"Incentives" is Astrology for Nerds
A short follow-up rant to intentions-based reasoning.
As explained in a recent article, reasoning about people’s intentions is a natural instinct that leads to many mistakes.
Indeed, reasoning about people’s intentions is hard. We do not have access to them, and very often, we do not have access to our own either. Furthermore, the mechanics of people’s thoughts are extremely complex, our internal lives being so rich and all.
My recommendation is to focus on aspects of reality that are easier to study, like actions. Actions are easier to study by virtue of being much more objective than intentions, and much harder to fake.
Alas, people do not heed my recommendations. And in the meantime, intentions are too complex to, with our current methods, properly scientifically reason about them.
So people still reason about intentions. But not scientifically.
Some approaches are non-scientific, like deferring to intuition, authority or personal experience. I actually like those approaches when they are self-aware. I sometimes use them myself.
For instance, when I must contend with guessing the intentions of someone, I will try to talk to them and probe them, as well as consult people who I have found better at studying the relevant social dynamics at play.
But too many are pseudo-scientific.
"Incentives" is Astrology for Nerds
There are many pseudo-sciences of intentions and personality, that are about guessing the intentions of others to predict their behaviour.
Freudian psychoanalysis and astrology are great examples of such pseudo-sciences.
Freud was a neurologist who presented his theories for how “the unconscious” mind worked before our modern understanding of science. He uses many medicine-inspired scientific sounding words, portrays his anecdotes as case studies, and big abstractions seemingly coming from generalising observed behaviour.
Similarly, astrology predates our modern understanding of science. It was deeply intertwined with astronomy, but has already been understood as a pseudo-science for centuries already. Nevertheless, it still draws its scientific sounding vocabulary from astronomy, and is always weasely about the prediction it purports to make.
Both are quite post-hoc and interpretive. They both help to make sense of what happened, but are not really fit for predictions.
They both overreach, trying to explain too many things from a single framework, from relationships and job opportunities for astrology, to dreams and politics for psychoanalysis.
Same for “incentives”.
It just feels like yet another iteration of the same thing.
Astrology right now is more “woo” coded, “spiritual but not religious” et al.
Freudian psychoanalysis feels more “old-school continental intellectual”.
And “incentives” is just the nerd version. To be clear, “nerd” is not a slur. I am much more of a nerd than I will ever be woo or old-school continental intellectual.
But nerds do use “incentives”, “efficient markets”, or “equilibria” to explain just about everything.
Even when it is contradictory.
Someone uses their money for pleasure? It’s their incentives.
To make more money? It’s their incentives.
For their family? It’s their (evolutionary) incentives.
We can make any post-hoc explanation for someone’s behaviour sound scientific by just appending “It’s the incentives” to it.
Someone we like does something bad, like a friend working at an AI company? Don’t punish them. It was their incentives.
Someone we dislike does something good? Don’t reward them. It was their incentives.
Incentives can justify any behaviour, including our own and how we react to others.
Something has been done already? It’s because the markets are efficient, and so they did the thing.
Something has never been done yet? It’s because the thing is impossible, as the markets are efficient.
Something good has been done by companies? It’s because the markets are efficient, and people want good things to happen.
Something bad has been done by companies? It’s because the markets are efficient, and the bad thing wasn’t made illegal.
Something illegal has been done by companies? It’s because the markets may be a little bit too efficient.
We can make any post-hoc explanation for why the world is the way it is sound scientific by appending “As predicted by the markets being efficient”.
Even when there are no actual markets!
Like evolution, “idea markets”, or plain old tournaments, where nothing is being traded.
They’re all metaphorical markets. Woah.
I can complain a lot about economics, but as an outsider, I’ll assume the best, and that whatever bullshit economists write on Twitter is not representative of their usual academic rigour.
So, I’ll assume that although most of academic economics is shit (as per Sturgeon’s law), there’s some good work. That the good work has well-described/mathematical models, with clear domains of validity, and honest reporting of its experimental validation.
Assuming so, I can say that the “incentives” talk is to economics what astrology is to astronomy. Bullshit explanations trying to cover up vague intuitions with a veneer of science coming from a respected field of science.
Conclusion
There’s a specific bullshit way in which “incentives” is used, which is well addressed in the article “No it’s not the incentives, it’s you”. I recommend the read.
Rant over.
Have a nice day!
Those are indeed inappropriate examples of using incentives to reason. I wonder if incentives are just another useful concept that have been overused into being a bad word by rationalist crowds.
Why is education so dysfunctional? because nobody cares. Why does nobody care? they are given no reason to care. But actually some do care a lot, how come they have such a hard time making education less dysfunctional?
My personal understanding is that education has deeply dysfunctional incentive structures, and have become accountability-avoiding machines. I have always said they need to be given systematic reasons to care in order for the people who legitimately care to be given leverage to achieve their positive change.
How do you explain how education is so dysfunctional and how it might be improved without relying on incentives?
Nice post but I disagree. Incentives are mechanistic. If [X level of performance] then [y level of comp].
The guesswork is after mapping a person/entity’s incentives. The forecaster then assigns weight to each variable: helping family, increased consumption, social status. This weighting is guesswork usually.
But it can be good guesswork! Past behavior/actions as you say, are good predictors of the future. We are able to add another layer of awareness in mapping incentives.
A man works late every holiday season to reach his bonus. We can assume he weighted the bonus highly. But this year, his child is sick/the bonus is smaller. What might he do?
Your model says he will work late. I would guess otherwise.