Artificial intelligence, the contemporary darling of technologists and investors alike, appears to have a hidden and unavoidable failure mode. The Achilles’ heel of AI has always been the raw materials used to create intelligence and such information is historical, existing and, known. AI, thus, can be considered to be an efficient mechanism to aggregate and summarize known information to aid tasks or answer questions based on them. Many are fascinated by AI’s capabilities currently as it is beginning to settle into using existing information at massive scale.
But, as we go forward we will enter a regime where future information is only created by AI agents. As AI eats only existing information, we may enter a loop driven by information regurgitating AI using its own outputs. In spite of grand claims of thinking AI and “artificial mind!,” we have seen no evidence for anything that can create something new. It is not the newness that matters, because purely random experiments can certainly create new stuff, but rather it is that if such newness extends logic, innovation and creativity. Human experience, thus far, shows no instance of ground breaking innovation by extension of known information or by random experimentation. It is possible that AI can get stuck eating its own excrement for ever.
There appears to be a hidden failure mode for AI. This could be good news for humans who may ultimately control the evolution of AI.