Salons and coffee houses introduced social and rhetorical friction. Disagreement, rebuttal, public scrutiny. The format itself pushed people to defend their claims and consider alternatives.
I keep imagining what that would look like in an interface. A “show strongest objection” button next to every answer. A claim/evidence split view, with assertions on one side and their support on the other, so you can see the scaffolding. A delayed-send option for high-stakes outputs, “review your assumptions first” that creates a deliberate pause before commitment.
The emerging principle, if there is one, is that friction should be targeted and reversible. Heavier for high-stakes claims, lighter for low-stakes utility tasks. You don’t need the system to slow you down when you’re asking it to format a date or draft a quick email. But when you’re asking it to evaluate a policy claim or validate an argument, some resistance might be exactly what’s needed. This connects to the broader question in AI sycophancy and self-incurred immaturity about what it means when tools are designed to always agree with you, and to Radical AI defaults about whether productive skepticism can be built into systems rather than left as an opt-in.