What Personally Frightens Me About the Development of AI Assistants

While text generators based on language models compete in the speed of producing shallow articles about job displacement, I too have something to say about the arms race of equipping humans with artificial assistants (which, for simplicity, I will henceforth refer to by the loosely relevant but well-established term "AI").

I am not afraid that I will be thrown out on the street because AI will start writing code and designing systems better than I can. It does not even matter why I am so calm: whether it is because my unique experience — in certain areas — cannot be obtained by ingesting and processing all the wisdom of the internet, or simply because I am a carefree fool. That is not the point.

There are things related to the frenzied spread of AI among the masses that genuinely, truly concern me.

Story One: The Danger of Skill Automation

To set the stage, allow me to share a couple of stories from my career as a driving enthusiast. About twenty years ago, a young lady and I went on vacation to a rather mountainous area. When I was planning the route, I asked if she had a driver's license: her license was perfectly in order, accompanied by assurances of ten years of driving experience. As a result, I set the upper limit of daily mileage at a thousand kilometers, but when the rental agency told us that for such terrain we needed a manual transmission (or a truly premium automatic, which they did not have in stock) — it turned out that my companion could not drive a manual. Not a catastrophe — we adjusted the route — but a bad taste lingered.

I too switched to an automatic long ago, but once every six months I rent a cheap stick shift and drive about five hundred kilometers, just to keep the skill alive.

The car I drive now has a ton of electronics; in particular, sensors for vehicles in adjacent lanes. If some expendable fellow on a motorcycle is riding alongside, the left part of the dashboard starts flashing like a Christmas tree, subtly helping me extend the fellow's life. "Convenient!" I said at the dealership. "Very convenient!" I kept saying for a couple of years. Until the light bulb burned out.

No, I did not stop checking the mirror before changing lanes (fortunately, losing that skill in a couple of years is not easy). But certain changes in my behavior did occur: I now glance at the mirror a couple of tenths of a second later (no motorcyclists were harmed in the process). And I turned off that disco light show for good.

Story Two: Memory Degradation in the Digital Age

People my age will confirm that in the early nineties we knew by heart the phone numbers of every girl who was easy to reach and a couple of friends. I still remember my home phone number to this day. Memorizing that many unrelated digits taught us the ability to quickly invent mnemonics. It trained our memory. It cleaned out the carburetor.

Even now I am capable of memorizing the bank account or credit card number of the person standing in front of me in the checkout line (I am joking, of course — it is easier to wait for the old lady outside and take her card by force).

Is this a useful skill? I do not know, but people with university degrees in white coats say that exercising your memory is not harmful for the prevention of Alzheimer's and other people with hard-to-pronounce surnames.

Debunking the Popular Argument

The favorite argument of people who are not frightened by AI development at all consists of a mangled Ford quote (about "they would have asked for a faster, more enduring horse") and the example of calculators not having killed mathematics. Ford said so many things in his lifetime that I will not undertake to discuss his remarks seriously, but about mathematics — I do have something to say.

Neither calculators, nor the abacus, nor the counting frame, nor computers have destroyed mathematics, of course. But the heterodyne of adequacy among people who encounter numbers in everyday life has been turned down to minimum.

I do not use food delivery services because I put together the week's menu directly in the store: the zucchini are fresher than the eggplants today, the bell peppers have started to wilt, and those lamb ribs over there are suddenly begging to be on the table instead of the usual beef tenderloin. I will start choosing food from a catalog right after I switch entirely to non-alcoholic beer and inflatable women.

So, every third time I witness the scene of "ring this up, I will tell you later if I am also taking that." I have long since been reading restaurant menus from left to right, and I am confident I have enough money for whatever I can physically carry out of a grocery store. But I continue, for mental exercise, to calculate the running total as I place items in my cart. And my error is at most within rounding.

Is this a useful skill? I do not know. But a glance at the dashboard is enough for me to know when it will be time to refuel, how much to authorize on the card to avoid a refund, or how big a certain log file will be in a month. Without making a special effort, without pulling a calculator from my pocket, in the background — I get an instant, free extrapolation of everything that can be extrapolated. That is convenient.

The Main Fear: Lazy Thinking

Back to AI. Naturally, I use assistants. Tests, documentation, JSON wrangling, and a pile of other tasks that require zero brainpower — why not delegate? And I do delegate. But I seriously fear that fairly quickly I will turn into that very girl who could only drive an automatic.

When I caught myself with my fingers typing "please extract file.tar.gz to /tmp directory," I felt uneasy. It is longer and slower than tar xf file.tar.gz -C /tmp. I lost efficiency. Even setting aside the fact that tomorrow I will forget about the -C flag, and the day after that — about xf.

The laziness that throughout my entire conscious life was the engine of progress, helping me find new, non-trivial, unconventional solutions for optimizing everything around me — that very laziness has led to my dumbing down over the past several months.

Caring About Code Quality and Learning

When I look at ugly (but working) code in a code review, I always try to talk to the person who wrote it, to show how it could be better, more elegant, more idiomatic, faster. When I see nested if statements in my pet project, generated by a soulless machine, I usually sigh furtively and shamefully press the button "yes, go ahead, pollute my project with ugly spaghetti." Because the only value in rewriting such code is to teach the person who wrote it to do better. And the assistant will forget everything as soon as the session ends.

That is why I am not afraid that AI will replace us: understanding why in this particular project RabbitMQ is preferable to Kafka is something it will not be able to do in the foreseeable future, if only because of context window limitations (and a million other reasons), and good developers are paid precisely for that — not for JSON wrangling.

I am not afraid of becoming dumber, but I am genuinely anxious about becoming so lazy that I start asking the assistant to "list the files in this directory" and turning a blind eye to the increasingly severe primitivization and spaghettification of code.

Final Reflection

"The calculator killed the ability of the vast majority of people to do mental arithmetic; pocket smartphones killed the ability to memorize information; AI is killing the ability to think."

Not even the ability, but the desire. And that desire is the only thing that still distinguishes us from an amoeba.