rtificial intelligence (AI) systems are unlikely to achieve human-like cognition unless they are connected to the physical world through robots, according to a study published in the journal Science Robotics. The researchers from the University of Sheffield emphasized the significance of incorporating principles from evolution in the design of AI systems to enable them to acquire knowledge and understanding similar to human cognition.
The study acknowledged the remarkable capabilities of current AI systems, exemplified by ChatGPT, which employ large neural networks to solve complex problems like generating coherent written text. These networks enable AI to process data in a manner inspired by the human brain and learn from errors to enhance accuracy.
Despite the similarities between AI models and the human brain, the researchers highlighted crucial differences that hinder AI systems from attaining biological-like intelligence. The first disparity lies in the fact that real brains are embodied in physical systems, such as the human body, which directly perceive and interact with the world. This embodiment grants meaning to brain processes that disembodied AIs lack, as they can recognize and generate intricate patterns in data but lack a direct connection to the physical realm, rendering them oblivious to their surroundings.
The second disparity concerns the organizational architecture of human brains, which consists of multiple subsystems arranged in a specific configuration shared by all vertebrate animals. In contrast, AI lacks this architecture, limiting its potential for developing human-like cognition. The study emphasized the crucial role played by the brain's interactions with the real world in overcoming challenges, learning, and evolving throughout the process of natural selection.
The researchers emphasized that the design of AI systems often overlooks the interplay between evolution and development, which has been vital in shaping the human brain's intelligence. Professor Tony Prescott, from the University of Sheffield, highlighted the importance of building AI systems with architectures that mimic the learning and improvement processes of the human brain, using their connections to the physical world. While acknowledging the significant advancements achieved by AI models like ChatGPT, he noted that their design methods are unlikely to lead to full human-like cognition.
In conclusion, the study suggests that AI systems will only develop human-like cognition when they are integrated with robots to establish a direct connection to the real world. By incorporating architectural designs that learn and improve akin to the human brain, AI systems may better replicate the intelligence honed throughout evolution. This research sheds light on the potential future direction for AI development, emphasizing the need to bridge the gap between disembodied models and the embodied nature of human intelligence.