"Accessing the Real World through Robots to Achieve Artificial Intelligence Cognitive Abilities"

2023-06-13

According to research from the University of Sheffield, connecting artificial intelligence systems to the real world through robots and applying evolutionary principles is the most likely way for AI to achieve human-like cognition. Current AI systems, such as ChatGPT, utilize large-scale neural networks to solve complex problems, such as generating understandable written text. These networks teach AI to process data in a similar way to the human brain, learning from mistakes to continuously improve accuracy. Despite the similarities between these models and the human brain, Sheffield researchers point out important differences that hinder their ability to achieve biologically-inspired intelligence. Firstly, the true brain exists within a physical system, the human body, which directly perceives and interacts with the world. The presence of a body gives meaning to brain processes in a way that is impossible for AI without a physical form. While this AI can learn to recognize and generate complex patterns in data, it lacks direct connection to the physical world and therefore lacks understanding or consciousness of its surroundings. Secondly, the human brain consists of multiple subsystems organized in a specific architecture, similar to all vertebrate animals from fish to humans. However, this architecture is not applicable to artificial intelligence. Sheffield's research suggests that biological intelligence, like in the human brain, develops because of this specific architecture and how it utilizes connections to the real world to overcome challenges, learn, and continuously improve through evolution. According to this research, the interaction between evolution and development is rarely considered in the design of artificial intelligence. Professor Tony Prescott, Professor of Cognitive Robotics and Director of the Sheffield Robotics at the University of Sheffield, says, "ChatGPT and other large neural network models are exciting developments in the field of artificial intelligence, demonstrating that challenging tasks like learning human language structure can be solved. However, if we continue to design these types of AI systems using the same methods, they are unlikely to develop full thinking capabilities like the human brain." "If AI systems are built based on learning and improvement architectures similar to the human brain, utilizing connections to the real world, it is more likely to develop human-like cognitive abilities. Robot technology can provide these connections for AI systems, such as through sensors like cameras and microphones, and actuators like wheels and claws. This way, AI systems can perceive the surrounding world and learn like the human brain." Sheffield scholars point out recent progress in developing AI for controlling robots. For example, a powerful approach is using recurrent neural network models, which consist of multiple feedback loops and are trained to better predict possible future events. These models have made significant progress in enhancing robot adaptability. However, the research indicates that AI in robots still has a long way to go in terms of collaborating different brain subsystems as part of a broader cognitive architecture, far from being like a real brain. Dr. Stuart Wilson, Senior Lecturer in Computational Neuroscience at the University of Sheffield, states, "In recent decades, efforts to understand how the real brain controls the body have led to exciting developments in robotics and neuroscience by creating artificial brains for robots." "After reviewing these efforts primarily focused on how the human brain learns, we believe the next breakthrough in artificial intelligence will come from closely mimicking how the real brain develops and evolves." Source: https://techxplore.com/news/2023-06-ai-gain-human-like-cognition-real.html

RECENT AI NEWS

RECENT AI TOOLS