Yesterday I published a model of a reputation engine for ancap that was submitted to me for review, and today I will tell you what I think about it
I will start, as usual, with a definition. Reputation is the conclusion drawn by a potential counterparty regarding a certain subject and the prospects of interacting with them, based on the full extent of their knowledge about that subject. Thus, reputation is not a property of the subject. My reputation in the eyes of socialism-chan and my reputation in the eyes of minarchism-chan are two completely different reputations, and the difference between them is no less than the difference between the concepts of “execution” and “a fine”.
Many have heard about the Chinese social credit system. The state collects a vast amount of the most diverse data about a citizen, then using an incredibly complex and opaque algorithm, it puts two and two together and concludes how reliable this citizen appears to be to the state. The higher the rating, the more freedoms it is willing to grant the citizen; the lower it is, the more it restricts them. The goal of this entire system is social engineering—the cultivation of the perfect human of the future. I don’t know if market offers for “boosting” ratings have appeared yet, but if not, they simply must appear soon.
Everyone knows about the Russian Unified State Exam (USE) system. The state forces schoolchildren to take a test of monstrous proportions, and then forces universities to admit applicants based on a single criterion: the result of this test. I don’t need to tell you what phenomenal results are demonstrated in this test by natives of certain national republics.
Examples can be multiplied, but I think it is already clear that the forceful imposition of one’s own assessments of someone’s reputation on others is not the true Tao. It’s not even about the fact that such systems will be attempted to be cheated, but about the coercion itself. In ancap, however, any system that collects certain data about people and derives assessments from it can be implemented only if people voluntarily begin to use that system.
How do I envision a near-ideal reputation system? Something like this:
— Siri, look at the proposal from Bitarch; can this guy be trusted?
— I have performed an express analysis of the proposal itself and what is available about Bitarch in open sources. Based on your previous preferences, Ancap-sama, the probability that you will trust him is 85%. The probability that he will betray your trust should cooperation begin is 7%. You specified a trust threshold of 80%; may I respond to him affirmatively, or would you like to review the details?
— Give me the details.
— Hi!
Of course, just as people try to get the maximum score on the USE or the maximum social rating in China, they will want to optimize their requests for the reputation assessment system I use—in this example, Siri. But they will strive to earn my trust in the same way now, when a lot of data has to be collected manually.
The more routine I want to remove from my life, the more typical decisions will be entrusted to my virtual consigliere. But first, one way or another, I will have to train this neural network on my preferences; otherwise, the result is unlikely to consistently please me.
As for the system you proposed, people will be willing to take its recommendations into account only if it somehow becomes very important to them that a person, for the sake of increasing their reputation, be willing to contribute to charity. I doubt such a quality would concern me if, for example, I am looking for a one-night stand or choosing a model of hiking boots. But if I need to find a home for a litter of homeless kittens, then that’s a different story.