Human Attention / Synthetic Subconscious
On Being Absorbed
Every account of the machine layer is written from the inside — your stack, your attention, your smoothed days. Flip the camera. To everyone trying to reach you, that layer is a border.
The Camera Flips
Every essay I have written about the synthetic subconscious holds the camera in the same position: behind your eyes, looking out. The layer absorbs the predictable so the world reaches you pre-smoothed. It meters surprise, guards attention, pays your small tolls, and hands you back the hours. That account is true, and it is exactly half the picture.
Flip the camera. To everyone who is not you, you are not the protagonist of this arrangement. You are part of the Actual — the unvetted world arriving at someone else’s door. Your call, your email, your favor to ask, your good news: each is a small candidate surprise in another person’s day. And increasingly, standing between your bid and their awareness, is a layer that was installed for one purpose — to keep the Actual from charging its owner too much.
The Reality Equation does not care which side of it you stand on. For a decade I have written it from the perspective of the person paying attention. But when someone else’s machine meters their world, you are in the numerator. Their layer’s job is to make you cost less — to receive you, price you, and resolve as much of you as possible before its owner ever feels the charge.
The synthetic subconscious does not only mediate the world for its owner. It mediates its owner for the world.
Talking to the Layer
Consider what already happens to an ordinary bid for attention. The phone rings and is answered by an assistant that interviews you before a human decides whether to pick up — Google shipped that capability in 2018, and it is now unremarkable. The three paragraphs of careful context you wrote arrive as a one-sentence summary above a collapsed thread. The warm reply that comes back in their voice was drafted by a machine and approved with a tap; the team that built Gmail’s Smart Reply reported that its suggestions accounted for roughly one in ten replies sent on mobile — and that was 2016, when the suggestions were three words long.
None of this is malicious. Each step is a reasonable answer to a real problem: most bids on a modern person’s attention are extraction attempts, and a layer that screens them is doing honest work. But notice what has quietly changed about the transaction. You addressed a person. You were processed by a model of the person — a model trained on their history, including their history with you. It knows how they usually answer you. It answers you that way.
Being handled is not the same as being heard — and the transcript no longer tells you which one happened.
That is the unsettling part: not failure, but success. The interaction is polite, fast, and resolved, and precisely because it is smooth you cannot tell whether a person was present for any of it. Presence used to leave fingerprints — delay, typos, the specific roughness of someone answering for themselves. The layer wipes the surface clean. Reaching a person stops being evidence that a person was reached.
Figure 01
The path of a bid
- Screen
- Rank
- Summarize
- Defer
- Draft
- Decline
- File
- Escalate
The Arms Race of Reaching
Gatekeeping is not new. Herbert Simon named the economics in 1971 — a wealth of information creates a poverty of attention, and attention must therefore be allocated — and the powerful have always hired the allocators: the switchboard operator, the private secretary, the aide who decides which letters reach the desk. For a century, being buffered from the world was a privilege of the important. The first genuinely new thing about the synthetic subconscious is that everyone gets the staff.
The second new thing matters more. A human gatekeeper applied judgment; the layer applies precedent. A good secretary could be persuaded, could take a risk, could recognize the visitor the boss did not yet know they needed to meet. A model admits what resembles what its owner already valued. You are no longer admitted by being important, or urgent, or right. You are admitted by being precedented — by looking like the kind of thing that got through before.
When attention is guarded by a model of the person, you no longer reach people by surprising them. You have to surprise their precedent.
Run that forward and you get an arms race. Senders learn what layers pass, and optimize — subject lines engineered for a model’s taste, favors dressed as the categories that historically escalated. Layers learn the optimization and raise the bar. It is search-engine optimization aimed at persons, and it converges the same way: the easily counterfeited signals inflate until they are worthless. What survives are the costly ones — the signals a model cannot cheaply fake on your behalf. Showing up. Handwriting. A mutual friend willing to spend their own standing. Time, visibly spent. Erving Goffman taught that we have always performed for each other’s expectations; the difference is that the audience is now literally a model of those expectations, and it keeps a ledger.
The Unabsorbed Channel
This is not an argument against the layer, any more than “Calibration Debt” was an argument against smoothness. A world where every stranger can charge you attention at will is not a human world either; the zero-attention economy exists because most bids really are extraction. The question is narrower and more personal. A relationship, at bottom, is a standing right to charge each other attention — the people you love are precisely the ones entitled to surprise you. If their bids are screened, normalized, and summarized like everyone else’s, the relationship is being repriced by a machine, and neither of you agreed to the new rate.
So the discipline, as usual, is not refusal. It is deciding — deliberately, in advance — who never has to talk to your layer.
Publish your true channel
Tell the people who matter which door still opens onto you directly — the number, the address, the knock that no model stands behind. A true channel kept secret is indistinguishable from no channel at all.
Keep an always-through list
A short roster the layer must pass untouched: no screening, no summary, full fidelity, at any hour. Let some people always charge you retail. The length of that list is an honest measure of your life.
Answer something yourself, daily
Presence only leaves fingerprints if there are fingerprints. Reply, occasionally, in a way the model of you would never have chosen — longer, stranger, more generous than your precedent. A visibly human answer is becoming a gift, precisely because it might have been delegated.
Ask for the person, plainly
The mirror duty. When it matters, say so at the border: this one is for you, not your layer. We will need social words for this — the machine-age equivalent of marking an envelope “personal” — and the people who use them sparingly will be the ones whose envelopes get opened.
The first decade of this technology will be spent negotiating what our layers do for us. The second will be spent negotiating what they do to each other — billions of models of persons, screening bids from billions of models of persons, with the humans meeting only at escalation. The point of having a synthetic subconscious was never mutual anesthesia. It was to be spared the world’s noise so that when a person finally reaches you, you are actually there to be reached.
Sources
Simon, H. A. (1971). Designing organizations for an information-rich world. In M. Greenberger (Ed.), Computers, Communications, and the Public Interest (pp. 37–72). Johns Hopkins Press.
Kannan, A., Kurach, K., Ravi, S., et al. (2016). Smart Reply: Automated response suggestion for email. Proceedings of KDD ’16. https://doi.org/10.1145/2939672.2939801
Goffman, E. (1959). The Presentation of Self in Everyday Life. Anchor Books.
Google (2018). Answer and screen calls with the Google Assistant — Call Screen launch, Pixel 3 event.