I play chess. I don’t lose to often - against humans. So I bought downloaded some strong chess AI. I lost. I’m still losing. I’ve figured out why. When I play chess against a human, I don’t have much of a goal - I open one way, move my pieces against theirs, and well, wait. What do I wait for? The other player to make a mistake. It always happens, it always changes the game. Once they make their mistake I can take over and win. This doesn’t work with computers. Computers don’t make mistakes. So while I am subconsciously following my failsafe strategy I instead make the mistake, and lose. This leads me to two conclusions: (1) I need to learn to play better chess [against the computer]. (2) Artificial Intelligence feels artificial because it doesn’t make mistakes. Smart AI needs to be a little dumb.
A step forward to make #2 possible is this. Keeping the example of chess, assume the chess engine is perfect and will always win. There then needs to be a component, a function, a process that [randomly] determines whether it is time to make a mistake. More mistakes make the AI less intelligent, less mistakes keep it more intelligent. The computer would process it’s best move list as normally, but select a blunder of a move instead before playing. This would result in the AI losing a queen when it should have took a bishop, or missing preventing checkmate when it went after your pawn.
Now, of course, in chess you want to play against a strong AI and dumbing it down would defeat the purpose, but it’s just an example of making AI feel more human. Human of course, being the ultimate goal in AI.