Its even harder with Alpha zero because it uses vast power to train itself but only relatively low levels to actually execute. Nothing approaching an equivalent for standard stuff.

Interestingly, it runs on graphics cards because neural networks use linear transformations and, since 3-d rendering depend on these operations, graphics cards are hard-wired to do them. So there's a question, perhaps with no clear answer, about how to create equivalent hardware for this engine and traditional engines.

Perhaps to accomodate AlphaZero there could be an "open division" competition by wire, where any program can run on any hardware and call its moves in. Classical time controls would be set by the tournament and contestants could deal with that however they choose. If AlphaZero moves instantly, too bad! Google could run AlphaZero on anything it wants, and the NSA and Bulgaria could check in with some astronomical number of cores... (who knows, the resulting program might run more accurately than Topalov in the first cycle of San Luis).

It might sound like a joke, but it is not: the revolutionary techniques used to create Alpha Zero, the famous AI chess program developed by DeepMind, are now being used to engineer an engine that runs on the PC. 
(...)
https://en.chessbase.com/post/leela-chess-zero-alphazero-for-the-pc

If this AI works for chess so well, then any subject will be solved like this, e.g. climate, material science, nanostructure technology - ?!?!