Boston Dynamics has taken its already very promising robot dog, Spot, and given it a voice. The company shared on its blog that engineers working with Spot trained an AI model that integrated ChatGPT and other AI models, allowing the robot to act as a talking dog tour guide across the Boston Dynamics campus.
The engineers at the company say that a robot tour guide was a simple way to test the concepts of a talking robot that could walk around, look at objects around it, and then describe them using VQA or caption modeling. The new system utilizes ChatGPT to help make its narration possible, and this is something we’ve seen in another ChatGPT-powered robot.
Because Boston Dynamics’ talking dog robot didn’t just utilize fixed responses, it could look at what was around it and describe it using the large-language model (LLM) that powers ChatGPT. This also allowed it to field questions from the audience and respond to their queries without simply responding with canned answers.
Of course, like any LLM, there was a concern about “hallucinating,” which is where the AI spits out things that aren’t factual or are even made up or incorrect entirely. The team also utilized some engineering techniques to help ensure the talking dog robot wouldn’t spiral out of control or give any distasteful responses to the audience.
Seeing the success of these kinds of robots will only push others to create even more responsive talking robots. And now that ChatGPT is continually expanding, we’ll likely see more companies taking the stance that Boston Dynamics has. Luckily, Boston Dynamics does everything it can to keep its robots from being used for bad, so there is some hope there.
If you’re concerned about the robots eventually rising up and AI overthrowing humanity, though, this news is probably going to be a bit terrifying to read about.