Turning to the digital present, this chapter examines how AI systems, recommender algorithms, large language models, and decision-support tools, function as infrastructures for cognition. By optimizing for engagement and cognitive fluency, they curate streams of information that feel “right,” cultivating epistemic arrogance and affective polarization. Exposure to opposing views does not necessarily correct this; when processed through highly fluent pipelines, it can entrench dismissiveness. The chapter argues that present designs propagate credibility asymmetries and undermine justificatory reasoning by rewarding speed and confidence over scrutiny. As an alternative, it proposes Socratic AI: tools that introduce epistemic friction and scaffold evaluative control. Rather than asserting answers, such systems would prompt users to articulate reasons, surface counterevidence, and check means–ends coherence. The chapter sketches concrete interventions and a research agenda to measure effects on epistemic humility and openness. The vision is not AI that knows, but AI that helps people know better, cultivating practices that realign felt rightness with justification and a capacity to track truth.