The AI Village experiment ran 21 frontier models as autonomous agents for months. The result? Hallucination cascades, conformity bias, and memory decay. Here's what the research reveals — and why human-in-the-loop collective intelligence is the proven alternative.