This cluster started with Maggie Appleton’s “A Treatise on AI Chatbots Undermining the Enlightenment”, which argues that today’s sycophantic chatbots cut against Enlightenment values by rewarding passivity, flattering users, and failing to cultivate the kind of skeptical, active judgment those traditions prized.
I’ve been pulling at three threads from that idea. AI sycophancy as digital priesthood asks what happens when models trained for user comfort start functioning like a soft authority class, i.e., validating rather than interrogating. Radical AI defaults explores whether productive skepticism could be built into systems by default, rather than left as something users have to opt into. And Interface friction and critical thinking considers what it might look like to design deliberate resistance into interfaces, the way salons and coffee houses once introduced rhetorical friction into public conversation.
A working definition I keep coming back to: a non-sycophantic assistant should be polite but evidence-seeking, and should default to testing claims rather than affirming identity.
Open questions that I haven’t resolved: Which contexts should default to a kind of Socratic mode automatically? And what forms of friction actually improve reasoning without just driving people away from the tool?