AI Research is Dead, Long Live AI

Overview

The landscape of artificial intelligence in 2026 is defined by a profound paradox. While the technology permeates every sector of society, the traditional academic field of AI research is widely considered to be in a state of terminal decline. This analysis explores why the discipline has hollowed out, not because progress has stalled, but because the core ambition of the field—creating general intelligence—has seemingly arrived, leaving no frontier left to explore.

The Death of a Discipline

Recent commentary, inspired by Mike Cook’s observations on the current state of the field, highlights a stark reality: much of modern AI research has degenerated into derivative application rather than genuine discovery. A significant portion of contemporary papers involve creating new agent pipelines for human tasks, where the “novelty” often amounts to chaining prompts around large language models. This shift suggests that the field has collapsed under a mindset that believes artificial general intelligence (AGI) has already been achieved in some weak form.

If the release of ChatGPT represented a Kuhnian scientific revolution, the current era is “normal science” that is simultaneously socially disruptive and environmentally taxing. The intellectual labor of inventing new knowledge has been replaced by the tedious process of making AI appear to possess it. Researchers have moved from pioneering the unknown to interpreting the outputs of statistical oracles, maintaining the illusion of intelligence while the substantive science fades away.

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, keep the gods fed, and warn of the coming destroyer god.

This transition has turned researchers into interpreters of a phenomenon they no longer control. The focus has shifted from understanding how intelligence emerges to managing the societal and environmental consequences of its deployment. The field is no longer about pushing boundaries but about policing the periphery of an already-achieved horizon.

The Hollowing Out of NLP

For scholars in natural language processing (NLP), the experience is not one of being left behind, but of being completely subsumed. What was once a distinct and rigorous discipline has been absorbed by the broader, less demanding category of “AI.” Tasks that required deep linguistic understanding are now dismissed as vestiges of symbolic logic, relegated to the marginal field of computational linguistics.

The ease with which NLP was swallowed illustrates the true nature of the AI field’s historical goals. AI research has always prioritized the “applications of intelligence”—activities that Enlightenment rationalism can parse and that hold significance for capital and the state. When LLMs began passing Turing-style tests, they suddenly became plausible candidates for every human-replacing application researchers had ever dreamed of. Subtopics like natural language understanding or speech recognition were rebranded as components of AI, while syntax parsing and semantic tagging were discarded.

My department is formally named the Department of Informatics and Operations Research, a distinction that reflects the separation between optimization algorithms and what is now called AI. Yet, in the current climate, even this distinction feels archaic. The rise of LLMs has allowed AI to claim any task involving language, effectively erasing the specialized contributions of NLP researchers who spent decades building robust, interpretable models.

The Theological Turn

The reason AI research engages in constant goalpost-moving is often misunderstood. It is not merely about achieving difficult technical benchmarks; it is about fulfilling a cultural prophecy. The field has undergone a discursive shift where AI has moved from an eschatological promise to a present reality. Funding is no longer secured by pointing to a future Messiah of silicon, but by pointing backward to 2022, when chatbots became believable enough to be treated as stochastic oracles.

This shift carries heavy implications for how the field operates. The “intelligence” exhibited by these models is fundamentally statistical, rooted in language patterns rather than genuine reasoning. However, this distinction has become irrelevant to the broader mission. The prior NLP subtopics that relied on explicit logic are now seen as obstacles, while the chaotic, probabilistic nature of LLMs is celebrated as the ultimate achievement.

Consequently, the community has entered a phase of maintenance rather than innovation. The remaining tasks are:

• Interpreting the outputs of increasingly powerful models.
• Ensuring the continuous consumption of vast computational resources.
• Preparing society for the existential risks associated with these systems.

Conclusion

AI research is not dead because it has failed; it is dead because it has succeeded beyond its own intellectual capacity to sustain it. The field has traded the hard work of understanding intelligence for the easier task of managing its applications. As NLP and other specialized domains are hollowed out and absorbed, the remaining community functions more like a priesthood than a scientific discipline.

The arrival of capable language models has closed the book on the classical AI research agenda. What remains is a cycle of interpretation, resource consumption, and risk management. For those who entered the field hoping to solve the mystery of mind, the prize is a mundane administrative role in the age of the oracle.