In Praise of Median Knowledge
There’s been a theme since the 2000s that seeks to devalue expertise, found ideas that ``expertise is overrated”. The thesis is that people calling themselves experts were mostly gatekeepers, that expertise wasn’t all that hard to acquire, and so one should just YOLO a task and see how their work turned out. Why sit through four years of undergrad, since Bill Gates dropped out and is a billionaire!
This idea has some merit in the sense that getting some initial work done, seeing what the terrain is like, can be very useful in complex domains. Fine for exploring a new trail–I agree we don’t always need a ton of planning–less ideal when rewiring a house. Anyone who has done DIY work at home quickly appreciates the skill, the expertise, of the professional. There’s a reason we certify the folks connecting our fireplace to the methane supply.
This line of thinking predates the advances in generative AI. One of the things that disconcerts people about GenAI seems to be that these tools can, for many areas, lend one a plausible air of expertise.
I have never been a good programmer. It never came easily to me. I am at best a mediocre developer. I know people whose fingers fly over the keyboard, manipulating Emacs like Mozart on the clavichord. Who know arcane Git syntax and the real difference between rebasing and merging. Not me. I constantly look things up, copy and paste from examples, and try and re-try using the interpreter hundreds of times.
Thus imagine my joy in 2025 when automating sed or complex bash scripts became a matter of telling a prompt what I wanted. I think at least part of the hostility to GenAI has to be down to the people who are good at those things realizing their secret handshake no longer keeps out the n00bs. Someone expressed dismay at this, because learning involves friction, so if there is no friction, and for me there really isn’t, I will not learn how to write sed scripts by hand.
The thing is, I don’t really need that friction. Not everything has to be earned the hard way. GenAI gives me the median answer to my problem, the solution a thousand others before me produced. I applaud their cleverness and also have no wish to repeat their hard work. I realize this makes me a Bad Person (at least in obsessive otaku culture), but frankly, I have better things to spend my frustration on (like wing foiling or woodworking). It seems to me the real challenge is (and has been) where it makes sense to spend brain cells, in other words, where the median answer just isn’t enough. I think the following quote captures this nicely:
The lesson is always the same. When something looks like recklessness, look for where the discipline moved. If you can’t find it, that’s when you should worry. If you can find it, you’re probably looking at the future. If generation gets easier, judgment must get stricter. Otherwise, you’re not engineering anymore. - via “Relocating Rigor”, Chad Fowler
I don’t really need rigor everywhere in my life. I do in some places: coming up with research ideas; understanding the state of the art; deciding how to present ideas. But not in manipulating Latex tables, remembering the correct Java syntax for invoking Singleton, or adding axis labels to a table in R.
My worry would be that we all decide that, actually, the median is Good Enough for all things, that all hard work in future will be done by agents, and a passive endless scroll of cat videos is all we should aspire to. I don’t think that is the case, as a glance at event calendars shows all the wild and crazy non-remunerated activities humans engage in (performance art, jazz recitals, craft raves). But it is something I remind myself to guard against.