The Frontier Era
The era in which deep learning rose to dominance, autonomous AI emerged as a research program in its own right, and the human-centered counter-tradition began to formalize. The intellectual ground for Hi-Centric-AI ripened even as the discipline itself remained unarticulated.
Models become the unit of analysis.
The Frontier Era opens with the rapid emergence of deep learning — image recognition, speech, translation, then language modeling, then increasingly capable foundation models. The new research program took the model as its primary unit of analysis, performance benchmarks as its evaluation criterion, and autonomy as its design goal. By the end of the era, frontier model development had become the dominant public face of artificial intelligence, with research programs pursuing systems that match or exceed human performance through unsupervised execution.
This is the autonomous AI program. It is responsible for most of the technical advances that define the contemporary AI landscape, and Hi-Centric-AI takes no oppositional stance toward it. The two disciplines are addressed to different problems.
Human-centered AI emerges as a named research program.
In parallel with the autonomous program, a counter-tradition formalized through the era. Stanford's Institute for Human-Centered AI, founded in 2019 under Fei-Fei Li and James Landay, gave the older HCI/IA tradition new institutional weight. Shneiderman's Human-Centered AI (Oxford, 2022) consolidated the framing. Wei Xu's HCAI methodological work, distill.pub's Artificial Intelligence Augmentation, and the broader HCAI literature developed alongside the frontier-model work — articulating principles, surveying methodologies, and offering a philosophical alternative to autonomy maximalism.
By 2024, the intellectual ground for Hi-Centric-AI was substantially in place. The lineage was named, the counter-tradition was institutionalized, and the failure modes of generic autonomy maximalism in regulated professional domains were increasingly visible. What was missing was a discipline of methodology — a working, applied articulation of how to actually build artificial intelligence around named human cognition in serious professional practice. That articulation is the work of the era to follow.