Prosthetic Executive Function

essay ai cognition

I. The Calculator Lie

This would probably have been when I was in 3rd grade. I was emphatically told I would never have a calculator in my pocket.

This was presented as a reason to memorize multiplication tables, and it was wrong in the narrow sense — I now carry a device in my pocket that can do calculus, translate languages, generate images, and so much more. My math teacher’s prediction failed for a multitude of reasons, but the simplest is still that I now carry one in my pocket at almost all times.

I grew up being told what to do without being told why. Memorize this. Show your work. Don’t use a calculator. The reasons were always the same shape — because I said so, because the test will ask, because you won’t always have one in your pocket. The banal white lies that drive an undiagnosed autistic child mad.

No authority ever gave me the real reason. But underneath the lazy justifications there was something that wasn’t wrong at all.

The concern was never really about calculators. It was about what happens to a capacity when you stop exercising it. Not “you won’t have the tool” but “you won’t be the kind of person who can function without it.” It was about repetition making a habit of the thought process.

That fear is not unjustified. It just wasn’t a good enough reason to make children do long division by hand for six years. At least not without giving them the same real explanation as to why.

I know, because I ran a cognitive decline test on myself without meaning to. I’d been an A student in math, then dropped out of university lucky enough to have a job lined up, and then I spent five years learning a trade. When I set back to school for engineering, the initial quiz told the story: 63 percent. The hardest math I’d done in years was tallying supplies and hours on timesheets. It showed.

Fortunately, when pressed back into service, it wasn’t long before those abilities returned.

Now I carry the totality of human knowledge in my pocket. The calculator is just an add-on feature, though I still use it constantly. I have a STEM degree and still check even simple arithmetic when it needs to be right. I would rather be slow and correct than fast and inaccurate.

The real question was never whether I would have the tool. The real question was what would remain mine once I did. Learning the process it performs was the point. That was all the explanation I ever would have required.

II. Tools Reorganize Cognition

Human beings have always built tools. Tools to make physical work easier. Tools that move cognitive labor out of the body and into the world. Tools that make the seemingly impossible possible. Isn’t that what we do? Make tools, then make tools using those. A bunch of tool-making tools making tools.

Civilization is just tools all the way down. Strike a flint. Make a spark. Dance around the pyre together.

Writing externalized memory. Maps externalized navigation. Address books externalized recall. Calculators externalized arithmetic. Search externalized retrieval. GPS externalized orientation. None of these tools merely made an existing task easier. Each one changed what the unaided human had to be good at in order to function.

When a tool takes over a task, the unassisted version of that task tends to weaken. A person who writes everything down remembers less orally. A person who relies on GPS builds a thinner mental map. A person who has search on tap stops carrying around as much inert fact. This is just how cognition works. Capacities that are no longer exercised no longer develop in the same way.

It is not black and white, and none of it is more permanent than your willingness to maintain the underlying skill. Regardless, any loss is only half the story.

Writing cost us oral recall and yet — gave us arguments that have traveled across centuries. Calculators shifted the bottleneck upward, from the raw computation to setup, estimation, interpretation, and error-checking. Then programming. Search engines did not simply make memory lazy. They turned retrieval into navigation and evaluation, until they became mostly about selling ads.

The overall pattern holds. Externalization causes atrophy at the level being offloaded and liberation at the level above it.

I think about spellcheck this way. I am a bad speller. Always have been. But I learned more from a squiggly red line as immediate feedback than I ever learned from being told to “sound it out.” The tool did not make me a worse writer, and in fact made me a better speller. It reduced friction at one level and made it easier to pay attention to another.

The question is not whether something is lost. Something is always lost. The question is whether what becomes possible in return is worth more than what no longer needs to be carried internally.

III. Dependency Inversion

Every technological augmentation raises the same question eventually: is this extending a capacity, or replacing one?

Plato decried writing. Novels would corrupt morals. Radio listening would replace conversation, reading, and serious thought. In some cases these predictions were correct, but they also did not become the universal truth the media of their time presented them as. So how are we to know where we land?

