AI: should computer opponent simulates the wrong choices?

Current AI version uses cards/battlefield scores to make best choices (it's simulates game for current priority and tries all available actions, but targets/choices are same in all simulations). It's adds bad experience for users.

Example problem 1: AI boosts same creature all the time (it's most valuable for him) or try to destroy same opponent creature all the time (it's most dangerous for him).
Example problem 2: AI will not sacrifice own creatures/life for more powerful effects in the feature;

It's can be improved by adding error possibility. As example:
* 1% of errors (choose own creatures for bad effect and opponent's creatures for good effect);
* 10% of semi-random (boosts little creature instead big, destroy weak creature instead powerful).
* Error possibility can be changed by computer level.

The issue names no files, tests, or entry points. Start by locating the AI simulation and target-selection logic, then clarify how error rates should vary by computer level. Done should include less repetitive choices, occasional suboptimal choices, and support for sacrificing resources for stronger future effects.

Written by the indexing model from the issue text.