Incentive Infrastructure
A new idea machine
Humanity needs to incentivize better incentives. We need infrastructure to beat Moloch.
We still live in a society with poor outcomes. We spend time, money, and energy on nostrums—ideas that promise to solve problems but have no real impact.
Why? Because we fix surface-level problems without fixing their underlying incentives.
As Charlie Munger, the patron saint of incentives once said, "show the incentive and I'll show you the outcome."
Look at the problems of the last few decades. They're all incentive problems. The Great Financial Crisis was caused by bad incentives for bankers to underwrite risky mortgages. No skin in the game. A moral hazard. Meanwhile, the Climate Crisis was caused by companies that were incentivized to ignore the externalities of fossil fuels. No carbon tax. Just pollute.
Many of the problems of so-called late-stage capitalism and neoliberalism are problems of incentives. We live in an addiction economy. Where TikTok, Facebook, and YouTube are incentivized to addict us to screens. Where CPG companies are incentivized to addict us to food and cigarettes. More than 50% of deaths every year are commerciogenic, like obesity, smoking, and many cardiovascular diseases.
Sure, companies are known to be rife with incentive problems. But it's not like nonprofits or government are free of them. The nonprofit industrial complex is rife with corrupt, rent-seeking behavior—signaling impact while lining their pockets. Plus, many nonprofits form dependency relationships. In housing, rent control just subsidizes demand while restricting supply. In substance use, harm reduction often encourages local minima for users and their communities.
Government is no better. Around the world, we have voting systems that incentivize and give power to charismatic megalomaniacs. Social media, gerrymandering, and two-party systems all incentivize polarization. Meanwhile, politicians themselves are incentivized for Twitter likes, not policy. But even the best policy rarely gets done. Bureaucrats are incentivized to follow procedures, not drive outcomes. We're left with a multi-trillion dollar monolithic government that mostly just sits around until it brinksmanships its way to the next debt ceiling.
All of the institutions feed off each other. Companies lobby corrupt politicians for private gains. Nonprofits feed off government grant programs too. Meanwhile, governments decrease corporate taxes in a global race to the bottom. This causes brain drain and a loss of state capacity, which further decreases the trust in government and desire to fund it. The systems degrade.
Many of our biggest industries are plagued by incentive problems. Healthcare charges for treatment not prevention. Doctors are paid for expensive medicines not cheap ones. Education is incentivized for graduation rates not actual learning. Universities form cartels to price fix and restrict supply.
We often talk about Baumol's cost disease for healthcare and education. But it jacks up prices in law and finance too. Lawyers work on an hourly cost-plus basis, colluding against their clients to run up prices and force a society where everyone is lawyered up. Finance acts like a leech on the economy, pulling an ever higher percentage of GDP and corporate profits each year.
The recent dockworkers strike was caused by incentives against automation. Unions generally are one of the worst cases of incentive rot. Rubber rooms hold NYC teachers who can't be fired. During the pandemic, teachers unions held out on in-person teaching for nearly twenty-four months, causing at least two years of learning loss. Unions can be a good response to greedy owners or capricious politicians. But they can decay.
Science degrades when we incentivize proxies like the h-index, which causes endless grant cycles and only incrementally legible research. Meanwhile, our national debt is ballooning, caused by politicians buying votes now at the expense of our future children. In addition, oil companies find the cheapest way to greenwash their way into ESG funds, all while continuing to manufacture doubt about the harm from fossil fuels.
So yes, the main incentive of any system is to perpetuate itself. And we have many such systems.
In addition, we can view many of the problems of other idea machines as problems of incentives.
Effective altruism is focused on helping animal welfare, people in extreme poverty, and future generations. Why are these fields neglected? Because those people have no voice, no buying power. There's no incentive to help them You can arbitrage impact by helping those without a voice. In addition to neglectedness, effective altruism is also focused on cost-effectiveness. Why? Because information asymmetries in donor-funder relationships optimize for cheap emotional signals of impact rather than quantifiable outcomes. Again, an opportunity for impact arbitrage.
The abundance agenda is focused on increasing supply of public goods, rejecting state capture, and developing state capacity. The reason why we're mired in incapable states that just cater to niche interests is simple—they're incentivized to do so. Institutional economists have been studying this for years. There's the iron law of oligarchy (Michels), the dictatorship of the small minority (Taleb), indifferent majorities (Olson), and winning coalitions (Bueno de Mesquita). These works all tell us that in order to beat vetocracy, we need to construct political institutions which are incentivized for the provision of public goods.
Finally, incentive problems are at the heart of progress studies. Why don't we have progress? Because the NIH spends $50B on god knows what. Because solar buildout is stifled by NEPA permitting. Because nuclear is blocked by the NRC. Because people are incentivized to be conformist around a pessimistic view of the future. Because we're incentivized to anchor from a slightly better present, instead of dreaming about an abundant, ambitious future.
Taking a step back, the incentives framing is so powerful because it is based on evolution and theories of universal Darwinism. Evolution is the selection of inherited variation. Fundamentally, selection is an incentive for survival. Over the last four billion years we can look at any pattern and ask: what incentivized this? What was being selected for?
The incentive framing has different terms in different fields.
In biology we call it selection. Selection for fitness. For energy consumption. For competition. Selection for bacteria and viruses and plants and animals and monkeys. For brain size and tool use. For co-domestication. As we begin to take more control over Earth's biosphere, we need to be mindful about what we're selecting for. If we don't incentivize biodiversity, carbon mitigation, and natural capital, we will lose these things. If incentivize our microbial world into an arms race, we will get more antimicrobial resistance. If we have incentives for low-cost DNA synthesis and drug design, we will get both the COVID vaccine but also bioengineered pathogens.
In the social sciences and in culture, we have many terms for incentives. Psychology is the study of behavior. What causes it? Biases. Motives. Skinner's Behaviorism. Meanwhile, economics is the study of incentives. Microeconomics looks at what incentivizes individuals. This started with rational actors but eventually complexified into behavioral economics. Macroeconomics is full of amazing incentive ideas too. Rent-seeking behavior. Tragedy of the commons. Adverse selection. Principal-agent problems. Ricardo's comparative advantage. It's incentives all the way down.
Finally, our computation and AI can be viewed through the lens of digital evolution and incentives. Stochastic gradient descent selects for low loss functions. Backpropagation is a great way to find them. These systems are incentivized to succeed at whatever minimizes their loss function. Indeed, this is the worry of AI alignment. Can we iterative construct robust incentives that dodge instrumental convergence while maximizing corrigibility? I'm not sure. But it's incentives again.
Fundamentally, incentives and evolutionary selection answer "why" questions. Why does an organism survive? Why do humans act this way? Why do AIs produce these outputs? By focusing on the why, incentives get to the causal underpinnings and root causes of reality. It's a powerful frame.
I'm not sure the best ways to solve our incentives problems—to create excellent Incentive Infrastructure.
Many of the groups above are already doing great work on it.
But we should spend more time on it.
Perverse incentives are tricky, antimemetic beasts that hide in plain sight and are excellent at dodging our attempts at removing them. Positive incentives are hard to design but have long-lasting positive effects if done correctly.
To come back to our patron saint, Charlie Munger: "Never, ever, think about something else when you should be thinking about the power of incentives."