The Greedy Fallacy
The Orange Line on fire in Boston. Credit: John Gosselin
There is a concept in computer programming called a greedy algorithm: an algorithm that accomplishes a task by solving a problem in several stages, and at each stage making the most optimal, or “greediest”, decision. For example, breaking down change at a cash register is often just a matter of choosing the largest coin, subtracting that coin’s value from the amount of change you still have to provide, and repeating, which results in making change with the optimal (fewest) number of coins.
Greedy algorithms tend to be great: they’re often extremely simple, intuitively correct, and run quickly. However, they don’t always find the optimal solution. Consider gradient descent, a fast greedy algorithm for finding a minimum value of a function. It essentially treats the function like a surface and rolls a ball from one point into the nearest pit, which represents a local minimum point on the function. This is significantly faster than brute force computation, but it doesn’t find the lowest point, just a low point.
A bird’s eye view of gradient descent with multiple balls. Notice that some balls find the global minimum in the south, but many other balls don’t. Gpeyre, CC BY-SA 4.0 https://creativecommons.org/licenses/by-sa/4.0, via Wikimedia Commons
So the greedy algorithm finds a solution, but not the optimal solution. Nevertheless, the various flavors of gradient descent are ubiquitous in machine learning because speedy computation beats a perfect solution.
This kind of approach to problem solving, where we just do whatever is best right here and right now, has a time and a place. Unfortunately, “greedy behavior” is often misapplied so often in everyday life that I call it the greedy fallacy. The idea is fairly intuitive: running a marathon by sprinting will get you into first place at the start of the race, but you’ll soon tire out and be overtaken by marathoners who take the entire race into account, not just the present moment. Even if you reach the finish line, you almost certainly won’t come in first place, and you probably had a miserable experience in the meantime.
The greedy fallacy is when we expect an optimal solution from a greedy approach, when the greedy approach isn’t guaranteed to be the best or even work at all. It’s a bit broader than “killing the golden goose”; while that’s a specific case of a greedy strategy backfiring, greedy behavior doesn’t always result in total disaster or self-defeat. It might not even result in a strictly undesirable outcome: it just isn’t the best. A typical characteristic is also that greedy behavior is unpleasant or eventually requires some kind of difficult and costly course-correction.
A CEO might be tempted to cut costs, including wages, benefits, and workplace safety, to grow her profit margin. Unfortunately, this can result in workers quitting, workplace disasters that result in lawsuits or regulation, and — worst of all — unionization from her workers. All of these hurt her bottom line, and probably wouldn’t have happened if she hadn’t pursued profit so aggressively. Video games like Anthem and Cyberpunk 2077 are two examples: both were scathed by critics and disastrously underperformed financial expectations in large part because the developers were forced to brutally crunch.
The greedy strategy also results in bad products, too. Enshittification was coined to describe why the quality of our products is declining. The most common complaints are about pop culture. Star Wars fatigue caused Solo: A Star Wars Story to bomb at the box office despite positive reviews and resulted in several planned movies being cancelled. Magic: The Gathering fans have been complaining about a glut of new products for years, with many worried about the longterm health of the game. Treating art as monetizable intellectual property often leads to it being damaged in the eyes of fans.
Tech debt and cyberrisk are the direct results of the greedy fallacy: taking the time to invest in tech and cyber infrastructure cuts into quarterly profits, so at every point in time the incentive is to put off necessary maintenance. Public infrastructure is no different. Boston’s MBTA has suffered after decades of underinvestment and is at the point of frequent delays, derailments, and endangering its passengers. After decades of chronic underinvestment, it will cost more to repair the system than if it had been adequately funded from the start. It’s like the Vimes boot theory, but the cause is strategic poverty, not economic poverty.
In a more extreme example, the Qin dynasty punished some crimes with slave labor; other crimes, like allowing prisoners to escape, were punishable by death. One can reasonably infer these laws were enacted with the intent to increase the output of the state workforce and protect Qin rule. When one overseer, Liu Bang, accidentally allowed prisoners to escape, he realized it was in his best interest to free all of the prisoners and overthrow the Qin dynasty.
The common denominators here are bad incentives. The pursuit of short-term profit over long-term profit is, by definition, approaching business with a greedy strategy; in some sense, it’s a problem of too much “rationality”. But there are also bad incentives in politics: a politician’s approval might suffer if they devote a large portion of the budget to necessary repairs. Our pathetic lack of response to climate change is probably the most pressing consequence of our culture of greed; at no point is it profitable to do anything until most of the damage is already done.
It’s also worth noting that things like public underinvestment and tech debt aren’t necessarily signs of the greedy fallacy: we might know that a greedy course of action is suboptimal, but time or resource constraints might require us to take it anyways. Both gradient descent and A* don’t always give the best solution, but they do it quickly and without a massive resource cost.
To conclude, my point here isn’t to propose solutions to greedy thinking and the conditions that incentivize it. it’s just to provide a language to talk about this flaw in our decision making. It’s kind of like therapy: now that we have a name for our bad habit, we can talk about it and figure out how to stop.
La Carte de l’Enfer. Sandro Botticelli, Public domain, via Wikimedia Commons