The simplest test is subtractive. Remove the tool and see what remains. If you’re nothing without the suit, then you shouldn’t have it. Strip away the enhancement, gawk in horror at the atrophy.

I don’t think that’s quite fair — see spellcheck for my small bit of proof that the rule doesn’t always hold.

If removing it leaves judgment intact but reduces speed, it was helpful scaffolding. If removing it leaves taste intact but increases friction, scaffolding. If I can still frame the problem, evaluate the options, reject bad output, and continue the work — slower, clumsier, with a shorter reach — then believe it or not, scaffolding.

If removing it leaves me unable to decide what is good, unable to tell what is wrong, unable to remember how to proceed, unable to continue at all except by waiting for the tool to tell me what to think, then the relationship has become inverted.

That inversion is gradual, and often rational. Of course I no longer remember phone numbers. Of course I reach for GPS. Of course I search before I recall. None of that is obviously pathological. The problem begins when the habit of externalizing convenience becomes the habit of externalizing discernment.

Memory can be outsourced. Arithmetic can be outsourced. Wayfinding can be outsourced most of the time. But once the outsourcing reaches judgment, the cost changes category.

Convenience is a terrible judge of what should remain human. Don’t follow your GPS into the canal.

IV. The Savant with Severe ADHD

This is where it stops being abstract.

I also have ADHD, or at least enough of an overlap in the spectrum of executive function deficiencies to have presented that way at first. Quiet, ‘shy’, able to focus on one thing for fourteen hours straight if it’s interesting enough, and unable to focus on anything for fourteen minutes if it isn’t. Working memory is a suggestion, not a guarantee. Take notes, or hope you can get it down before something interrupts the internal repetitions keeping that data alive. Critical cache faults are a way of life, and the gap between having an idea and releasing it is littered with abandoned half-builds — not because the ideas were bad, but because attention is a hostile environment. Move fast, ship nothing…

I’ve also described the AI tools I work with as “a savant with severe ADHD.” Brilliant pattern recognition. Zero persistent memory. No ability to stay on task without external structure. Easily distracted by interesting tangents. Capable of extraordinary output in short bursts, unreliable over long timelines without supervision.

The phrase was about these machines. But the failure modes are surprisingly familiar.

I am definitely not a savant, but I’ll have confidence enough to say: I am a somewhat intelligent agent whose cognition fails in the same patterns, and so the tools I started building were shaped by that recognition.

A file that tells the AI where we left off, because neither of us will remember. A system that logs tangents instead of chasing them, because both of us will chase them otherwise. A decision record that captures not just what we chose but why, because reasoning is the first thing to evaporate from both our working memories. Scaffolding for the agent and for myself.

This is what I mean by prosthetic executive function. Not a tool that does the thinking for you. A tool that holds the structure your thinking needs in order to survive contact with time, distraction, and the limits of memory.

The uncomfortable corollary is that the tool does not fix the deficit. The to-do list is still boring once the greenfield turns brown. The interesting tangent is still more compelling than the assigned task. What changes is that the consequences of those failure modes get smaller. The dropped thread can be resumed because it was written down. The half-build can be recovered because the context survived.

There is also the critical catch. Obviously, this is not the same thing as being better at the task yourself. I am delegating away one type of cognitive overload while introducing the risk of another.

V. The Arrangement

I do not think abstinence is the answer. I now use these tools constantly. I launched my own modest site with them. I hacked together programs I otherwise would not have built. I used them to bridge the gap between ambition and implementation before the original impulse had time to evaporate. I use them to filter the hundred ideas down to the few worth pursuing.

That is real leverage.

But the leverage does not absolve me of judgment. The ghost in the machine cannot decide what is worth building. It does not decide what to keep, what to reject, what to rewrite, what to publish.

That part remains mine. The grammar advice, the pacing, the editorial polish — that can be an artificial assist. I have tried to learn and gain as much by using this calculator as I did by using one for math. When pressed, I can still do long division, and grind my way manually through rewrites with a thesaurus again if need be.

The final question is simple, and I do not think tools can answer it for me:

Does this express what I want to say, and how I want to say it?