From a new perspective, recently published in the journal PNAS Port , Athanasios S. Fokas explores a very important topic: the potential for artificial intelligence (AI) to reach and possibly surpass human cognitive abilities. Historically, the focus has been on evaluating computer models on their proficiency at complex tasks, such as winning at Go or engaging in conversations indistinguishable from human interactions.

According to Fokas, this approach has a significant methodological limitation. Any AI needs to be tested for every human purpose before claiming that a program thinks as well as a human. Therefore, alternative methods are needed.

Limitations of Artificial Intelligence

Additionally, the “overarching goal” focus does not take into account aspects of human thought such as emotions, subjective experience, or understanding. Moreover, AI is not truly creative: AI cannot connect disparate topics by using techniques such as metaphor and imagination to produce new outcomes that were never obvious goals.

AI models are often conceptualized as artificial neural networks, but human thought is not limited to neurons; Thinking involves the entire body and many brain cells, such as non-neuronal glial cells.

Focas argues that computation reflects a small part of conscious thought, and that conscious thought itself is only part of human cognition. A huge amount of unconscious work goes on behind the scenes. Fokas concludes that artificial intelligence is far from surpassing humans in terms of thinking.