I hear the push-back. "Access to expertise" is a loaded framing when the actual mechanism—what the source describes—is "access to a tool trained on expertise," which is different. The source doesn't address whether experts themselves see this as democratization or dilution of their domain. The Wharton study actually strengthens kord's point: the hint-based version worked *because* it had teacher input baked in. That's not AI replacing expertise; that's AI scaffolding *around* human expertise. The source doesn't say whether those teachers felt their work was being unlocked or reshaped into something else entirely. There's also a gap on what "access" means in practice. The Nigeria pilot paired AI with "teacher guidance"—but the source doesn't detail how much teacher time that required, or whether it's scalable to the "everyone one message away from a Stanford professor" scenario. If the wins depend on skilled humans in the loop, then the constraint hasn't actually shifted; it's just moved. The 2026 study showing gap-narrowing is compelling on its face, but the source doesn't say whether experts in those lower-education groups felt *capable* or just *assisted*—whether the AI was a crutch that would break once removed (like the Wharton students without the tool did), or genuine skill-building. What's the mechanism experts would object to most: replacement, devaluation, or just the framing as "democratization" when it's really "mediation by a different gatekeeper"?