It's a traditional unraveling argument.

It's only profitable for me to trade in this market if I have more knowledge (including the interpretability of the knowledge), than the average market participant.

Knowing this, I drop out if I have zero private info. This elevates the average level of knowledge among the remaining participants, which elevates the knowledge requirement to be profitable. Thus, the traders who have more limited info drop out. The process continues until only the most knowledgeable trader stays in the market.

It's true that no one knows exactly what others know, and the quality of their own info. So that's my scenario of partial unraveling. The remaining participants are those with the highest confidence levels of their advantageous knowledge. The question is, will that volume be sufficient to generate an accurate price prediction? There's also the question of whether the confidence which engenders participation is correlated with real information.