The Winograd Apostasy

In the AI Story

The social cost was real. Winograd was accused of abandoning productive research for philosophical obscurantism. Heideggerian vocabulary—Zuhandenheit, Vorhandenheit, Geworfenheit—was treated as pretentious obfuscation by researchers fluent in predicate calculus. The argument that computers could not model intelligent behavior was interpreted as counsel of despair. John Markoff called Winograd 'the first high-profile deserter from the world of AI'—a characterization capturing both the drama and the institutional betrayal. Winograd was not a philosopher who'd never built systems and could be dismissed as ignorant of technical realities; he was the builder of SHRDLU, and his critique came from inside, which made it uncomfortable and difficult to dismiss.

Yet the apostasy was not refusal. Winograd did not leave computer science—he pivoted through it, establishing Stanford's Human-Computer Interaction program and mentoring Larry Page, whose Google would become one of history's most powerful AI engines. The insight that computers cannot understand but should support human understanding became, through students who absorbed it, part of the intellectual DNA of companies building the modern internet. Google's original insight—that simple statistical techniques over vast data could produce results sophisticated AI systems could not—is recognizably Winogradian in its pragmatism. The apostate built something more durable than SHRDLU: a framework for asking the right question, which was never 'Can machines think?' but 'What do we need machines to do, given that they cannot think?'

Origin

The transformation unfolded across a decade. Winograd spent the mid-1970s attempting to extend SHRDLU's approach, publishing technical work on language as cognitive process. But each extension to broader domains hit the same wall: the moment the domain opened, the moment sentences could mean more than one thing depending on inaccessible context, the formal methods collapsed. The failure was not technical but categorical. Dreyfus's What Computers Can't Do (1972) provided philosophical vocabulary for an intuition Winograd's engineering experience had generated: the difficulty was not insufficient computation but fundamental misconception about what computation could accomplish. Flores—Chilean engineer, Allende minister, imprisoned after Pinochet's coup, exiled philosopher—provided the constructive alternative: a theory of action, communication, and coordination offering a different foundation for computing's role in human life.

Key Ideas

Rationalistic tradition critique. The unexamined foundation of classical AI—that knowledge is forming correct representations and intelligence is manipulating them—was itself a form of Vorhandenheit, mistaking theoretical stance for whole cognition.

Understanding as being-in-the-world. Genuine comprehension is not representation-formation but a mode of existence—inhabiting a world of purposes, tools, and commitments mostly transparent until breakdown forces them visible.

Design reorientation. If machines cannot understand, the question changes from 'How do we make computers think?' to 'How do we design computers that support human thinking, communication, and coordination?'

The defector's legacy. Winograd's students, including Larry Page, absorbed the critique and built companies whose pragmatic statistical methods vindicated his skepticism about classical AI while creating the infrastructure for its successor.