Put ChatGPT in C-3PO to make AI as intelligent as humans, scientist says

Connecting multiple AIs together and putting them in the body of a robot, similar to Star Wars’ C-3PO or R2-D2, is the most likely way AI will become as intelligent as humans, according to a scientist from the University of Sheffield.

In a new book, Professor Tony Prescott, Professor of Cognitive Robotics, argues that for AIs to match or exceed human intelligence they need to have an architecture that is similar to the human brain—an approach known as robotic intelligence—in which several AIs all specializing in different abilities are all linked as part of one larger intelligence system.

AIs are intelligent, but they are mainly trained to be good at one thing—for example, generative AIs that can produce text, images or video.

Large language models (LLMs), such as ChatGPT, don’t understand meaning—they produce text based on math and processing. They understand the relationship between words but they don’t understand how words relate to the world.

The Sheffield professor says that for LLMs to get this understanding they could be connected to other AIs—such as vision AIs that produce art.

For AIs to match or exceed human intelligence, they also need to understand the world around them and acquire self awareness. Professor Prescott believes the best way to achieve this is through multiple connected AIs, all specializing in different tasks, being embodied in a robot, as this will enable them to interact with their surroundings, learn and experience the world as humans do.

“Scientists have argued for decades over whether AI will ever match or exceed human intelligence, but now that AI is advancing at an unprecedented rate, this question has become ever more important,” Professor Prescott said. “If AI is to develop human-like intelligence, then how it will likely do this is another part of the debate that is currently dividing opinion.

“There have been some major breakthroughs recently, particularly in deep learning and generative AIs such as ChatGPT, but this isn’t real intelligence like we have as humans. Aspects of human intelligence, such as scientific discovery, aspects of creativity and imagination will remain beyond AI until we build it with a cognitive architecture that is similar to the human brain.

“For AI to match or exceed human intelligence, it also needs a body to help it gain a genuine understanding of the world around it—something which it can only get through robotics.”

Connecting multiple AIs together and putting them in the body of a robot, similar to Star Wars’ C-3PO or R2-D2, is the most likely way AI will become as intelligent as humans, according to a scientist from the University of Sheffield.

In a new book, Professor Tony Prescott, Professor of Cognitive Robotics, argues that for AIs to match or exceed human intelligence they need to have an architecture that is similar to the human brain—an approach known as robotic intelligence—in which several AIs all specializing in different abilities are all linked as part of one larger intelligence system.

AIs are intelligent, but they are mainly trained to be good at one thing—for example, generative AIs that can produce text, images or video.

Large language models (LLMs), such as ChatGPT, don’t understand meaning—they produce text based on math and processing. They understand the relationship between words but they don’t understand how words relate to the world.

The Sheffield professor says that for LLMs to get this understanding they could be connected to other AIs—such as vision AIs that produce art.

For AIs to match or exceed human intelligence, they also need to understand the world around them and acquire self awareness. Professor Prescott believes the best way to achieve this is through multiple connected AIs, all specializing in different tasks, being embodied in a robot, as this will enable them to interact with their surroundings, learn and experience the world as humans do.

“Scientists have argued for decades over whether AI will ever match or exceed human intelligence, but now that AI is advancing at an unprecedented rate, this question has become ever more important,” Professor Prescott said. “If AI is to develop human-like intelligence, then how it will likely do this is another part of the debate that is currently dividing opinion.

“There have been some major breakthroughs recently, particularly in deep learning and generative AIs such as ChatGPT, but this isn’t real intelligence like we have as humans. Aspects of human intelligence, such as scientific discovery, aspects of creativity and imagination will remain beyond AI until we build it with a cognitive architecture that is similar to the human brain.

“For AI to match or exceed human intelligence, it also needs a body to help it gain a genuine understanding of the world around it—something which it can only get through robotics.”