We’ve finally entered the spooky world of AI robots, and the latest is a robot receptionist powered by ChatGPT’s large language model (LLM). The robot was designed by engineers at Heriot-Watt University and Alana AI. It runs on OpenAI’s GPT-3.5 and even has a humanoid face with an animated mouth and eyebrows.
The robot has been designed to interact exclusively with users at the UK National Robot Aquarium. To eliminate the possibility of the robot giving false information, engineers scanned the robot aquarium’s website and added the information to a special database that the robot can access when answering questions.
The result is what you’d expect from something using GPT-3.5, which is still very impressive for a large language model. In a YouTube video, one of the engineers interacted with the robot receptionist and asked her questions. The robot responded after a moment’s pause.
The robot is quite convincing, at least when it comes to answering basic questions. Engineers say this is the first time anyone has developed an LLM that has knowledge about a single subject and can communicate verbally and non-verbally. No matter how scary the robot looks, it is truly impressive.
When someone talks to the robot, the robot converts the speech into text and then processes it; This explains the momentary pause we mentioned above. It then works out how to respond, both with its voice and with appropriate facial expressions and movements, providing a more realistic response.
The robot receptionist’s answers can be conversational and appropriate to the audience. However, it is definitely a very robotic presentation. A lot of it still feels like your phone’s text-to-speech feature reading things to you. Still, it is impossible not to be impressed by what is done here. We’re sure this won’t be the last attempt to introduce GPT and other LLMs to robotic receptionists.
47473
