We’ve finally entered the world of creepy AI robots, and the latest is essentially a robot receptionist that runs off ChatGPT’s large language model (LLM). The robot was created by engineers at Heriot-Watt University and Alana AI. It works off OpenAI’s GPT-3.5 and even features a humanoid face with a moving mouth and eyebrows.
The robot was designed solely to interact with users at the UK National Robotarium. To help alleviate the possibility that it could provide false information, the engineers scraped the website for the robotarium and added the information to a special database that the robot can access when answering questions.
The result is about what you’d expect from something utilizing GPT-3.5, which is still very impressive for a large language model. In a YouTube video, one of the engineers interacts with the robot receptionist, asking it questions. It responds after a moment of delay – I can only assume the robot is processing what the person has said.
The robot is very convincing, at least at answering basic questions, though it looks uncanny. The engineers say this is the first time someone has created an LLM knowledgeable about a singular subject and capable of verbal and nonverbal communication. It’s really impressive, despite how creepy the robot looks.
When someone talks to the robot, it converts the speech into text and then processes it – which would account for that momentary pause I mentioned above. It then deduces how it should respond, both with its voice and the appropriate facial expressions and movements, providing a more lifelike response.
The answers from the robot receptionist might be conversational in tone and appropriate to the audience, but it’s still a very robotic delivery. Most of it still feels like you’re having your phone’s text-to-speech reading things off to you. Still, it’s hard not to be impressed by what has been made possible here, and I’m sure this won’t be the last attempt at bringing GPT and other LLMs to robotic receptionists.