Tarn Adams Rejects 'AI' Label for Dwarf Fortress

In a recent statement highlighted by industry observers, Dwarf Fortress co-creator Tarn Adams expressed frustration with the modern linguistic shift surrounding artificial intelligence, noting that the term AI has been stripped of its original utility in game development. Adams explained that he has been forced to alter his vocabulary when discussing the complex, emergent systems that govern the non-player characters in his simulation game.

They took the letters from me, Adams remarked, lamenting that he can no longer use the term dwarf AI to describe the pathfinding and decision-making algorithms of his virtual citizens. Instead, the developer now refers to these systems strictly as dwarf behavior, noting that the characters simply misbehave sometimes rather than operating under what modern audiences might mistake for machine learning or neural networks.

For game developers and software practitioners, Adams' distinction highlights a growing friction between traditional game AI—which relies on deterministic state machines, utility systems, and heuristic search algorithms—and modern generative AI technologies. As the public increasingly associates AI exclusively with large language models and neural networks, traditional developers face a challenge in communicating how classic, rule-based simulation systems function.

Dwarf Fortress, which Adams has developed for over two decades, is famous for its deep, simulated world where individual dwarves possess distinct personalities, memories, and desires. Despite the complexity of these interactions, the game relies on intricate procedural generation and rule-based programming rather than modern deep learning, making the distinction between programmed behavior and modern AI a crucial one for technical clarity.