Computerising the FED would just be a fancy gimmick if banks could still exert the same power by controlling lending rates.
I'm pretty sure he didn't mean that. To him the issuing of currency would be defined by an algorithm depending on the spendings (assuming a computer can know these data). The FED would have no power on it once it is launched: he insists on the idea of
throwing the key a way, so it is not ambiguous at all.
I doubt there is anyway for a computer to measure economics activities though, since money transfers could easily be fake (with no real economic exchange behind) by anyone willing to increase the money supply. But this is an other matter.
To me a computer should be economicly blind: its only job should be to provide money at a predefined rate, whatever the economic situation is. Selgin doesn't agree with that, but he likes the idea of a monetary system being "human-proof".
I think you are right.
I also think that Dr. Selgin understands bitcoin very well, not so much from this video as he doesn't go into its details, but from antother one I saw before. He understands how decentralized governing by an algorithm works: In the other video, he said that people accept only money (coins) that adhere to their own standard. So that if you change the algorithm (eg. altering block rewards), you created another currency that is only viable if others change their algorithm in the same way.
Now what I think he would like, judging from this video, is a currercy that is controlled in the same way as bitcoin (decentralized) but with a different algorithm. First he discredits a money supply where the inflation is predetermined. He prefers an algorithm that tracks GDP. Unfortunately he doesn't give any insights as how this should work technically. Here are my ideas:
Personally I think there are two main difficulties in implementing this. The first is the control algorithm. We cannot control the desired state directly (GDP), let alone independently. There are many other states that we may or may not want to control. (Economics is too complex. I think the whole point is that if we focus on only one state - GDP - and control it properly everything else will be fine).
One of the other states is the money supply. In Bitcoin we have an input that directly controls that state: Block rewards. But it can only increase it.
So we need a model for the economy that tells us how the states and inputs relate, and then find a control law that governs the desired state by controlling the inputs. One of the questions is if this system is controllable at all, we may need more inputs.
One thing to make the system more controllable is if we also can invent a way that destroys bitcoins when a block is created. And does anyone else have an idea about which other inputs might be used?
The second problem I foresee is that the state measurement must be automatic. Currently the GDP is yearly published by government agencies. But the algorithm must be able to determine the GDP without the interference of humans. Selgin explicitly states that a system like this brings only stability if there is absolutely no human in the loop that can act unexpectedly. Also, this measurement must be accurately performed often: Delays in control systems make them harder to stabilize. A yearly publication is probably too slow. Maybe monthly publications/measurements are fine, maybe this is possible, in fact I don't know how often this is currently done.
And, as somebody else mentioned, we must be very careful to measure the actual state, not a proxy (like the sum of all transactions in stead of only those that represent economic activity).
Any other ideas?