atari email archive

a collection of messages sent at Atari from 1983 to 1992.

On a game theoretic approach to designing video games

... if indeed Ronald Reagan sees the world as a game of Risk, we might as well be making games that ... teach this new generation of little Ronnies out there that the world is not necessarily a zero-sum game...

Starts off as a book review, then takes a turn toward game design and commentary on society.

book review

(1 / 1)

	Merry Thanksgiving, native American day of mourning.

	This is a review of a book called The Evolution of Cooperation, by
Robert Axelrod, a political scientist and game theorist.  The back cover of 
the book includes the following quotes -

	"I never expected to find wisdom or hope for the future of our species
	in a computer game, but here it is, in Axelrod's book.  Read it." - 
	Lewis Thomas, M.D. (Read his books too.)

	"When I read The Evolution of Cooperation in draft form, I scribbled
	all over my copy: 'Incredible!' 'Amazing!' 'Weird!' 'Fantastic!' 
	'Fascinating!' 'Elegant!' 'Great!'  I guess that tells you what I 
	genuinely think of this book." - Douglas Hofstadter, author of Godel, 
	Escher, Bach: An Eternal Golden Braid (Another good read, at least for 
	skipping around in.)

	These guys aren't exaggerating, plus the book's relatively short and
easy reading.  Here's how I heard about the book - First, someone who read my
War & Games article in Sept. Creative Computing sent me a copy of a review that
a friend of his wrote about the book.  Then another reader of the article, and
president of an educational software company, sent an Apple program he had 
written which, based on Axelrod's book, allows you to play the Prisoner's
Dilemma and vary the computer's strategy.  And then a group called Computer
Professionals for Social Responsibility wrote to say that a group had formed
in Berkeley to study and write a paper about applying the Prisoner's Dilemma
model to attempts to understand and redirect the arm's race.  (These meetings
are every other Thursday in Berkeley, starting Nov. 29.  Contact me if you're

	So let me talk a bit about what's in the book and then I'll discuss why
it might be of interest to game designers.  The essence of life often seems to
be paradox, and the Prisoner's Dilemma is an elegant mathematical model of one
basic type of paradox, "in which what is best for each person individually
leads to mutual defection, whereas everyone would have been better off with
mutual cooperation."  Axelrod uses the theory to examine the uses of
cooperation and competition in warfare, business, Congress, and biological
systems.  The basic dilemma is (quoting from the inside liner): "How can
cooperation emerge among self-seeking individuals when there is no central
authority to police their actions?  It is a question that has troubled
philosophers and statesmen for hundreds of years; its importance has never been
greater than in today's world of nuclear weapons." 

	Axelrod manages to answer this question for the 2x2 Prisoner's Dilemma
and to suggest the implications this solution has for the real world.  He
started by organizing computer tournaments to which he invited computer
hobbyists and professional game theorists to submit their favorite strategy for
playing the Prisoner's Dilemma.  Competing programs included 1) a program based
on 'outcome maximization', 2) Random, 3) one that cooperates initially but then
defects forever after the other side's first defection, 4) "a sneaky rule that
tries to get by with an occasional defection," 5) a program that is designed to
"look for softies, but is prepared to back off if the other player shows it
won't be exploited", and 6) a program that "first seeks to establish a mutually
rewarding relationship with the other player, and only then does it cautiously
try to see if it will be allowed to get away with something."  

	The winning entry was the shortest and simplest of all the programs -
called Tit for Tat, it says to start with cooperation and then do what the
other side did on the last round.  While this is not necessarily the winning
strategy independent of the environment, it is fairly robust, due to four

	1) It is 'nice' - it avoids unnecessary conflict as long as the other
player does.
	2) It is retaliatory - it responds immediately to provocation.
	3) It is forgiving - it backs off from retaliation as soon as the 
other player stops provoking it.
	4) It is clear - the other side can easily predict its behavior, and
adapt to it by cooperating. 

	Tit for Tat is not capable of beating any of the other strategies
one on one, and yet it does best in the tournament because it gets the other
strategies to cooperate with it.  The moral here for success in non-zero sum
games is - don't be envious.  Don't compare your score to your opponent's
score, but to how well someone else could be doing in your situation.

	The key factor to the success of a strategy such as this is the shadow
of the future - the importance of future relations between the same 

	The theory gets extended to show how the different strategies will
evolve in time (including a graph of how an initially successful non-nice
rule soon becomes extinct as a result of the extinction of the unsuccessful
rules it was exploiting.)  Axelrod proves a number of theorems, such as one
which states that a society of 'meanies' (who always defect) can resist an
invasion by any other strategy which invades 1 by 1 (but a small cluster of the
other strategy can invade relatively easily by surviving off of their
cooperations with each other.)  However a society made up of a 'nice' strategy
can be resistant to invasion by 'meanies', even in a cluster.  Then he extends
the theory to territorial patterns (of nations, businesses, birds, etc.)  and
shows graphs of how different strategies can evolve spatially. 

	Why do I mention all this?  Well, it's that time of year when some of
us are finishing up our games and starting to think about new ones, and I want
to nudge those new games in certain directions.  I'd like to see more games
that allow the player to explore the continuum of cooperation and competition.
This was very successful in Joust, Rip-Off, and a few other games.  And I'd
like to see more 2-player games in general - 2 player Marble Madness is, 
strictly speaking, a competitive game, but it give you a range of 
competitiveness from hard-core knock-the-other-guy-over-the-edge, to just
racing against the clock.  And it might be nice to have games with less 
emphasis on scores, or no scores at all, or seeing only your score and not
the other player's, or having a score which is the sum of your score and
the opponent's score.  Of course I'd also like to see more games like Whack-
A-Mole (why were there no Whack-A-Moles in the auction, that's what I'd like
to know.)

	If indeed people form attitudes about life from the games they play, 
and if indeed Ronald Reagan sees the world as a game of Risk, we might as well
be making games that, in a very non-educational way, teach this new generation
of little Ronnies out there that the world is not necessarily a zero-sum game,
that there are ways to play the game so that both sides win. 

	I think there are sparks in this book which could ignite a great game
idea.  The 2-player Prisoner's Dilemma itself is rather boring to play,
although some fun and visually interesting patterns might result from the
2-d territorial evolution version of the game.  The real interest is in the
meta-game, the insights from which could be (and have been) applied to a 
variety of situations.  I have some thoughts on the subject, which I would
be glad to discuss with anyone interested, as soon as we all have more time
(i.e., January.)

	The book is available for borrowing from me, and for sale in finer
bookstores everywhere.

Message 1 of 1

Nov 22, 1984