In distributed constraints programming, many researchers have been interested in confidentiality in multi-agent systems. One of the most known techniques is lying and biphasic communication. In this context you can have many ethical problems:
How should an ethical-agent be protected against such reasoning?
How can we save or protect fundamental rights of agents?
Who will be responsible for unexpected consequences of this false information?
How can we deal with non-ethical agents?
Thats an interesting question and in line with some recent works I saw at last years AAMAS. First, I think that mediating or representation agents could be a solution to the confidentiality problem, too, as they can easily reflect opinions of still anonymous agents. If you introduce lying, you can in parallel introduce a cost functional for the utility, so lying will dissatisfy the agent on the long-run. Another interesting aspect is forecasting, when the agents can predict that a certain lie will cost less satisfaction than the outcome of a decision might gain, he will follow the lie. Perhaps the method from Tarapore et al. http://www.ifaamas.org/Proceedings/aamas2013/docs/p23.pdf could also help to isolate such lying agents, especially if that behaviour is abnormal to the rest of the group.
Best regards, yours
Marcus
Edward Snowden worked on it for two years before he attempted a solution; just say'n!
Thanks Marcus Neuer for your answer, Now, I'm investigating the contribution of Tarapore and al. it seems to be interesting. Mediation is also a good idea to manage ethical agents, but the problem is how to detect liar agent ! So, we have to determine some parameters of abnormality in distributed constraints resolution (message contradictions, #backtrack message, ...)
- Badges
- Science method