AI Research Is Dead, Long Live AI: A Critique of the Post-ChatGPT Field

Summary

This personal essay argues that modern AI research has become intellectually stagnant after ChatGPT, while acknowledging that the boundaries of AI have always shifted. The author contrasts their experience in natural language processing with Mike Cook’s account of older AI fields being left behind: NLP, they argue, has instead been absorbed into the current AI category. They claim that a large share of current papers presents an “agent pipeline” for a human task, often meaning a collection of prompts built around a GPT-style model. In the author’s view, this trend rests on the belief that some form of artificial general intelligence has already arrived, leaving researchers to apply presumed intelligence across society rather than discover new capabilities. The essay also criticizes the social and environmental costs of this direction and says that the field increasingly makes AI appear to possess knowledge instead of creating knowledge. It describes ChatGPT as a discursive turning point: chatbots became believable enough to be treated as stochastic oracles, shifting researchers toward interpreting model outputs, sustaining the systems, and debating catastrophic risks. The author further argues that this shift has reclassified areas such as language understanding and speech recognition as AI while relegating other NLP work, including semantic tagging and syntax parsing, to computational linguistics.