Join the discussion
Write your take first — we'll ask for email only when you're ready to publish.
- Hacker News
- > In the eternal game of cat and mouse between the goodhearted people trying to fix the incentives and the Goodhart’d people trying to exploit the incentives, I am rooting for the cats.
What lovely writing.
by nerdsniper - It's worth thinking about incentives from the standpoint of feedback control theory. Typically, you have an input, an output, a measurement of the output, and an output goal. There is some lag between the input and the output, and there may be lag between the output and the measurement of the output used to drive the controller. The output will oscillate somewhat, more if there is more lag. Too much lag will result in oscillation so bad that convergence never happens. Much of classic control theory is devoted to understanding when that will happen and what to do about it.
Incentive systems often suffer from error and excessive lag between the output and the measurement of the output. This is a known problem in quality control. See Deming, the Toyota Production System, and continuous improvement. Even if you're measuring the desired output in a reasonable way, lag in that measurement will cause oscillation, and may prevent convergence. School grades and work performance reviews are examples of systems with high lag between output and the measurement of output. That happens even if the metrics are good, but not immediate.
If the measurement method is noisy, it may have to be filtered before use. This adds to lag. (Think about how a trailing moving average works.) Many incentive systems also have that property. School tests are intermittent noisy measurements filtered out by averaging.
These are known problems with incentive systems, but not well-known problems. Even if you're measuring the right thing, you can hit these problems.
by Animats - I suspect that the author and many of the commenters have experienced the "bad project manager" effect.
MANY people have dealt with terrible project managers and therefore assume all project managers are bad. This is because great project managers are very rare.
The same is true about incentive schemes. Most of them are improperly designed and/or get hit by Goodheart's law. [0]
As a counter point, this article [1] about designing incentives for farm labor is excellent and shows that with careful thought, incentives can dramatically increase output.
I've also been reading Deming's last book [2] and he is VERY against sales commissions. Some of this is probably bad design (e.g. 100% of the commissions for a sale go to one person) but I do feel he makes a valid point that everyone at a firm should benefit from helping the customer. Highly recommend it as an alternative to the more mainstream business books.
0 - https://en.wikipedia.org/wiki/Goodhart%27s_law
1 - https://archive.is/https://slate.com/culture/2008/08/an-amaz...
by alexpotato - > My real gripe with this line of thinking is it implies that incentives should be obeyed
One individual can ignore incentives. The point of that diagnosis is that you can't build a system on the premise that most people will behave well despite incentives, because many won't. It's a point about systems, not individuals.
by gfrecvh - I remember this book: Punished by Rewards, Alfie Kohn
"The basic strategy we use for raising children, teaching students, and managing workers can be summarized in six words: do this and you'll get that. We dangle goodies (from candy bars to sales commissions) in front of people in much the same way that we train a family pet. The quick fix of rewards may seem to be effective, but manipulating people with external incentives actually kills their interest in what they are doing and lowers the quality of their work."
- The game theorists/economists have studied this stuff in depth under the rubric of "mechanism design" or "market design". The relevant concept is "incentive compatibility" where the agents in the system do the right thing (whatever that was intended to be/designed for) because any deviation from that costs them more than playing nice.
The Wikipedia article [0] is not yet great but does give the two classic examples of "second-price auctions and a simple majority vote between two choices". The Gibbard–Satterthwaite shows the difficulty in extending this to more choices (in that Arrow's theorem sort of way).
The best thing I've read on this stuff is Alvin Roth's "Who Gets What and Why" (2015) [1] which is worth a read in any case. Repugnant markets!
[0] https://en.wikipedia.org/wiki/Incentive_compatibility
[1] https://en.wikipedia.org/wiki/Alvin_E._Roth#Market_design
- Jobs are the simplest and most striking example of incentives. "Do this and you will get money, which you can spend on food and shelter".
Where a person living in abundance can choose to ignore monetary incentives, most people cannot. If you want to make behavior X be what a broad selection of people do, the only available way today is incentives. And a lot of problems we face today are exactly this sort of collective action problem - recycle, vote for good leaders, don't shit in the well.
Yes, it would be better that we lived in a world where most people were not "losers" (I.E. needs to work for money), but we're certainly not there yet.
by ball_of_lint - I really like the article but I feel like ignoring incentives always costs something. Only people who can afford the cost get to look principled, which turns having money into looking like having character. Everyone he praises in the article has/had a safety net: Sumner was a Harvard-educated senator with inherited standing, the writer himself still had a career after dropping his uni title.
The fish metaphor has the same issue. The people dying on the sidewalk outside his conference have seen worse than he has, they don't lack moral clarity but rather the freedom and capital to rebuild their lives around it. His essay treats seeing the fish and acting on it as a single moment when the acting part requires you to have capital/social safety net.
by augment_me