Reputation Police

The goal of reputation police is to perform periodic or ad hoc inspections in order to detect malicious patterns in rating or staking activity. There are at least two problems to be solved along the way:

● Natural cooperation versus fake cooperation. A malicious attacker may control many agents on the network, which give each other high ratings and transfer money to each other in order to artificially inflate their reputations. We need to be able to differentiate rings of agents created maliciously from natural ones. For instance, if agent A provides agent B with image recognition service and agent B serves agent A with text summarization, and then they make payments to each other, there is no way to discern whether the services are fake or not. We cannot discount ratings made between A and B simply because they form a loop; doing so would discourage members of the community from providing mutual services. The only way to resolve this is to flag the presence of any circle or ring of this kind and then inspect the involved agents, audit their source code, or perform an “undercover investigation,” ordering services from these agents and checking that they perform valid services.

● Temporally spanning cooperations. Alice may pump Bob’s reputation up in January, and then in July, Bob pumps up Alice’s reputation. This may happen organically, or they may be collaborating to manipulate the reputation system. It should be possible to figure out if cooperation is fake or natural using the pattern recognition incorporated in our design.

Once suspicious agents are found, fraud may be confirmed with manual agent system and code inspections, automatic or manual undercover investigations, and other enforcement activities. If a suspicious agent is confirmed as fraudulent, preventive measures can be enforced in a manual, semi-automatic, or automated manner. Either the agents will be excluded from Osiris or their activity will be publicly reported to the community so that the providers of staking reputations can recall their stakes or highly reputable agents can be appointed to use their stakes for “corrective downvoting” against the guilty agents.