Cognitive Debt
Creating problems for future us
I’m Tim Gorichanaz, and this is Ports, a newsletter about design and ethics. You’ll find the latest article below, followed by Ports of Call, links to things I’ve been reading and pondering recently.
Humans have a funny relationship to the future. On one hand, every action we take is oriented toward it. Often we sacrifice comfort and ease today to have even more comfort and ease tomorrow. That’s the whole basis of agriculture (plant now, eat later), which is the whole basis of civilization.
But on the other hand, we take on debt like there’s no tomorrow. Debt is the whole basis of money, which is also the whole basis of civilization.
We go into debt in all sorts of ways besides money, too. There’s infrastructural debt, like when a city rebuilds a bridge in the cheapest possible way knowing they’re just going to have to do it again and at a higher cost later. There’s relationship debt and home maintenance debt. In 1992, Ward Cunningham coined the term “technical debt” to describe our human penchant for digging holes for ourselves in the world of software development. Technical debt is when a developer or company chooses the cheap and easy path today rather than paying more for a better, longer-term solution. Like financial debt, technical debt accrues interest—higher maintenance costs and fixedness. Taking the cheaper route today can end up more expensive in the long run.
Enter Cognitive Debt
The new debt on the block is cognitive debt, invited by broad access to generative AI systems.
Today at least 20% of office workers use generative AI as part of their jobs (in effect outsourcing some of their work, subsidized by venture capital, to the underclass of data labelers and content moderators working behind the emerald curtain). While these office workers may have been promised that AI would reduce their workload, many are finding that it just intensifies their workload. Employees are working at a faster pace, taking on more tasks and working for more hours each day. Unsurprisingly, this is proving unsustainable, leading to cognitive fatigue, burnout, poor decision-making, and turnover. (Not to mention, the productivity gains, if there truly are any, are hard to pin down.)
The spread of AI agents will only make this situation worse. Instead of working step by step alongside an AI product, users of AI agents are only confronted with the outputs from complex tasks.
In one sense, AI agents turn us all into managers (a dynamic that has been unfolding at least since the 1930s), which maybe doesn’t seem so scary. But whereas in management you can ask your direct-report to explain their decisions and provide you with the right amount of context and information to refine the output yourself as needed, AI agents are imperfect predictive systems that are not fully inspectable. They do not have knowledge, per se, nor do they reason, per se, even if they generate outputs that appear to result from knowledge and reasoning. The distinction may seem philosophical and irrelevant, but it’s the difference between building your house on a solid foundation and building it on a sand dune.
The term “cognitive debt” was coined in a study on essay writing in education. The researchers had three groups of participants: the first wrote essays without any tools, the second used classic search engines, and the third used an LLM. The first group exhibited the most brain activity while working and expressed the most ownership over their work. The group relying on an LLM struggled to accurately quote their own work. “Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels,” in other words, they take on cognitive debt. The authors write:
Cognitive debt defers mental effort in the short term but results in long-term costs, such as diminished critical inquiry, increased vulnerability to manipulation, decreased creativity. When participants reproduce suggestions without evaluating their accuracy or relevance, they not only forfeit ownership of the ideas but also risk internalizing shallow or biased perspectives.
Looking outside education, Professor Margaret-Anne Storey has written a series of blog posts on cognitive debt in the software workplace, showing how it interacts with technical debt. “Technical debt lives in the code. Cognitive debt lives in people.” When you have both together, you have quite an indebted system. (The Hacker News thread discussing her latest post is also interesting.)
Lessons for Reducing Cognitive Debt
If cognitive debt was defined by analogy to technical debt, perhaps there are lessons from the world of treating technical debt that can be applied here.
One experienced programmer shares their recipe for success on Reddit: “Don’t create tech debt unless you absolutely have to. Important first step.”
Once you’ve created debt, you must know how much debt you’re in, and you should set a deadline to repay it. Prioritize repaying it. Set aside, say, 10–20% of your time to focus on doing so (rather than taking on further debt).
In the software world, teams also benefit from regular code and architecture reviews to identify potential bottlenecks—i.e., technical debt that may have been taken on unknowingly or whose impact was not foreseen.
These lessons, I think, can be readily applied to the world of cognitive debt: Don’t take it on unless you absolutely have to. Schedule time to digest your work, reflect, and catch up. Channel your AI use toward giving yourself a four-day workweek rather than a six-day one.
Debt can be easy to ignore, for a time. But like bacteria, it multiplies quickly. If you don’t pay it back, eventually you’ll face bankruptcy, which happens, as Ernest Hemingway famously put it, “Gradually, then suddenly.” If you go financially bankrupt, at least you still have your mind. If you go cognitively bankrupt, I’m not sure what you have left.
Ports of Call
A technique for producing ideas: I’m working on an academic project, which gave me cause to revisit a book I first learned about and loved back when I was studying advertising in college: A Technique for Producing Ideas, by ad executive James Webb Young, first published in 1940. Young’s technique is interesting to reflect upon in this age of generative AI because he emphasizes that new ideas rely on a broad base of knowledge.
Camera clips: A few months back I got a new camera, a Sony a6700, and I’ve been slowly getting lenses and accessories. I love Peak Design’s camera clips, which can securely hold your camera on a backpack strap or a belt. Expensive but worth it.
Calculating volume: I have been obsessed with Solvej Balle’s series of short novels, On the Calculation of Volume. They’re somewhere between sci-fi, thriller and philosophical treatise. There are seven volumes total, and in English the fourth just came out last month. If you’re curious but not ready to commit, this essay gives a nice introduction.