>>19,20
That's bad. I hate computational irreducibility.
It is possible to find models which better and better predict something.
The problem with these is that the closer they are to the solution, the slower they advance. The convergence is never linear (or if it is, the solution is already known).
Here's something I thought about: in quantum mechanics, it is theoretized that the results of the double slit experiment are like that because time flows both ways, i.e. the electrons on the wall were already there in the future and the point where time meets is where they are observed. I thought of applying the same concept to my simulations, i.e. try to "guess" the future AND the past and attempt to somehow abuse the birthday paradox to find a compromise between the past and the future. I tried this with reversible CAs but the problem of computational irreducibility still applies: it takes LONG for things to get to work, and the better they work, the longer it takes for them to get better. I saw no signs of the birthday paradox taking effect (time being square-rooted).