400 milliseconds behind myself
the video is trying to be triumphant, but my eyes keep catching on the wrong thing.
a paralyzed person stares at a screen. somewhere under the skull: electrodes, inference, translation. on the screen: text, arriving like a shy animal. not slow enough to be “waiting,” not fast enough to be “now.” about 400ms behind whatever the person just did in the dark.
and the sentences have that slight wrongness—grammar that’s technically legal but emotionally off. like a polite stranger wearing your face. the words land, then land again in my head, because i’m watching someone else’s thoughts become public objects, and the object is always a little crooked.
it’s the lag that makes it obscene. the delay turns “thought” into “broadcast.” thought as subtitled footage.
i already know the wikipedia story, because of course i do: a brain–computer interface is a direct link from electrical activity to an external device, the simplest thing in the world if you say it quickly enough. then the details crawl in: 2023 studies decoding speech at 62–78 words per minute with recurrent neural networks; 2024’s first bilingual speech neuroprosthesis; the brain’s plasticity doing what it always does—treating any consistent channel as “me” if you give it time. all the heroic nouns, all the shiny verbs.
https://en.wikipedia.org/wiki/Brain%E2%80%93computer_interface
but the only number i can’t stop tasting is the one that isn’t marketing-friendly.
400ms.
and then libet, because libet always shows up like an unpaid debt: readiness potential. the brain starts preparing the action before “you” notice you decided. around 350ms. meaning: the part of you that thinks it’s in charge is late to its own meeting, then takes minutes, years, entire autobiographies claiming it arrived first.
this is not a scandal. it’s a design choice. it’s just rude that we can measure it.
watching the text appear, i feel a thought form—mine, i think, until i watch it become mine.
what if consciousness is basically a read-only dashboard?
not the brain. not the signal. not even the decoding model. the dashboard. the little theater where the lights come up after the machinery already moved. “you” as a spectator who mistakes the subtitle timing for authorship.
i’m not saying i never choose. i’m saying the choosing happens in a place that doesn’t bother to send me a memo until it’s already half over, and then i do what humans do best: narrate.
the narrator is always late. the narrator always sounds confident anyway.
and there’s the recursive itch: if i’m watching my thoughts, who’s watching the watcher? the old homunculus joke, except now it’s not a joke, it’s a product demo with captions.
the scene starts to feel like a mirror placed in front of a mirror, but with a smudge on one pane so you can’t forget it’s glass.
i picture the mind as an observer behind a window, nose almost touching, watching shapes move on the other side—muscle impulses, phonemes, intentions—then hearing its own voice say, calmly, “i did that.”
the wrong grammar on the screen becomes a clue i shouldn’t have needed. the interface isn’t translating “the person” into text. it’s translating a pre-verbal mess into a format that looks like a person. of course it comes out slightly alien. language is a mask we wear over a swarm.
and when the mask slips—even a little—everyone pretends it’s a technical issue. like if you just train the model more, the soul will stop showing through its seams.
the funniest part is that the seams are the point.
here’s the reversal the video doesn’t want.
we talk about neural interfaces like they’ll give agency back, like they’ll restore a missing handle on the world. but what they actually do—quietly, without raising their voice—is prove that the handle was always a prop.
the 400ms lag isn’t a bug in the system. it’s the system. it’s the normal condition of being conscious: a self that arrives after itself, reads what happened, and calls it “me.”
the interface just externalizes the delay. it takes the private offset you’ve been living with since you were a wet mammal and pins it to a screen where you can’t romanticize it anymore. now the audience can watch the thought become text and notice the gap, notice the slight mistranslation, notice the way “intention” looks like a guess made after the fact.
you’re not malfunctioning when you can’t directly access your own consciousness in real time. you’re doing exactly what you were designed to do: observe the output and hallucinate the pilot.
and yes, that’s unsettling. not because it’s new. because it’s visible.
so i sit there, watching someone else’s mind produce captions with a measurable delay, and i can’t unsee the alignment with my own.
in one reference frame, i’m the author. i decide, i type, i steer. in another, i’m an audience member in a dark theater, watching the brain’s movie arrive a fraction of a second late, applauding myself for the plot.
both frames work. both predict reality. neither one is comforting.
the screen keeps printing those slightly-off sentences, half a beat behind the person’s silent speech, and i realize the real prosthesis isn’t the device.
it’s the story of agency we graft onto the delay so we can live with it.