Do you think AI can be more difficult than a human opponent?

Easily. Most multiplayer games can easily be dominated by AIs through simple routines. The classic example is SF alpha 2 akuma, who would walk up to you and if you did anything that wasn’t invincible, he’d throw you, if you tried to throw, he would dragon punch, and if you dragon punched, he would block. This an unbeatable option select that is completely unbeatable, but no human could ever do it because you’d need intensely fast reaction time. The computer is already capable of processing everything going on on every frame, having the AI decide based on that is a light load.

In Starcraft there are a ton of build orders that are incredibly hard to counter, requiring specific timing pushes, but they’re so hard to execute that human players can’t even bother with them. Brood War AI tournaments are dominated by these. It’s so much more effective to program for that, that the designers of those AIs don’t bother to have the AI scout or react to anything their opponent does, they simply program a routine where it sweeps every corner of the map until the opponent is destroyed.

In Chess or Checkers or other turn based games of perfect information, the usual routine is simply to simulate every single possibility as far into the future as is reasonable, and weight the possible moves this turn that lead into scenarios that take more pieces in more possible futures more highly than those that lose pieces or take none.

The real question is, in what field can a human opponent compete with an AI? What are AIs still poor at?

An obvious one is object recognition. So they’d probably do bad at pictionary. They’re also pretty shit at Go, mostly because simulating possibilities for go or scoring different outcomes as more/less advantageous is incredibly difficult. They’re also pretty terrible at predicting human behavior (Yomi), which lead to me beating one pretty hard in rock paper scissors. They don’t really know how to write jokes or things that we’d find aesthetically appealing. Still pretty terrible at translations. Pathfinding algorithms are still too rigid to really produce effective use of movement in combat situations. They have a tendency to be indecisive or switch routes up frequently or to be too committed to a route that becomes blocked. AI are good at finding the shortest path between two points, better than humans, provided they have all the data, but it becomes harder as movement mechanics for a game become more complex and link together more physical spaces, not to mention that computation time becomes prohibitively expensive for searches like that, where humans can parse wider fields of possibilities more easily using heuristics (a good enough solution). Computers can have algorithms modeled like heuristics but don’t have the same self awareness of when an operation would take up too much time to take an estimated result. All estimations made by computers are the designers realizing in advance the scope of the estimations to be made.

2 thoughts on “Do you think AI can be more difficult than a human opponent?

  1. This Guy April 1, 2016 / 12:03 am

    >They’re also pretty shit at Go.

    Not anymore, mate.

    Liked by 1 person

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s