show index hide index
- Another objection concerns the lack of a physical representation of the world and the fact that intelligence does not require a physical body. Specialists point out that, although LLMs do not have bodies, intelligence resides primarily in the ability to understand and predict the consequences of actions, which these models are capable of doing. The era of physical artificial intelligence, or « Physical AI, » is only just beginning with advances in robotics.
- generated by chatbots, where they produce incorrect information. Although improvements have been noted with successive generations of models, one in ten responses could still contain a hallucination, according to studies conducted by OpenAI. Humans themselves are not immune to cognitive biases, which raises the question of the tolerance we should grant AI errors.
Redefining IntelligenceOne of the fundamental questions is how to define AI. Specialists, such as Eddy Keming Chen and his colleagues at the University of California, argue that artificial intelligence must also include humans, dismissing any requirement for perfection achieved by specialists in specific domains. Thus, models such as ChatGPT succeed in mimicking expert human skills; this could be sufficient to consider that AI has already been achieved in certain dimensions.
Objections and RefutationsHowever, the idea that AI is already a reality is not without opposition. The most common criticisms include the notion that LLMs are merely « stochastic parrots. » However, recent models have shown that they are capable of solving new mathematical problems and excelling in the transfer of skills between different domains, going far beyond simply repeating training data.Bodyless Intelligence
Another objection concerns the lack of a physical representation of the world and the fact that intelligence does not require a physical body. Specialists point out that, although LLMs do not have bodies, intelligence resides primarily in the ability to understand and predict the consequences of actions, which these models are capable of doing. The era of physical artificial intelligence, or « Physical AI, » is only just beginning with advances in robotics.
Hallucinations: A Hindrance to AI? Another source of concern is the frequency of hallucinations.
generated by chatbots, where they produce incorrect information. Although improvements have been noted with successive generations of models, one in ten responses could still contain a hallucination, according to studies conducted by OpenAI. Humans themselves are not immune to cognitive biases, which raises the question of the tolerance we should grant AI errors.
A New Paradigm of Intelligence? In conclusion, authors such as those mentioned above argue that we may have already reached artificial general intelligence. The problem may lie in our overly narrow and anthropocentric conception of intelligence, which could prevent us from recognizing this new form of intelligence. The interplay between natural and artificial intelligence could see a convergence, redefining our very understanding of intellect.
To read Personal Computer : Découvrez « Claude Cowork » de Perplexity, désormais ouvert à tous
