Competition on Mulitple Dimensions Paul Sztorc (But mostly: Ken Arrow, Tom Schelling, and Eliezer Yudkowsky) 3/4/2018
Competition is of preeminent importance – when it takes the form of Darwinian Evolution, it is a full-blown force of nature. One cannot understand life, nor modern society, nor knowledge itself without some awareness of competition.
“Multi-factor” competition is the focus of this small note, because this phenomenon is ultimately responsible for most of modern society’s largest and most stubborn problems.
Competition usually can solve any problem, given enough time. However, when competition is on many simultaneous factors, its ability to improve our lives slows down, and sometimes even halts completely or moves backwards.
Competition typically works as follows: a “supplier” lays out some “options” for a “customer”, and the customer selects the one which (s)he most prefers. This makes the customer as happy as possible, and it also punishes the options that were not selected. Thus, competition excels on two fronts: it makes the best of an imperfect situation today, and it punishes suppliers for their errors. In a modern economy, everyone is both customer and supplier of something. The result is, literally, optimal: everyone gets to consume the most that they can, and everyone is encouraged to refine their supplies as much as possible.
The problem of “multi-factorness” occurs when the “customer” must make a selection based on many criteria at once.
Now, in a trivial sense, this situation I describe is inevitable. Any option of sufficient complexity will require multiple dimensions to fully describe. For example, a hamburger can have various flavors of meat, various types of bun, and various condiments and extras.
We could measure the relationship between each of these factors (meat, bun, condiments, extra) to a variable “taste”, and then measure the relationship between “taste” and “willingness to pay”. And then we could announce a problem of multi-factorness.
Or could we? In this example, all of those criteria could be efficiently described by a new criterion, namely: “the owners skill” at running the restaurant. Does the owner hire good chefs? Are they well-trained and well-motivated? Are good incredients purchased? Is the equipment well-maintained?
So, the act of “describing the hamburger’s bun” is really just the act of “[describing the hamburger-store-owner’s skill] when it comes to hamburger buns”. Hence, we are truly explaining only one thing, not many.
For example, notice that, if the owner sells famously terrible buns, (s)he may “make up for it” by making some other aspect of the hamburger better. Perhaps the meat is made to be better, or perhaps the burger is cheaper, or perhaps the service is faster. As each of those aspects is improved, the customer will become more and more satisfied. While that is necessarily true here, it is not necessarily true of the situations I am about to describe.
( A preview example: imagine that you are choosing where to eat, but I announce that I will violently decapitate you if you order anything but a hamburger from Burger Restaurant. We can assume that you will choose to order a burger from Burger Restaurant. Now, the better the burger tastes, the happier you will be. But suppose I instead improve the threat, by stating that [if you choose non-hamburger] you will merely have all of your limbs cut off, but be left alive. Your enjoyment of the total experience will not improve. If I made the threat worse (perhaps by threating to also kill your entire family), your enjoyment of the burger would not decrease. Improvements in one do not “make up for” deficiencies in the other. )
So, we are not ONLY interested in cases where customers make 1 selection based on >1 criteria. The criteria need to be significantly different criteria. They must not reduce to a single factor.
We can designate “supplier’s competence” as the first factor. This factor is tame and does not present us with any problems.
The second, problematic factor is almost always “network effects”. For example:
A canonical example would be a case where you, instead of having dinner alone (above), intend to have dinner with friends. Now your decision on where/what to eat will be based on two things: your desire to have the best-tasting meal, and your desire to eat at the same place that your friends are eating. Perhaps you really do not want a hamburger, but hamburgers are what a super-majority of your friends overwhelmingly prefer. You are likely to show up at the hamburger place anyway. Thus, you may be superficially “free” to make a choice in the “free market” of options, but you may still come off rather dissatisfied.
In fact, state intrusion into the free market is just one special case of this more general phenomenon. Of course, state intrusion would be more than mere “network effect”, as laws are not only “followed by most people” but also actively enforced by courts and police. But, all the same, you are free to “choose” from a global set of options that includes both “legal options” and “illegal options”; and the fact that a hamburger may be “illegal” does nothing to change how well it tastes.
In this way, the law persuades people not to do things, without changing how inherently-enjoyable those things are. Since the latter is in most cases an impossibility, this is (seemingly) the only way in which “the law” could work.
Under certain conditions (some amusing, others tragic and horrifying), every single member of a group can simultaneously be held hostage by the second factor.
For example, imagine a family reunion has historically taken place at “Restaurant R”. But imagine further that, secretly, each individual family member no longer enjoys the experience at “Restaurant R” (perhaps it now has new owners, or the owners have become lazy). And finally imagine that this family has certain shortcomings: say that they have poor communication skills, a low tolerance for dissent, a history of blindly deferring to authority figures, no appreciation for creativity or discussion, no optimism that life can be improved (and especially not through rational debate and criticism). In such a family, most family members will keep their dissatisfaction to themselves. Criticism of the choice of “Restaurant R”, if voiced, will not be appreciated for what it is (an earnest desire to help the family) and will instead be mis-interpreted negatively (as disloyalty to the family or disrespect to its decision-making authorities). As a result, it is plausible that the reunion will continue to take place at “Restaurant R” year after year, imprisoning the entire family in an experience that none of them actually enjoys.
This is an example of “loyalty”. While network effects grant one a benefit for joining a larger group, loyalty grants one a benefit just for proving their association with an arbitrary existing group. In the Facebook example above there is a little of both: facebook is the largest social networking site (network effects); and anyone who threatens to leave facebook.com is, in part, threatening to leave their friends (loyalty)!
Other Examples: Vendor “Lock-In”, Format Wars, US News College Rankings, “Trendy” Grant-Supported Academic Research, many Terrible Government Policies (esp. War on Drugs / the FDA / delayed Marriage Equality – decades of torture and death, sustained by public embarassment at having initially committed to the wrong decision).
“Bitoin Hivemind” by Paul Sztorc (2018)
“Inadequate Equilibria” by Eliezer Yudkowsky
“usually, when things suck, it’s because they suck in a way that’s a Nash equilibrium” (p32)