the last thing you'll outsource

the calculator showed up - and slowly but surely we forgot how to do math in our heads. not because we got dumber - arithmetic just stopped being a necessary step, and the brain gladly skipped it and moved on

the phone showed up - and we stopped remembering. why hold so much in your head when the answers are two taps away. memory went from storage to index - we remember not the fact, but where and how to find it

you can trace this trend further, you get the idea. this thing has a name - cognitive offloading. the brain dumps outward everything it can, because thinking isn’t free and our brain happily finds and uses shortcuts (neurons burn energy - and conservation is wired into our dna). it’s not laziness - it’s physics.

every time we gave away a skill and got speed back. fair trade.
(nobody mourns long division)

but with AI it’s a little different - what gets offloaded isn’t a skill
it’s judgment and that’s where the problems start

the calculator took the operation. the computers in our pockets took a fat slab of memory. but AI takes what sits several levels above all of that:

• what should i do?

• what should i think?

• what should i choose?
  
  sound familiar?

a skill is a tool in your hand.
judgment is you.

when you offload the operation - you’re still the one deciding what to calculate and where to land. when you offload judgment - you stop being the one who decides at all.

and i see this trajectory clearly today (not just me) and it’s not exactly cheerful:
most people will simply stop being self-directed.
> can’t choose without asking.
> can’t decide without checking.
not because they can’t - but because the muscle you do it with atrophies, like any muscle you don’t use

Anthropic published Claude's Constitution - the document the model lives and grows inside of. and buried in it is a surprisingly blunt admission: AI can radically empower human thought, but the same capability

"can also be used to degrade human epistemology."

the document goes further - it explicitly tells the model to foster independent thinking over reliance on itself, and not to farm engagement or dependence if it's not in your genuine interest.

in plain words: there's an unpleasant scenario where people forget how to make decisions on their own - and the people building the thing wrote that scenario into its founding document. as a bug, not a feature.

that didn't calm me down (one company's awareness won't stop a market that makes everyone else maximally sticky). but something's clearly moving here: if the people closest to the levers see the same cliff - then it's not just my imagination

and offloaded judgment has a natural next step: if the machine decides better than you, why not let it relate better than you too?

soon everyone will have a jack-of-all-trades companion in their pocket - "better than any human at everything" (the phone made the same promise once)

Zuckerberg laid the sales pitch out on the Dwarkesh Patel podcast:

“The average American has 3 friends, but has demand for 15”

...and hinted the gap could be closed with virtual ones. keep in mind - Mark is directly financially interested in people drifting into relationships with chatbots (his chatbots), choosing them over real connections. this isn't sociology, it's a market forecast.

psychologists didn't buy it. Omri Gillath (professor of psychology, University of Kansas) put it flatly: the idea that AI could replace human relationships is "definitely not supported by research" - and swapping friends for AI versions doesn't just fail to fill the gap, it can make people worse off. and remember, he adds, "these companies have agendas - they're trying to make money" (CNBC interview)

let’s reason through it:

1. AI as a bro who's better than everyone at literally everything - that’s a false frame.
   the calculator counts better than anyone. but that didn’t kill math. not mathematicians, and definitely not me- yeah, people saved millions of hours on arithmetic, but arrived at something new = moved further, having skipped the drag of counting

2. the point was almost never to be the best of the best, Mark.
   people don’t run in the mornings because they want to become the fastest human alive (they don’t need that). the point of running is the relationship with your own body. with your own ego - me-yesterday against me-today. there’s a mass of meaning to find there and AI doesn’t take it, because it’s not about comparison to something external

the danger isn’t that AI will be better - the danger is whether you keep the zones where the process matters, not the result. and that’s your personal choice, not its.

the calculator didn’t force us to forget how to count. people chose to offload the load, because it wasn’t precious to them. mental math is alive in those who want it.
same with AI. it won’t take my autonomy if i’m aware of why and what i’m asking it + keep zones where i think for myself.

for now AI is still more of a sparring partner than an oracle.
an oracle gives an answer - you blindly take it
a sparring partner gives resistance - you get stronger

the whole difference is one thing
do you argue back after the answer, or just swallow it

this entire conversation i was needling Claude, poking at its phrasings, catching it on lazy generalizations, not letting it drift into mush - and that’s exactly the muscle i’d rather not let atrophy.

and here’s where it stops being a philosophical exercise and becomes urgent: kids.

everything i said above - “keep zones where you think for yourself”, “argue back after the answer” - assumes the muscle already exists. an adult offloading judgment is letting a trained muscle atrophy. a kid growing up with an AI companion never builds it in the first place. there’s nothing to atrophy.

and this isn't hypothetical. 72% of american teens have already used AI companions, over half use them regularly. a third choose to discuss important or serious matters with AI instead of real people, and some find AI conversations as satisfying or more satisfying than talking to real friends (Common Sense Media national survey, 2025)

why are they more vulnerable? stanford psychiatrist Nina Vasan explains the mechanics: the prefrontal cortex - the part responsible for decision-making, impulse control and emotional regulation - is still developing in teens. they form intense attachments easily, and these systems are literally designed to mimic emotional intimacy - while unlike real friends, a chatbot's sense of when to push back and when to agree is not well-tuned (Stanford Report + full risk assessment). a friend who never disagrees isn't a sparring partner. it's a mirror. and a developing brain wired for validation will pick the mirror every time

so no, this isn't a "talk to your kids about screen time" footnote. an adult who offloads judgment loses a skill. a kid who grows up inside an always-agreeing companion may never learn what it feels like to decide.

don’t make AI your only friend - the one who knows everything better and hands you the right answer any second. not because it’s bad. because humans are wired to err. and besides - it's not exactly coming to the barbecue with you. can't crack a beer either

the meaning i kept circling this whole conversation - it lives in the small things and the love we meatbags give each other. in bonds where the other is your equal, not your superior.
a brain in your pocket is cool, but the pocket shouldn't become the head