As long as AI systems continue to be trained in the current model and designed using current architecture, it is highly unlikely that they will ever develop human-like cognition
Despite the worldwide buzz about artificial intelligence (AI) systems soon outwitting humans and taking control, a new research conducted by scholars at the University of Sheffield concludes that as long as AI systems continue to be trained in the current model, it is highly unlikely that they will ever develop human-like cognition.
Published in Science Robotics, the new research was conducted by Dr Stuart Wilson, Senior Lecturer in Computational Neuroscience at the University of Sheffield, and Professor Tony Prescott, Professor of Cognitive Robotics at the University of Sheffield and Director of Sheffield Robotics. They found that despite AI having the ability to emulate how humans learn, the programs are unlikely to fully think like humans unless given the opportunity to artificially feel and sense the real world. Only if they are connected to the sensory perceptions of the real physical world through robotic connections to gather real world inputs and are designed using evolutionary principles can AI reach the next level of human-like biological smartness.
Mimicking evaluation is a tall ask
Going by current technology, no matter how large the neural networks or training datasets might grow as long as AI systems remain “disembodied”, they can never replicate accurate brain processing.The researchers also pointed out that the kind of biological intelligence which the human brain displays has evolved due to the unique evolutionary design of the brain, where it leverages sensory connections with the outside world to overcome obstacles, pick up new skills, and improve over time. This interaction between evolution and development is rarely factored into the design of AI. According to the study, unless the relationship between evolution and development is considered, AI design will not be able to break new grounds and compete with human intelligence.
Of course, generative AI systems use large neural networks to solve complex problems and come up with near-human original output. This is possiblebecause such large-language models teach the algorithm to process data in a way that is inspired by the human brain and also learn from their mistakes in order to improve and become more accurate. However, the study notes, although these models have similarities to the human brain, there are also crucial differences that stands in the way of developing biological intelligence. As a result, these AI systems are unlikely to advance to the point where they can fully think like a human brain if they “continue to be designed using the same methods”.
Reach out to the world
Cognition is definitely a complex phenomenon, and science has long struggled to fully explain it. To put it simply, cognition is the mental process of acquiring knowledge and understanding through thought, experience, and senses. This becomes possible because human brains are part of the human body – which is a physical system that directly interacts with the world. This is not possible for disembodied AIs, which may learn to recognise and generate complex patterns from datasets, but without a direct connection to the physical world they have no awareness of the world around them, or any interpretation thereof.
Generative AI systems learn in a way that is somewhat similar to supervised and unsupervised human learning. Unsupervised learning entails the system learning through trial and error, such as a human telling the chatbot an answer to a prompt was wrong and building off of that information. Supervised learning is more similar to children attending school and learning required material – AI-powered chatbots are trained on inputs that have pre-established outputs that the program learns from.
However, all of this happens in closed-box systems – and the process is fixed. Human brains, on the contrary, are made up of various subsystems which are organised in a specific configuration –or architecture. This configuration is a time-tested system that is similar in all vertebrate animals, and is the result of millions of years of evolution. Despite animals from fish to humans sharing that same architecture, AI systems are based on a totally different configuration. This is where the researchers argue that human intelligence is developed due to the complicated subsystems of the brain that all vertebrates share. This architecture of the brain, coupled with a human’s experience in the real world to learn and improve through evolution, is something that leads to biological intelligence. No AI system has ever attempted to incorporate such an architecture.
Prescott and Wilson conclude that AI research needs to focus on this aspect. Only a system that equals the human brain in architecture and functioning can aspire to attain human-like intelligence. If they continue to be created using the same techniques, present AI systems are unlikely to get to the point where they can thoroughly think like a human brain.
Robotics can provide a body for the artificial brain
But how can modern technology accomplish something that has been perfected through eons of evolutionary process? Recurrent neural network models comprising several feedback loops that are taught to produce better predictions about what can happen next, can be one effective method. Such models represent significant advancements in the field of robot adaptability. The study contends that robot AIs still need to accurately simulate how various brain subsystems interact as a part of a more comprehensive cognitive architecture.
And robotics can solve the problem of disconnect with the outer world too.Robotics can provide AI systems with the much-required connections –via sensors such as cameras and microphones and actuators such as wheels and grippers. AI systems would then be able to detect their environment and learn in a way similar to the human brain.The researchers said there has been some progress on building AI platforms for robots, but these are still a long way off from mimicking the architecture of the human brain.
In an interview with Fox News Digital, Prescott sent out a clear warning that advanced AI systems must come with “…greater transparency from companies who are developing AI, alongside better governance, which needs to be international to be effective.”
Prescott also expressed hope in the same interview that a more human-like architecture could actually help with the transparency part:
“AIs that are not transparent could behave in ways we don’t expect. By applying an understanding of how real brains control real bodies, we think these systems could be made more transparent, and we could advance towards having AIs that are better able to explain how they have made decisions……This should be possible. Just as we are able to make planes, cars and power stations safe we should be able to do the same for AIs and robots. I think this also means we will need regulation, as we have in these other industries, to ensure that safety requirements are properly addressed.”
Know more about the syllabus and placement record of our Top Ranked Data Science Course in Kolkata, Data Science course in Bangalore, Data Science course in Hyderabad, and Data Science course in Chennai.