One Day The Hardest Levels Of Your Favorite Games Won’t Have to Resort to Cheating

Shanif Dhanani
Aug 28, 2017 · 2 min read

I’m currently going through a paper where a few researchers describe their attempt to build a reinforcement learning agent to play StarCraft II. In a short sentence early on in the paper, they mention the following:

“StarCraft II includes a built-in AI which is based on a set of handcrafted rules and comes with 10 levels of difficulty (the three strongest of which cheat by getting extra resources or privileged vision).”

That got me thinking — in virtually every game that I’ve ever played, on the hardest levels, the AI basically “cheats” to compensate for the fact that it doesn’t make good decisions.

Thus far, cheating AIs have been the only way for developers to make the difficulty of their games hard enough for the best players. This unfair advantage, where the computer can receive rewards that a human player just doesn’t get, is a decent enough way to approach game development, but it falls far short of making a fun game for your everyday player.

The game of Civilizations (one of my favorites) is notorious for this. On the highest levels, AIs start out with more units, have faster build times, and get more production and more rewards than the human player ever will throughout any point in the game.

This requires the human to change their strategy. The mindset goes from one of thinking and planning and reacting to hacking the game’s vulnerabilities wherever possible (i.e. in Civilization 5, starting on a small map with no city states and playing against Venice).

Thus far, there hasn’t really been a way for game developers to make the difficulty of their games higher without having to resort to cheating. But as reinforcement learning continues to improve and make progress, one day, turning up the difficulty setting on your favorite game will simply increase the “intelligence” of the automated agent that you’re playing against.

This will be great for games and gamers, since it allows you to hone your strategy as you increase your difficulty, rather than you having to resort to hacks and completely different strategies as you progress through the difficulty settings.

I think we’re still fairly far from that day. In the paper, the researchers mention that in a 1v1 game of SC2, their agent wasn’t really able to make great progress.

But we’re getting there, and I have a lot of high hopes for the ability of reinforcement learning to change a large part of our everyday worlds, and hopefully for the better.

)

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade