I'd like to write a program, which uses Quantum Electrodynamics to simulate an electron in the presence of a point charge (ie. an hydrogen atom).
Next, I'd like to run two instances, put them in the same space and see if the simulation spontaneously forms an hydrogen molecule.
I'd then like to add a second electron, and see if the result was an Helium atom.
Keep adding electrons until the simulation can do 6 and 8 electrons (carbon and hydrogen).
Now, I'd like to put all three types of simulated atoms (H,C,O) in the same space and see if the result is organic chemistry. At this point, additional optimisations might be necessary, maybe along the lines of starting the simulation by computing simplified models and using those where full accuracy is not required.
I'd then like to continue building the system up, adding optimisations as necessary. Keep adding atoms, and seeing if ever more complex molecules result. Extend to atoms with additional electrons as required (eg. need to add Nitrogen for DNA bases).
At some point, custom hardware would be required. First GPUs/FPGAs, then custom silicon, and whatever else is state-of-the-art at the time.
See how far the optimise/extend cycle can be pushed. Amino acids, DNA, proteins, cells, organs, organisms?
Who knows how far it would get? It might remain an interesting toy, at the 0 level. The closer it is to simulating a human, the closer it is to an 9 or 10 on the "Yudkowsky Ambition Scale".
---
update: spelling