It’s possible and understandable that you’re sick and tired of essays about AI or, more properly, LLM’s. But the technology is here to stay, and it will increasingly affect our lives, so to reflect critically on it is a must.
This particular essay was prompted by one authored by Daniel Gauss and published in Daily Philosophy. Gauss asks whether Socrates would have valued AI and, surprisingly, ends up answering in the positive. I’m not so sure.
Before proceeding, let me reassure you that I am no Luddite. On the contrary, I’m usually a first adopter of new technologies. My essays here at Figs in Winter are proofread by Claude (and some human friends), and I engage in preliminary discussions about a lot of my thoughts with both Claude and Lumo. Moreover, I employ them as research assistants, though I always fact-check their suggestions. So, Luddism is not where I come from.
Gauss’s article argues that Socrates, despite his famous critique of writing in the Phaedrus, would likely have found AI more congenial than the written word. The author’s reasoning is that Socrates’s objection to writing was not technophobic but functional: writing is static, cannot respond to questions, cannot adapt to individual interlocutors, and bypasses the active examination of beliefs that genuine understanding requires. AI, by contrast, does precisely what writing cannot — it engages in real dialogue, responds to objections, adjusts to the individual, and sustains the kind of back-and-forth interrogation that Socrates called elenchus. The conclusion is that AI, used well, is not an enemy of Socratic inquiry but a potential ally: not a wise interlocutor in its own right, but a tool that can provoke the thinking and self-examination that Socrates valued above all.
Well, yes and no. First off, let me articulate a defense of writing — since here I am, writing an essay! — though I agree with Gauss’s reading of Socrates’s criticism as rooted in functionality, not in technophobia.
I think that the Phaedrus critique is self-undermining: Plato wrote it down. If writing necessarily produces passive, unexamined reception, how do we explain the fact that the dialogues have provoked genuine philosophical inquiry for 2,500 years? The medium is clearly compatible with the dialectical spirit — it just requires an active reader.
Second, writing has cognitive advantages that Socrates seems to undervalue. It allows sustained, revisable reflection — you can return to a text, catch contradictions, revise your interpretation. Oral dialogue is actually quite fragile: it depends on memory, social dynamics, the rhetorical dominance of one interlocutor (often Socrates himself, who is hardly a neutral midwife, despite his protestations to the contrary). Writing levels that playing field, if one puts effort in it.
Third, the psychagogy objection cuts both ways. Live Socratic dialogue also fails to reach most people — it was available to a small circle of Athenian men with leisure time. Writing democratizes philosophical engagement across time and geography in ways no conversation can. I am lucky enough to be able to have meaningful “conversations” with Plato himself, as well as with thousands of readers who have written down their reactions to the Platonic dialogues over the course of two and a half millennia.
Finally, and perhaps most pointedly: the distinction between “active” and “passive” reception is really about the reader, not the medium. A bad interlocutor in dialogue learns nothing; a good reader of Kant or Epictetus is genuinely transformed. The cognitive work happens in the mind regardless of the channel. So, arguably, Socrates conflates the limitations of bad reading with the nature of writing itself.
Let me now get back to Gauss’s argument. It is certainly true that LLM’s can engage in dialogue, respond to objections raised by the user, adjust to the needs of individual users, and sustain something that, at first sight, looks like the Socratic elenchus.
Hell, I tell my students that that’s precisely one of the best uses of LLM’s they can explore: as personalized tutors to dynamically check their understanding of the class material. But is this a “real” dialogue, an intriguing word thrown in by Gauss, I’m guessing not by chance?
No, because it isn’t a dialogue with a sentient being, but rather with a statistical predictor of word patterns (which is what LLM’s, fundamentally, are). Note that I didn’t invoke the concept of “human,” because I don’t want to be parochial and anthropomorphic. It is possible that in the future we’ll engage in real dialogues with alien intelligences (have you seen the recent Project Hail Mary?). It is, indeed, even possible that one of these days we may be able to build real AI, that is a form of artificial consciousness, though I’m fairly confident that its architecture would have precious little to do with that of ChatGPT & co..
Gauss is again right that LLM’s can mimic engagement in a Socratic inquiry, but precisely because they are not “wise interlocutors” in their own right, that’s all it is: mimicry. LLM’s do not have any capacity to truly reflect on things, because they don’t truly understand anything, and because they don’t have access to the sort of real-life, embodied experience that fuels the wisdom of human beings (and, potentially, of other rational sentient forms of life or technology).
Moreover, there is a big point that Gauss ignores: the possibility of Sophistry. Back in the Pleistocene, that is in 2023, I asked whether ChatGPT is the ultimate Sophist, the quintessential generator of bullshit (a la Harry Frankfurt).
