inspired by Mike Cook’s 2026-09-22 article “Why I love AI” (MHTML archive)MHTML is a web archiving file format that combines an HTML page and its associated resources into a single file.
This may be showing my youth, but I am surprised to learn that route optimization algorithms were once considered a part of AI research; I’ve always seen it treated as a highly distinct field, sometimes even outside the realm of informatics/CS. My department is named Département d’informatique et de recherche opérationelle, or “department of informatics and operations research”, where “operations research” is about such optimization algorithms. If in a different life I’d had the time to become invested in AI research for 20 years before things became the way they are in 2026, I could totally see myself feeling left behind like Cook feels. But critically, just as he was seeing his field left behind, I was seeing my field of interest, natural language processing, becoming subsumed by this latest instantiation of “AI research”.
I resonate with Cook’s evaluation of the deadness of modern AI research. Easily half of AI papers today are “we created a new agent pipeline to do X human task”, where “agent pipeline” means a bag of prompts to GPT-n. The field has collapsed around this because of a mindset that fundamentally believes we’ve created “artificial general intelligence”—at least in some weak sense—and that what’s left is therefore to work out how to apply this intelligence to every domain of society. If the moment of ChatGPT was a Kuhnian scientific revolution, the new “normal science” is not only societallyfiona fokus, “I don’t care how well your ‘AI’ works”. 2025-11-25, personal blog. and environmentally devastatingJosh Axelrod, “The hidden environmental cost of AI data centers”. 2026-09-04, Deutsche Welle.Paul Schütze, “The problem of sustainable AI: a critical assessment of an emerging phenomenon”. 2024, Weizenbaum Journal of the Digital Society 4: 1. (doi:10.34669/WI.WJDS/4.1.4) but also incredibly boring. The task of AI science is no longer to invent any new knowledge, but to further the process of making AI seem to have that knowledge.Olivia Guest, “Models”. 2026-06-10, Zenodo.
I arrive at this conclusion differently than Cook because my field of NLP, rather than having been left behind by AI, has been completely hollowed out and replaced by three LLMs in a trenchcoat. My particular resulting disillusionment is perhaps why I depart from Cook on the reason why AI research engages in goalpost-moving in the first place. When Cook says “AI is a field built on trying to get computers to do things that they can’t do”, what that definition hides is the category used to define that set of tasks. For the field of AI, this has always been about the “applications of intelligence” which Enlightenment rationalism can parse human activity into, especially those which are significant to capital and the state. This is why NLP was so easily subsumed by AI: passing the Turing test suddenly rendered LLMs plausible candidates for so many of the human-replacing applications AI researchers have always dreamed of. The fact that these “AIs” are statistical models of language means that prior NLP subtopics like natural language understanding (NLU) or speech recognition (ASR) are now immanently recognizable as components of AI research, while other classical areas of NLP like semantic tagging or syntax parsing are discarded as vestiges of the era of symbolic NLP, left to be categorized as “computational linguistics”.
I think the most succinct way to describe this discursive shift in AI research is that in the moment of ChatGPT, the previously eschatological phenomenon of artificial intelligence has now arrived. AI researchers no longer point to a coming Messiah for their funding, but point backward to that moment in 2022 when chatbots became believable enough we collectively started treating them like stochastic oracles containing hidden knowledge. If the oracles are here now and the gods are speaking through them, all that’s left for the scientists-turned-priests of AI is to interpret their words,Mohamed Amine Ferrag, Norbert Tihanyi, Mérouane Debbah, “From LLM reasoning to autonomous AI agents: a comprehensive review”. 2026-06-01, IEEE Access 14: 84237-84285. keep the gods fed,Callum Cant, James Muldoon, Mark Graham, Feeding the machine: the hidden human labor powering A.I. 2024-08-06, Bloomsbury Publishing. and warn of the coming destroyer god.Ashley Capoot, Arjun Kharpal, “Experts weigh in as researcher says AI has more than 10% chance of ‘killing all humans’”. 2026-09-09, CNBC.