Robotics researcher and AI expert Rodney Brooks has recently voiced his opinion regarding the overestimation of OpenAI’s large language models, such as the popular chatbot ChatGPT.
In an interview with IEEE Spectrum, Brooks highlights the limitations of these models and argues that they are far from being capable of competing with humans intellectually. He believes that our predictions about the future of AI have been flawed and that we need to be mindful of the current constraints of AI technology.
According to Brooks, the notion that AI could attain artificial general intelligence (AGI) and operate at a similar intellectual level to humans is misguided. He emphasizes that these language models lack an underlying model of the world and have no real connection to reality. They merely rely on correlations between language.
The limitations of AI technology become apparent when attempting to generalize from a performance to a competence. While humans can quickly make judgments based on observed actions, AI systems struggle to logically infer meaning, even though they might give the impression of being capable of doing so.
Brooks asserts that the primary strength of large language models lies in their ability to generate answers that sound plausible, rather than providing accurate or reasoned responses. Unfortunately, this can easily mislead users into believing that the AI system possesses a deeper understanding than it actually does. He shares his personal experience of using these models for assistance with complex coding tasks.
Despite the models confidently providing answers, Brooks discovered that they were often incorrect. He would spend hours following the models’ suggestions, only to realize that they led him astray. This highlights the fundamental difference between intelligence and the ability to look up information.
While Brooks acknowledges the potential for future advancements in AI technology, he firmly believes that it will not culminate in AGI. He cautions against the risks involved in surpassing human intelligence with AI systems and suggests that it is better to remain mindful of the limitations of current technology.
Brooks’s cautionary stance reminds us of the current limitations of AI technology and the importance of understanding its constraints. As we move forward, it is crucial to manage our expectations and be aware of the risks associated with surpassing human intelligence through AI.