Social dilemma games need a facelift

The Prisoner's Dilemma is one of a handful of games that test the limits of cooperation.

The Prisoner's Dilemma is one of a handful of games that test the limits of cooperation.

Social dilemma games have been usually taught in terms of word problems to math classes, since many of them stem from game theory and a finite way to calculate a "good" strategy. In theory, the way we play these social dilemma games are how we approach living our lives. For example, The Prisoner's Dilemma (PD). This game, as depicted in the picture above, is fairly simple and involves two people. Each are taken into a separate interrogation room and are given the choice to either confess to the crime or stay silent. As the payout matrix shows, if both confess then they go to jail for 5 years. If they both stay silent then they both serve a single year. However, if one confesses and one defects, then the confessor goes free and the defector is locked up for 20 years. Of course, cooperation would allow for both people to get out of jail quickly (1 year < 5 years < 20 years) but that option to skip an entire year of jail serves as a nice reward, especially for an individual who really doesn't care about his or her partner.

Although the IPD shows a fairly important aspect of cooperation (the more you work together, the better off you can be) it inherently strikes up a competition. There is a score involved and it's possible to be "better" than your partner. In the end, no matter how well the partnership does against the average, the individual scores exist as forms of praise or shame. And I've definitely been in that situation in laser tag, where your team wins but you have the worst shooting percentage. It's a cop-out for why the team didn't do better. The fractal nature of competition blooms into bickering, and thus we find ourselves cycling back through retaliation and forgiveness.

So I began a dive into social dilemma games such as the IPD and found a handful, in hopes that there exists a game that not only includes cooperative coordination but also hides or masks the sense of individual praise.


Snowdrift (SD) is a game that assumes you and a soon-to-be partner are driving and find yourself stuck on a road now covered in snow, blocking your way. A snow plow or something will move the snow off the road but that wouldn't be for a very long time. So, you can both get out of your warm, cozy car and plow the road and be on your way. Or, one person can get out and plow the road while the other sits in the warm, cozy car.

Plow road Stay in car
Plow road 20 mins 40 mins
Stay in car 40 mins 60 mins

Now, in the point-earning scheme, defecting (staying in your car) will earn you more points, even though the cost of the points equals more time spent waiting. Also, in the point-earning scheme, there is no subtraction of points.

A plows road A stays in car
B plows road A: 200; B: 200 A: 300; B: 100
B stays in car A: 100; B: 300 A: 0; B: 0

Similarly, here is a possible point-earning scheme for the PD:

A confesses A defects
B confesses A: 200; B: 200 A: 300; B: -100
B defects A: -100; B: 300 A: 0; B: 0

As Rolf Kümmerli points out in his study, cooperation exists in more real-life applications. Snowdrift and Iterated Snowdrift (ISD) can be analogous to finishing a project: although one person slacks on work, the project will have to be completed. Using the project analogy for Prisoner's Dilemma, if one person works and one person doesn't then the person who works is rewarded and the person who didn't is punished. And although that sometimes is the case for classwork, in most real-life cooperation tasks, a single person isn't normally singled out. A team effort is usually required for success and a lack of team effort is generally what is to blame when things don't go according to plan.

Stag Hunt

Similar to SD, Stag Hunt (SH) has two hunters out hunting game. There is a deer off in the distance and if both hunters shoot for the deer, they will kill it and take it home. In this hunt there are also smaller game, in this case they are rabbits. Smaller game can be killed by a single hunter, but the deer requires both to shoot at it. So, in this case, defecting and shooting a rabbit brings home some game, but obviously not the big game.

A shoots deer A shoots rabbit
B shoots deer Bring home deer Bring home 1 rabbit
B shoots rabbit Bring home 1 rabbit Bring home 2 rabbits

And the point-earning payoff matrix:

A shoots deer A shoots rabbit
B shoots deer A: 200; B: 200 A: 100; B: 0
B shoots rabbit A: 0; B: 100 A: 100; B: 100

Unlike SD, defecting is always beneficial, but only half of what you could be earning each round. On the same token, shooting the deer (aka engaging in cooperation) is a larger risk in SH than plowing snow is in SD. SH is much more similar to PD except this dilemma considers the idea that both hunters aren't separated in different rooms. Then again, many of these games in lab settings are usually played against a computer interface, or at the very most an actor or accomplice (aka a confederate).

Pros and cons of these games in the lab

A large pro is the immense amount of research from all fields concerning these dilemmas. Also, the fact that they are fairly classic games and can be implemented relatively quicker than developing your own game.

A con against using the computer interface is the lack of actual social engagement. Then again, with a confederate, the interaction is very controlled and usually not side-by-side or face-to-face.

Being able to run two subjects at the same time (aka hyperscan two subjects) is a way to increase the naturalistic behavior while subjects are playing the games, but again this is usually done without any sort of side-by-side or face-to-face interaction.

The con for face-to-face and side-by-side interaction is the idea that the game can theoretically be "won". That is, pre-planning for upcoming trials becomes a factor that increases the likelihood of cooperation when cooperation may not have been catalyzed or even created when a part from each other.

Then again, the pro is that this is how people normally interact on a project.

Possible directions

It's not impossible to believe that there exists a study out there where people are playing one (or potentially all) of these games, first hyperscanned isolated and then in a room where communication can occur. I haven't sought out that study as of yet, but thus far I haven't found any indicating it exists yet (scoopers: at least link to my blog please!). A study involving all three of these games, both isolated and in the same room where interaction can happen, could be interesting. As Kümmerli found, ISD engendered almost 20% more cooperation actions than IPD, and that was isolated. It occurs to me that if a pre-planning stage exists (like in many cooperative video games) then the payoff and cooperation interactions will be much more increased.

With this idea, it seems as if the overarching goal for participants in an iterated version of any of these games seems a bit weak. They seem as if they are competing for total points -- which is generally a solid measure in games -- but there exists so many different aspects of even these simple games that could possibly be included in a larger payoff matrix. In most video games, and especially in games where cooperation exists, there are usually markers or indicators of progress, which usually are attached to some sort of in-game bonus. Call of Duty's multiplayer game mode has a killstreak function, where the better the killstreak (aka killing opponents without dying) then the better the in-game bonus (which usually correlates strongly with winning). Cooperation feedback isn't necessarily indicated clearly with just incremental gestures of earning points. Rather, if you block off trials, let's say by ten, and include a bonus after X amount of trials of cooperation within a block, then the incentive of cooperation should increase. Then again, if you include bonuses for defecting a certain amount of times within a block, you could see cooperation get lost.

Cooperation games I've found seem to be very easy to implement but almost too easy to find anything interesting about cooperation. Real life cooperation involves real life interaction, and not implied or inferred actions done multiple times over a computer. There are reasons why social aspects of video games increase both competitive and cooperative development within a team -- as shown by games won by teams with stronger cooperation (that's just based off of my countless hours of playing Call of Duty, not a study). Investigation into socially-driven cooperation can begin using these cooperative social dilemma games, so long as we frame and manipulate these games in order to study social interaction before and during the game, and not just the reactions to what happened in the game previous.

Leave a Reply