While AI chatbots like ChatGPT and Gemini have gained immense popularity for their conversational abilities, Meta’s chief AI scientist Yann LeCun believes these large language models have limitations and will never be as intelligent as humans. In an interview with Financial Times, LeCun said chatbots only have surface-level understanding and lack reasoning, planning and real-world knowledge that distinguishes human intelligence.
LeCun explained that although chatbots can answer prompts accurately based on their extensive training, they don’t truly comprehend language or grasp logic and causation. Their responses are simply outputs triggered by certain inputs in the data, rather than inferences drawn from accumulated knowledge. He noted chatbots cannot apply learning to new situations or think hierarchically – essential traits of human-level general intelligence.
Despite their use as helpful tools, LeCun warned against overestimating chatbots’ potential. Unlike humans, they have no persistent memory, cannot adapt or plan ahead based on past experiences. Their safety also depends entirely on training data quality rather than an innate ability to determine right from wrong.
Moving closer to human-level intelligence, Meta’s new research group is taking a scientific approach through ‘world modeling’. This involves teaching AI systems to understand how the physical world functions in a common-sense manner, much like humans. However, developing artificial general intelligence poses immense scientific challenges beyond today’s technology.
While chatbots entertain and inform, LeCun’s insights provide a realistic perspective on their limitations compared to human intelligence. Continued research holds the key to more capable, yet reliably safe AI systems in the future.
