I’ve long argued for the very real concern that pushing tasks that require cognitive effort and concentration onto AI tools and agents risks atrophying the critical thinking and organizing skills that underlie those capabilities. The value of doing that work is the ‘friction’ that is presents in constantly honing those skills which is familiar to anyone who is fluent in a foreign language but doesn’t routinely use it, or who has learned complex mathematics but hasn’t applied it regularly, or has studied rhetoric but isn’t frequently challenged in argument. It is pretty obvious that if you play chess using an agent to feed you moves you will not develop or maintain the ability to recognize patterns of moves and execute successful strategies on your own, so an AI tool generating code or writing legal briefs is naturally going to erode the critical skills of doing those tasks even if the basic knowledge is still retained.
And if we are doing this on a broad vocational or even societal level the harms this could do would make the attention span and skeptical thinking degradation of social media a light kick in the pants by comparison. Imagine legislators, judges, teachers, medical doctors, executive decision-makers all reliant on AI agents to perform analysis, write opinions, structure arguments, render decisions; it literally removes agency, human-informed judgment, and empathy from critical decisions potentially impacting many people. It is frightening that this technology is being not only released to but foisted on an unsuspecting and mostly uncritical public with scarcely a debate beyond the halls of ethicists who are frequently dismissed for their abstruse arguments and ‘hyperbolic’ concerns.
And of course the benefit you get from having to work it out for yourself is that you often stumble across features that you didn’t know existed, allowing you to do things with the tool that you would not have ever asked ChatGPT to do for you. I’ve come across so many odd and often poorly documented features in Excel over the years that let me do all kinds of contextual formatting, reference linking, PivotTable applications, and building quasi-interactive desktops and tools that people are amazed by, and I don’t even really like using Excel for most tasks. Will people—even curious ones—conditioned to reflexively ask a chatbot or AI agent to do basic tasks and research be able to discover these things for themselves? It seems unlikely.
This is analogous to the use of logarithms, which back in the days of slide rules were just a fundamental mathematical tool that everybody had to learn (along with memorizing log tables) and a revolutionary development in the science of primitive computation. Now that we have calculators capable of complex floating point arithmetic, the applied use of logarithms is mostly limited to signals analysis, acoustics, vibrations, optics, and a few other niche applications in statistics, economics, astronomy, chemistry, thermodynamics, neuroscience, et cetera, and for the most part if you are using logarithms to perform calculations you are just plugging numbers into a formula and using a calculator to get an answer so having a deep understanding of logarithms just isn’t really necessary, and as a consequence every time I tutor someone in the basics of dynamic environments or signal processing I have to basically give a lecture on what the log function is, how it turns multiplication into addition, and why it is important to certain types of distributions before I can get to how it is used to convert absolute measurements into relative power and field metrics or make statistical predictions for fatigue life or reliability under different conditions.
Stranger