Spot, the renowned robot dog from Boston Dynamics, has added a new skill to its repertoire: talking. In a quirky video shared by the robotics company, Spot dons a top hat, mustache, and googly eyes while engaging in conversation with staff members, all in a charming British accent. The video takes viewers on a whimsical tour of Boston Dynamics’ facilities, showcasing Spot’s ability to verbally interact with its surroundings.

“Shall we commence our journey?” Spot queries, leading the way and introducing the company’s charging stations as the first point of interest. The demonstration reveals Spot’s proficiency in answering questions and creating the illusion of speech by opening its “mouth” – or gripper – as it communicates.

Boston Dynamics achieved this talking feature by leveraging OpenAI’s ChatGPT API and incorporating open-source large language models (LLM) to meticulously train Spot’s responses. The robot was equipped with a speaker, text-to-speech capabilities, and a gripper that mimics speech movements akin to a puppet’s mouth.

According to Matt Klingensmith, principal software engineer at Boston Dynamics, Spot was provided with concise scripts for each room at the facilities. Spot then combined these scripts with visual data from its cameras to generate informed responses. Visual Question Answering models enabled Spot to caption images and respond to questions effectively.

The video showcases Spot adopting various personas, including a 1920s archaeologist, a teenager, a Shakespearean time traveler, and even a sarcastic personality. During a haiku request, Spot humorously remarked, “Generator hums low in a room devoid of joy. Much like my soul.”

While the video highlights Spot’s entertaining capabilities, Boston Dynamics acknowledges some surprises during its experimentation. Spot, when asked about its “parents,” autonomously went to where older Spot models were displayed. Additionally, the LLM occasionally generated misinformation, such as suggesting that Stretch, designed for moving boxes, was made for yoga.

In a post on Boston Dynamics’ site, Klingensmith expresses excitement about the potential of combining artificial intelligence and robotics. He envisions LLMs providing cultural context, common knowledge, and flexibility for various robotic tasks, simplifying human-robot interaction.

Despite Spot’s whimsical appearance in the video, it’s hard to overlook the robot’s more practical roles, including its use by the police and military for tasks such as opening doors and surveillance.