AI Philosophy: What Computers Still Can't Do
Share
- Hubert Dreyfus published What Computers Can't Do in 1972; What Computers Still Can't Do (MIT Press, 1992) is the updated edition incorporating a decade of neural network research.
- Bart Kosko's Fuzzy Thinking (HarperCollins, 1993) introduced fuzzy logic — systems that reason in degrees rather than binary true/false — to a general readership.
- Cyberspace/Cyberbodies/Cyberpunk (Sage, 1995) emerged from the Body & Society journal series, edited by Mike Featherstone and Roger Burrows, at the height of 1990s cyberpunk culture.
- Carl Mitcham's Thinking through Technology (University of Chicago Press, 1994) bridges philosophy of technology and engineering ethics, arguing that making and thinking are inseparable.
- All four titles predate the 2010s deep learning revolution and offer frameworks for skepticism that remain relevant in the ChatGPT era.
What Computers Still Can't Do: A Critique of Artificial Reason — Hubert Dreyfus
The foundational text for anyone who thinks AI skepticism started with GPT-4. Dreyfus, a Berkeley philosopher steeped in Heidegger and Merleau-Ponty, spent decades arguing that intelligence is embodied, contextual, and irreducible to symbol manipulation. The 1992 MIT Press edition responds to the connectionists who thought neural nets would finally prove him wrong — spoiler: he doubles down. His critique hinges on the frame problem (how machines lack common sense about what's relevant) and the situatedness of human know-how, which resists formalisation. Decades later, when LLMs confidently hallucinate facts, Dreyfus's arguments about pattern-matching versus understanding feel sharper than ever. Explore our current copy of What Computers Still Can't Do or browse more Philosophy books at Patina.Fuzzy Thinking — Bart Kosko
The counterargument to binary AI from inside the machine-learning tent. Kosko, an engineer and mathematician, wasn't rejecting computation — he was rejecting the idea that intelligence runs on ones and zeros. Fuzzy logic, developed by Lotfi Zadeh in the 1960s, allows systems to reason in shades of grey: "mostly true," "somewhat cold," "kind of fast." Kosko's 1993 book argues this mirrors how humans actually think and made the case that fuzzy systems (already controlling Japanese subway trains and washing machines) were more robust than classical AI. It's a different critique than Dreyfus's phenomenology, but it lands in the same place: strict binary reasoning is a poor model for intelligence. Explore our current copy of Fuzzy Thinking or browse more Philosophy books at Patina.Cyberspace/Cyberbodies/Cyberpunk: Cultures of Technological Embodiment — Featherstone & Burrows (eds.)
The moment when cyberculture felt like prophecy, not nostalgia. Published in 1995 by Sage as part of the Body & Society series, this anthology captures the mid-90s collision of cyberpunk fiction, VR hype, and posthuman theory. Contributors ask: what happens to bodies when identity migrates online? How does technology reshape gender, race, disability? It's less a direct AI critique than a cultural-theory excavation of what happens when we stop thinking of minds and bodies as separate. If Dreyfus questioned whether machines could think like humans, Featherstone and Burrows ask whether humans are already becoming machine-like — and whether that's liberation or erasure. Explore our current copy of Cyberspace/Cyberbodies/Cyberpunk or browse more Philosophy books at Patina.Thinking through Technology: The Path between Engineering and Philosophy — Carl Mitcham
The book that reminds you engineering is already philosophy, whether engineers admit it or not. Mitcham, a philosopher of technology, argues that the divide between making things and thinking about things is artificial. Every engineering decision — how to build a bridge, code an algorithm, design a user interface — embeds values, assumptions, and metaphysical commitments. The 1994 University of Chicago Press edition traces this argument through ancient Greek technē, Heidegger's tool-analysis, and 20th-century debates about technological determinism. It's not strictly about AI, but it's essential for understanding how technologies (including intelligent systems) shape what counts as knowledge, agency, and the good life. Explore our current copy of Thinking through Technology or browse more Philosophy books at Patina. These four titles represent the last serious wave of AI skepticism before the field pivoted to deep learning and rebranded as "machine learning" to escape its baggage. Dreyfus questioned symbolic reasoning; Kosko offered fuzzy alternatives; Featherstone and Burrows tracked the cultural fallout; Mitcham reminded us that philosophy and engineering were never separate. As of August 2026, Patina's philosophy collection includes rotating copies of all four — secondhand witnesses to a debate that never really ended.What is Hubert Dreyfus's main argument against artificial intelligence?
Dreyfus argues that human intelligence is embodied, context-dependent, and grounded in a pre-reflective grasp of situations — what Heidegger called "being-in-the-world." Symbolic AI (and later, neural networks) can't replicate this because they lack bodies, cultural backgrounds, and the ability to distinguish relevant from irrelevant information without exhaustive rule-sets. His critique, rooted in phenomenology, suggests computers can simulate reasoning but not genuinely understand. Patina's copy of What Computers Still Can't Do lays out the full argument across 400 pages of footnoted philosophical combat.
Where can I buy secondhand philosophy of technology books in Australia?
Patina Paperbacks stocks rotating preloved copies of philosophy of technology titles — Dreyfus, Mitcham, Kosko, and adjacent thinkers like Langdon Winner and Bruno Latour — and ships Australia-wide from Sydney. Our philosophy collection turns over regularly, so if you're hunting a specific edition (say, the 1992 MIT Press Dreyfus), check back or grab what's live while it's there. Free shipping over $29 takes the sting out of impulse buys.
Is fuzzy logic still relevant to modern AI?
Honestly, yes — just not in the way Bart Kosko imagined. Fuzzy logic fell out of fashion as deep learning ate the world, but concepts like probabilistic reasoning, uncertainty quantification, and gradient-based learning are fuzzy logic's intellectual descendants. Kosko's 1993 polemic against binary thinking remains a useful corrective when you're watching ChatGPT confidently hallucinate with no sense of "maybe." The toolset evolved, but the critique of strict true/false reasoning still lands.
What's the connection between cyberpunk fiction and AI philosophy?
Cyberpunk (Gibson, Stephenson, Sterling) and academic AI philosophy were asking the same question from different angles: what happens when minds, bodies, and machines blur? The Featherstone/Burrows anthology tracks how 1990s cyberculture — MUDs, early VR, posthuman theory — imagined identity as software you could upload, rewrite, or distribute. That's less a technical prediction than a philosophical premise about personhood and embodiment, themes Dreyfus and Mitcham were wrestling with in more sober terms.
Are pre-ChatGPT AI philosophy books still useful in 2025?
More than ever, actually. The technical details have changed — no one's defending GOFAI symbolic reasoning in 2025 — but the core questions (Can machines understand context? What is consciousness? What does embodiment mean for intelligence?) haven't budged. Dreyfus's frame problem, Kosko's critique of binary logic, and Mitcham's insistence that technology is never neutral all map cleanly onto contemporary debates about LLM alignment, hallucination, and whether ChatGPT "knows" anything or just autocompletes plausibly. These books are the philosophical receipts for arguments happening right now on Twitter.