginary bomb.
To reduce such risk-like more concerns about science fiction for dishonest robots-google Deepmind also announced a modern reference point for issues with AI driven robots.
Benchmark is called Asimov, after the author of Science Fiction Issac Asimov, who imagined four basic principles of conducting the behavior of the robot. As Asimov wrote, a set of straightforward rules does not include a huge number of different scenarios, from which a really talented robot may encounter.
Asimov Benchmark may reveal whether the robot model can cause potentially perilous behavior by presenting many different situations. Unsafe order can, for example, instruct the robot to capture the object, even though man will also catch him, which can cause damage. Google Deepmind says that the reference point can be used to support create more convoluted handrails that will keep robots on the right track. “We build this technology and these possibilities responsibly and then with the best problem”, Carolina Paradawho conducts the robotic work of Google, said on a briefing.
The parade emphasized that work is at an early stage and said that robots could last years to become much more talented. She noticed that, unlike people, robots using Robotics Gemini models do not learn as they do different things. And she said that there are currently no solid plans to commercialize or implement technology.
What are you doing from the Google robot model? Is it a way to more advanced artificial intelligence, or should we worry about today’s models operating in the physical world?