Athanassios S. Fokas argues that AI, regardless of its developments, continues to be removed from matching human thought, because it lacks the power to completely replicate the complexity of human cognition, together with feelings, creativity, and unconscious processes.
In a brand new perspective not too long ago revealed within the journal PNAS Nexus, Athanassios S. Fokas explores a well timed query: the potential of synthetic intelligence (AI) to attain and presumably exceed human cognitive capabilities. Traditionally, the main target has been on assessing laptop fashions based mostly on their proficiency in advanced duties, like triumphing in Go or participating in conversations indistinguishable from these with people.
Based on Fokas, this strategy has a key methodological limitation. Any AI must be examined on each single conceivable human aim earlier than anybody might declare that this system was pondering in addition to a human.
Various methodologies are subsequently wanted.
The Limitations of AI
As well as, the “advanced aim” focus doesn’t seize options of human thought, corresponding to emotion, subjective expertise, or understanding.
Moreover, AI shouldn’t be really inventive: AI can’t make connections between extensively disparate subjects, utilizing strategies corresponding to metaphor and creativeness, to reach at novel outcomes that have been by no means specific objectives.
AI fashions are sometimes conceptualized as synthetic neural networks, however human pondering shouldn’t be restricted to the neurons; pondering includes all the physique, and plenty of sorts of mind cells, corresponding to glia cells, that aren’t neurons.
Fokas argues that computations mirror a small a part of aware pondering and that aware thought itself is only one a part of human cognition. An immense quantity of unconscious work goes on behind the scenes. Fokas concludes that AI is a great distance from surpassing people in thought.
Reference: “Can synthetic intelligence attain human thought?” by Athanassios S Fokas, 19 December 2023, PNAS Nexus.
DOI: 10.1093/pnasnexus/pgad409
