- Microsoft is testing what happens when ChatGPT takes control of small robots.
- Researchers at Microsoft imagine a day when people can give instructions like "heat up my lunch" to at-home robots.
Microsoft's partnership with OpenAI, the maker of ChatGPT, has already set off an AI-fueled frenzy in the tech world, but now, a team of researchers at Microsoft is pushing the AI technology even further by allowing ChatGPT to break out of the confines of the computer, setting it free in the real world.
This week, a team of researchers at Microsoft released its early findings after allowing ChatGPT to take control of robots.
"Our goal with this research is to see if ChatGPT can think beyond text, and reason about the physical world to help with robotics tasks," the Microsoft researchers wrote.
Straight out of science fiction, the researchers at Microsoft imagine a day when the average person can bark instructions like "please warm up my lunch" at an at-home robot home assistant and watch it complete the task from start to finish.
In this report, researchers detailed how they used certain prompts to allow ChatGPT to take control of a small drone. By writing instructions in the ChatGPT chatbox, the researchers were able to direct the drone to find objects in a room like a "healthy drink," "something with sugar and a red logo" (in this case, a can of Coke), and a mirror for the drone to take a selfie in front of.
The reason ChatGPT is well suited to following these commands is the same reason that OpenAI's tech has made some software engineers nervous about the future: the AI chatbot can quickly translate natural language into code.
OpenAI has kept many industries on their toes since the company decided, as a last resort, to release its ChatGPT AI chatbot to the public in late November. Experts anticipate that a host of industries, including writing, education, and software engineering, will be disrupted as a result of the technology.
While Microsoft's researchers acknowledged that their work "represents only a small fraction of what is possible," they cautioned that these outputs from ChatGPT should not be deployed without "careful analysis."