cognitive prosthetic vs cognitive crutch

by Ray Svitla

Steve Jobs called the computer “a bicycle for the mind.” he was making a specific claim: the tool amplifies human capability without replacing the underlying ability. you still pedal. the bicycle just makes the pedaling go further.

but there’s another kind of tool. one that replaces the pedaling entirely. and when the tool breaks, you can’t walk.

the difference between a cognitive prosthetic and a cognitive crutch is the difference between an AI that makes you smarter and an AI that makes you dependent. and right now, most personal AI is closer to the crutch.

the bicycle test

a bicycle makes you faster without making your legs weaker. when the bicycle breaks, you walk. slower, sure. but you walk. the capability — locomotion — still exists in you.

a cognitive prosthetic works the same way. it extends your thinking without replacing the thinking itself. after using it, you’re better at the task than before — not because the tool did it for you, but because the tool helped you see patterns you now recognize independently.

examples of cognitive prosthetics:

• a calculator doesn’t replace your understanding of math. it lets you do math at a scale that would be impractical by hand. take away the calculator, you still understand multiplication.
• spellcheck doesn’t replace your language ability. it catches what you missed. you still know how to write.
• a good notebook system (emphasis on good) doesn’t replace your thinking. it offloads storage so you can focus on synthesis. take away the notebook, you still know how to think.

the common thread: the tool amplifies an existing capability. the capability survives the tool’s absence.

the wheelchair problem

a cognitive crutch replaces capability. after using it long enough, the underlying ability atrophies.

I’m not talking about physical wheelchairs — those are essential tools for people who need them. I’m talking about the metaphorical wheelchair: a tool used by someone who can walk but stops walking because the chair is easier.

examples of cognitive crutches:

• GPS that you follow turn by turn until you lose the ability to navigate by landmarks, mental maps, or intuition. take away the GPS and you’re lost in your own city.
• autocomplete that finishes your sentences until you stop forming complete thoughts before reaching for the keyboard.
• AI that writes your emails until you realize you can’t compose a coherent paragraph without it.

the common thread: convenience became replacement. the capability didn’t just go unused — it degraded.

the personal AI spectrum

personal AI sits on a spectrum between prosthetic and crutch. the same tool can be either, depending on how it’s used.

prosthetic use of AI memory: the AI remembers your past decisions and their outcomes. when facing a similar decision, it surfaces the relevant history. you still make the decision — but with better information. take away the AI and you’d remember some of this yourself, maybe less completely.

crutch use of AI memory: the AI remembers everything so you don’t bother remembering anything. your past decisions, your values, your patterns — all externalized. take away the AI and you genuinely don’t know what you decided last time or why.

prosthetic use of AI routing: the AI matches your current state to relevant approaches from the catalog. you evaluate the suggestion, understand why it was routed, and internalize the pattern. over time, you route better yourself.

crutch use of AI routing: the AI tells you what to do and you do it. you don’t understand why this approach was suggested for this context. you just follow. the AI becomes an oracle and you become a follower.

same tool. same features. the difference is whether you’re engaging with the tool or just consuming its output.

how atrophy happens

cognitive atrophy from AI is subtle. it doesn’t feel like losing a skill. it feels like efficiency.

“why would I think about this myself when the AI can analyze it faster?” this is a reasonable question. the answer is: because the thinking itself is the capability, and outsourcing it consistently causes it to rust.

it’s the same pattern Illich identified in convivial tools : the tool crosses a threshold from serving the human to replacing the human. the human doesn’t notice because the replacement feels like service.

you don’t notice your navigation skills atrophying because each individual GPS trip is more efficient than navigating yourself. you don’t notice your writing skills atrophying because each AI-drafted email is faster. the efficiency masks the loss until the day the tool is unavailable and you realize you’ve forgotten how.

the self.md design principle

the self.md architecture is designed to be prosthetic, not crutch. this is a conscious design choice with specific implications:

the file is readable. your self.md is a markdown file you can read, understand, and write yourself. it’s not a black box model trained on your data. if the AI routing disappears, the file still makes sense to a human — to you.

the routing is transparent. when the AI suggests an approach from the catalog, the reasoning is visible. “matched because: your current tension is autonomy ↔ structure, leaning structure. this approach addresses structure-seeking without surrendering autonomy.” you can evaluate the reasoning. you can learn to route yourself.

the journal builds self-knowledge. the .journal/ doesn’t just feed the AI. it feeds you. the act of writing about your tensions, your patterns, your shifts — that act itself builds the cognitive capability. the AI reads the journal too, but the primary audience is your future self.

the burn test is regular. the architecture is designed to survive its own deletion. if that’s not true — if you’ve become dependent — the burn test catches it early.

the honest question

here’s the uncomfortable part: most people prefer crutches.

crutches are easier. they require less effort. they produce faster results. and in the short term, the results are identical — a GPS-guided trip and a mentally-navigated trip arrive at the same destination.

the difference shows up over time. over years. in the accumulated capability or accumulated atrophy. in whether you’re a person who uses AI to think better or a person who uses AI instead of thinking.

self.md can’t make this choice for you. no architecture can. you can use self.md as a prosthetic — engaging with your own identity, learning your patterns, building self-knowledge that survives the tool. or you can use it as a crutch — letting the AI manage your identity for you, following its routing without understanding, outsourcing self-knowledge to a file.

the tool offers both paths. the architecture nudges toward prosthetic. but the choice is yours.

is your AI making you stronger or making you dependent? and how would you know the difference from the inside?

→ convivial AI — Illich’s framework for tools that enhance autonomy → the discomfort principle — prosthetics include friction → the three tests — the burn test catches crutch-dependency → the narcissism trap — the crutch in its most seductive form

Ray Svitla stay evolving