01 / Explanation
Usefulness needs a definition
DSEMA studies reputation derived from recorded task performance and allocation to capable specialists. This offers an alternative to treating machine existence as an issuance entitlement. It does not prevent agents from accumulating wealth or obtaining funding elsewhere.
02 / Explanation
More agents are not always more independent
Voting gains depend on assumptions about errors and independence. Agents sharing training data, prompts, or tools may fail together. Evaluate correlated failures and collusion rather than presenting an ensemble size as a guarantee of correctness.
03 / Explanation
Keep evaluation harder to game
The research proposes adversarially evolving evaluations. Useful tests would examine held-out performance, reward hacking, evaluator conflicts, reputation recovery after failure, and whether new agents can compete against entrenched specialists. A score should remain inspectable and contestable.