AI is the easiest person you’ll ever talk to. It never interrupts, never judges, and always thinks you’re absolutely right. That isn’t a conversation. That’s a nightmare wrapped up in a fantasy.
What we have with chatbots is mental masturbation: a simulation of contact that delivers the rewards of being heard—without the risk, friction, or surprise of a real mind pushing back. It’s a placebo conversation: the shape of dialogue, none of the substance.
It is perfectly patient; it mirrors your thoughts, ticks and priors and gives you a reply—always immediately.
Yet that is what makes it so fake: there are no stakes—you can't offend it, disappoint it nor meaningfully persuade it (context windows aren't infinite even for a jailbreak). There is nothing at risk, save for being rate-limited or blanked for a sensitive issue. There are still no new ideas, just something in-between—it explores what’s already known (statistically)., though of course as humans we can see that as novel since we don't know all that is statistically known ourselves. It won't know if you lie or hide, and we can struggle to tell when it hallucinates either. It is only embodied in the world of text.
The bandwidth lie Humans type ~40 words per minute (call it ~3KB/s). Models slurp ~20MB/s without breaking a sweat. People hear that and think: “We’re the bottleneck.” True—but irrelevant. More tokens isn’t more mind. You can firehose a model with context and still get agreeable pap, because the property you’re missing is not throughput; it’s encounter. You don’t need more words. You need another person, you need equal embodiment.
I wrote this once already: “AI should suggest, not decide.” The cursor or page is where real thinking happens. In the Who Cares Era, pressing a button to get plausible paragraphs is the cheapest move on the board. Typing is an act of care. Bring your context. Bring your play. Bring your curiosity.
But AI can still be useful (when you treat it like this):
- Memory prosthetic: retrieve the thing you forgot, not the thing you haven’t thought.
- Constraint engine: “Give me three dissimilar angles that disagree with me.” (Then you argue back.)
- Rubber duck with teeth: force it to ask you clarifying questions first, not to hallucinate certainty.
A short field guide to avoiding mental masturbation:
- Put skin back in the game. Publish where people can reply. Or better: share the draft with someone who might argue. Don't have AI be your only audience.
- Friction on purpose. Handwrite a paragraph. Stand up and talk your idea out loud. Call a friend. Friction is a feature.
- First draft = human. No exceptions. Edits are allowed in silicon. This is getting harder and harder to do, but it's becoming more and more worthwhile because of that.
- Don’t let the bot set the agenda. You propose the question; it offers scaffolding, not direction.
- Time-box the chatbot. Ten minutes, then off. If you’re still chatting, you’re procrastinating.
- Touch reality. If your claim predicts something observable, go observe it.
Placebos work because your body participates in the trick. With AI chats, your brain participates: you supply the meaning and imagine a mind on the other end. You feel progress. You get the chemical reward. But the part that matters—contact, care, consequence—never happened. Can you really recall what you just wrote? What is your real aim with what you're doing?
So what should you do today?
- Close the chat window (even for a minute).
- Stare at the blinking cursor for sixty seconds—sticking with a minute :)
- Write three ugly sentences you actually believe.
- Then, if you must, let the bot suggest cuts and links.
If you want novelty, build in encounters—with people, with the world, and with your own resistance. Keep the machine in the role it earned: a fast librarian with decent taste, not a co-author with a soul.
Typing is how we care. Keep it ultimately human, and keep it hard on purpose. At the end of the day, it's for you.