[dnautics]

Because it's a silly concept. The whole point of hashing as a proof of work is that verification is faster than generation... It's a one way function. If you intend to use scientific computation as proof of work, then in most cases, verification is just as difficult as generation, so now all you've done is wasted computation by repeating your calculation each time you're verifying. In other cases (protein folding e.g.) which might be a one way function if formulated correctly the difficulty of work is not meaningfully measurable which creates uncertainty in the underlying value of the coin or the cost of mining. Another problem is that it breaks distributed trust since how can you be really sure that the problem is legitimately solved if it's scientific unknown.

[wellboy]

Why not reward computation power by the amount of kwh they have spent, regardless of if their computation was worth it or worthless.

[boulos]

kWh isn’t a good metric of computational output though. You want to encourage people to do more useful work for you; if that means they should tune their code for a fancy GPU, then kWh probably won’t correlate (a single package server CPU is about 145W, a server class GPU about 300W, but the GPU usually performs way more compute).

[dnautics]

I'm sorry, but the world does not work the way that you want it to.

[mpeg]

Then you have to verify, and as he says, you end up duplicating the work.

[dnautics]

duplicating is very optimistic. Over a distributed network, to achieve trust, it will have to be (n) where n is the average (over all time) number of agents that care about the validity of the chain. Then don't forget to multiply by k^2 where k is the expected blockchain chain length over all time. As an approximation...