Designing reward structures is hard

| Monday, August 26, 2013
Did you know that President Obama has personally stopped 57 terrorist attacks? I made that number up. Maybe it's zero. Or ten thousand. And that's precisely the problem: it's really hard to reward prevention, or to reward people for preventing what would have happened otherwise (for the example, let's set aside the secrecy issues involved in security). Which of course leads me to Civilization.

I gain standing with city states for clearing camps, if they are targeted. I also gain standing every time I kill a barbarian. I gain standing for liberating workers. So of course I try my hardest to never do the first one. Why would I clear the camp for a short boost when I can farm the stream of barbarians for more rep over time? In fact, why kill them so soon? I'm better off letting them wander into the city state and kidnap workers, then killing them to liberate the worker. In fact, I'm no even rewarded for clearing camps, and I won't do so if my cities are safe, unless they are targeted by the city state. The problem is that I'm not rewarded for an outcome such as "no barbarian attacks for 20 turns" or "no workers captured for 10 turns". Instead I'm rewarded for removing problem, but only when those problems get to be bad, even when the terrible problem was readily preventable.

Similarly, other civilizations are glad if you join them in a war. However, if you win the war too much, they get suspicious. That's right: no destroying the warmongers. Apparently Montezuma is supposed to be defeated, but never permanently, as if he were the Joker. On the other hand, if you just leave them to die, that's cool. Roll in later and liberate their cities and they'll be eternally grateful. In fact, they'll be so happy that they'll vote for you for world leader, which I think is the only way to make that happen. Once again, you're better off letting terrible people do terrible thing and cleaning up afterward.

Often we try to reward the actions that typically lead to outcomes. Killing barbarians is a key step toward making a city state safe from them, and of course they're going to care only if you're killing their barbarians rather than the ones wandering into their rival's land. Similarly, clearing the camp is a way to permanently fix the problem. Yet the permanent solution is less rewarding than killing individual barbarians, despite the fact that clearing the camp is as effective as killing infinite barbarians and liberating every worker, or more so, since no worker turns are wasted.

The city state could reward outcomes, such as no barbarian attacks for 20 turns. Who do they reward? Usually the civ that clears that camp contributed something. But what if the one who cleared it just jumped in at the last moment and someone else had killed every barbarian leading up to it? The second civ certainly contributed more, what with bringing in three archers while the other guy had a scout wander through. How do you track that, damage inflicted on barbarians within two tiles of the city and camp? That seems rather complex, and in fact encourages farming. Players who are behind could try to block other civs to have a chance to get in damage themselves.

Designing reward structures is hard.

2 comments:

Kring said...

But in the end it works like real live. Why should the reward structure be changed?

Klepsacovic said...

It work like a real life failure. Why not try for a success? A flawed reality has never held back games!

Post a Comment

Comments in posts older than 21 days will be moderated to prevent spam. Comments in posts younger than 21 days will be checked for ID.

Powered by Blogger.