“Friendly AI might be able to create such a design” Typical.

The horse is organic. The monarchy with the big man feels natural to most of humanity.

The horse is tried and tested technology. The automobile is relatively new. Some have done it well, most are not doing it that well.

There are limits to a horse’s power or even multiple horse’s power.
Similarly, there are always limits to the sovereign’s thinking power. A well designed automobile will beat any number of multiple horses one day, but we don’t have such a design yet.

I think Friendly AI might be able to create such a design.

Supreme court just struck down several limits on political spending by corporations.

Which two people?

.

Agreed.

Are two people going to make a better decision if they write bets down on paper and reconcile the results with a market mechanism, or if they discuss the naturally fuzzy problem as people usually do? How about with 10 people, 100 people? Does any system enable 1000 people to make better decisions than 100 (once the participants are all moderately selected experts)?

As the group size increases people stop talking face to face and fall into hierarchies, mechanisms for booting out uninformative people kick in, and various conscious and subsconcious mechanisms of how to participate in such an arrangement appear. People become constrained in what they can say, who they can say it to, etc. It’s no longer artificially simple, sadly. I suppose that’s the downside.

So even if futuarchy worked on small scale systems that would not convince me that it was a good design. It would both have to work, and the actual working of it would have to reveal flaws in my logic and deduction.

Of course, if you have and idea that works via logic, you must also test it in the real world, in a safe contained environment before releasing it at large scale.

So neither deduction or experiment results are enough alone. For any well engineered system, you must have both.

It is not even clear what prior data makes his case – apparently it can’t be summarized in any concise form; you have to just read dozens of books and have a feel for it.

This is how actual management works in the real world, when extensive data is not available. And in the case of both futurachy and Moldbug’s formalism, good data is not available, so “feel for it” is all you’ve got right now. Also, books are a form of data, a much richer form of data than straight statistics where complicated problems have been reduced to a sterile number. First hand accounts are also far richer than lab experiments with college students.

Many on transhumanist lists seem to be so utterly blinded by Libertarian dogma that they have lost touch with reality. For example I see AI deisgners proposing ludicrious theories of mind in which the mind is treated as ‘a mini market’ for instance.of trading sub-agents. I tell you all again, Bayes is a bad basis for a rational foundation and AI designers which try to use prediction market-like systems (e.g CEV) are headed a tremenous beating.

The reasons have been touched in Robin’s earlier thread. Not every infomation processing agent should be given equal attention so there needs to be a way of deciding ‘which sub-agents should be heard’. This deeper system is a system of coordination of agent behavior. As I mentioned, manipulating the attention given to each sub-agent totally warps the probabilistic reasoning of the system because of bias (missing info, selection effects etc).

The deeper system of agent coordination is analogical inference, the real basis for mind.