Incentives & degenerate play
by Rip Dcb
When designed institutional incentives fail, their failure is predictable; when they succeed, their success is conditional. In trying to align the interests of the individual with those of the institution within which he operates, these types of incentives can only work provisionally, doing so by optimizing for a metric that the higher-ups have decided represents, in a necessarily approximate sense, the larger interests of the institution. They will–if they’re effective–shift one’s priorities around, pushing for one task to get done before another. But designed institutional incentives surely are not capable of reconfiguring the structure of one’s value system, meaning there’s almost always a gap to be bridged between the institution’s interests and the individual’s. A structural feature of most, if not all, superorganisms.
In this way, incentives are a hyperpragmatic attempt at ensuring that gameplay goes the way the game managers would like. They take as their premise that you can’t ever enforce the spirit of the law, only the letter, and that the letter is, at best, a rough and paradoxical approximation of the spirit. Suspended Reason gives a concise overview of this spirit-into-letter problematic in Surrogation:
When an institution wishes to set up an internal game, it must convert a desired spirit of behavior into a specified letter of law.
Letter—the specification of spirit—can attempt to capture some of the shapes and guises in which spirit manifests, but can never succeed in full. And yet spirit cannot be legislated, cannot be uniformly instituted as expectation, cannot tile itself across a superorganism.
So if you can’t enforce the spirit of your project, you can at least enforce its letter. Designed institutional incentives are, in theory, a short-term solution to the long-term problem of aligning to the letter, ensuring that the tasks that need to get done today in order to keep the shop lights on do in fact get done.
We might also think of these types of incentives as institutional forms of degenerate play. Possible Modernist gave a neat definition of degenerate play not so long ago:
In brief, degenerate play can be thought of as strategies which abide by the rules of the game, but go against the spirit, or simply as strategies that make the game less fun for everyone.
But there’s a potential caveat: do designed institutional incentives go against the spirit of their institutions in the same way that degenerate play goes against the spirit of games? In a certain sense, no, as they are more spirit-irrespective rather than spirit-spiting. Their relationship to the spirit is only via the letter as “the specification of the spirit”, and given that they’re meant to optimize along the lines of the letter, any behavior promoted by an incentive that plays by the spirit of the game would be an added bonus, not a feature of the incentive’s structure.
When designed institutional incentives fail, however, they can fail badly, tailspinning out and leading to a host of unintended consequences, such as thousands of firings and millions in fines. The autopsy on these failures usually reads the same: the metric was poor, crudely defined, compromised, etc., which allowed for people to take advantage of it in their own self-interest.
In these catastrophic cases, it’s not just the letter of the game that’s violated, but the spirit as well. When Wells Fargo managers decided that ‘number of accounts opened per customer per day’ was their optimal metric for measuring short-term financial success, they almost certainly did not intend for thousands of their lower-level employees to defraud–albeit to varying extents–tens of thousands of clients. But the employees did, and did so in violation of both the letter (US law as dictated by the SEC) and the spirit (making money without defrauding the client). Similarly, when public school teachers are both rewarded and punished based on their class’s performance on mandated state- and national-level standardized tests, they may only teach the testing material (including backroad testing shortcuts) irrespective of whether or not this approach actually teaches anything (see: The Wire, season 4, for insight into problems inherent in ‘teaching to the test’). And finally, in keeping with the educational theme, when adjunct professors, who are the most tenuously situated in an already tenuous labor market, are re-signed or let go based on their students’ reviews, they might focus more on being personable and grade more easily, in order to gain an edge in the eyes of their students.
Across all three cases the metrics chosen and their resultant incentive structures promote different degenerate strategies in order to win. Degenerate plays might even be the optimal plays when working under designed institutional incentives,1 given the effort necessary for payoff versus potential punishment.
This is usually where the analysis on incentives ends. Institutions will say, “We chose a bad metric”, we might respond, “Duh, all metrics are bad!”, and we move on. But what might incentives tell us about the institutions that chose them? They deliberated on and chose metrics to be representative of their own spirits. They may not have intended to promote, for example, company-wide fraud or grade inflation, but they did, and the metric they chose to optimize for selected those behaviors based on their fit.
In the space between subjective intent and objective consequences lie the host of practical associations that undergird why we do something and how we do it. Sometimes we verbalize these practical associations to ourselves in the decision-making process, but most times we don’t, because practice works intuitively, historically, and heuristically.
This is not to say that institutions are intentionally choosing bad metrics, but that behind the metrics they choose to optimize for, behind the incentives they then design, is a buried history of past metrics and incentives, past administrative decisions and hires, past letters and spirits. Incentives don’t fail (or exist) piece-meal, but rather as individual strategies embedded in a structure of other possible strategies: the metrics that could have been chosen and the designs that could have been deployed tell us just as much about an institution’s past and present as those that are selected. That designed institutional incentives can promote degenerate play isn’t just a defect of their structure, but also a window into the specific historical tensions between the spirit, letter, and the agents within that govern policy decisions and, ultimately, define institutions.
If we stop the analysis at “all metrics are bad, thus all incentives are flawed,” we might miss the social histories obfuscated and hidden by the predictable inefficiencies of superorganisms.
In the case of Wells Fargo, fraudulently opening accounts was not the optimal play in the grand scheme of life, but it was the optimal play within the internal game of ‘working at Wells Fargo’.↩︎