From the bickering droids in Star Wars, to the robotic companions and NPCs in video games like Fallout 4, talking robots have long been firmly in the realm of science fiction. However, that may have finally changed since the explosion of artificial intelligence in the past year. In fact, we might soon have chattering robo-dogs giving walking tours in a city near you courtesy of the mad scientists at Boston Dynamics.
In a video released Thursday, the company showcased a robotic dog infused with OpenAI’s ChatGPT. The bot can be seen speaking in a variety of voices and accents including a debonair British gentleman, a sarcastic and irreverent American named Josh, and a teenage girl who is so, like, over it.
“There’s been a lot of new advances around AI technology—specifically generative AI technologies. Things that are able to write poetry or paint pictures or chat with people,” Matt Klingensmith, principal software engineer for robot autonomy at Boston Dynamics, said in the video. “We were interested in seeing how we could use technologies like these for robotics.”
The robot was a result of a hackathon in which the Boston Dynamics engineers combined a variety of AI technologies including ChatGPT, voice recognition software, voice creation software, and image processing AI with the company’s famous “Spot,” the robot dog known for its ability to jump rope and reinforce the police state. The bot also had some upgrades including image recognition software combined with a “head” sensor that the engineers decorated with hats and googly eyes producing incredibly creepy results.
The team created a number of different versions of the robot including a “tour guide” personality that seemed to recognize the layout of the Boston Dynamics warehouse, and was able to provide descriptions and the history behind the various locations in the workplace.
“Welcome to Boston Dynamics! I am Spot, your tour guide robot,” the android said in the video. “Let’s explore the building together!”
In the video, the robot can be seen “speaking” and responding to different humans and a variety of prompts. For example, an engineer asked Spot for a haiku, to which it quickly responded with one. After Klingensmith said that he was thirsty, the robot seemed to direct it to the company’s snack area.
“Here we are at the snack bar and coffee machine,” Spot said. “This is where our human companions find their energizing elixirs.”
While impressive, it’s important to note that this robot can’t actually understand what it’s saying or what other people are saying to it. Like all other large language models, it’s been trained on a massive corpus of data and language in order to respond to prompts with its own predictions of what we might want to see or hear. It’s like the text predictor on your phone—albeit in a futuristic robot dog that can run, jump, and (theoretically) fight crime.
Still, watching Spot move around and speak to its human engineers is no doubt uncanny—and potentially a glimpse into a future where these types of robotics are even more commonplace. We might not be getting Rosey a la The Jetsons chattering away with us while she washes our dishes—but you might just see Spots giving tours at your next vacation destination soon enough.