 |
| Is This True? |
I want to be upfront about how this piece
came to exist, because it’s relevant to the last idea in it.
I was not conducting research. I was not
developing a framework. I was scrolling through my social media feed on a slow
afternoon, getting mildly irritated, and I started typing to Claude the way one
mutters to oneself while reading the paper — except the paper muttered back.
What I’d noticed: somewhere between one
in five and one in ten posts in my feed now ends with the same three words. “Is
this true?” It has become its own genre, its own aesthetic. Not journalism.
Mostly not genuine inquiry either. A performance of scepticism — the poster
signalling I’m not just swallowing this without the effort of actually
checking. The doubt is outsourced to whoever replies first. The thing gets
spread either way.
That observation started a conversation
that lasted a few hours and ended with something that looks, uncomfortably,
like a product concept. This is that concept, written up.
The Milk Vat and the Amusement Park
Two stories came up in the conversation
that I think describe the current state of social media better than any
analysis I’ve read.
The first is traditional — possibly
apocryphal, certainly instructive. A king asks citizens to contribute milk to a
communal vat for a grand public occasion. Each person, individually, reasons:
my single cup of water won’t make any difference among thousands of litres of
milk. The vat fills entirely with water. No single person was irrational. The
collective outcome was completely irrational.
That is the social media feed. Every
person posting a questionable claim thinks: my one post won’t change anything.
The vat fills with water. Individually rational, collectively disastrous.
The second story is stranger because it
involves someone actually counting. A water amusement park, according to a post
I once read and cannot now verify (fittingly), fills 10,000 litres each morning
and drains 12,000 by evening. Someone is measuring. The conclusion is
mathematically unavoidable. The park keeps running.
That’s the platform’s relationship with
its own misinformation data. They have the numbers. The pool keeps running.
The Elevator Pause
So what would change this?
Not suppression — the current approach of
shadow-banning and algorithmic demotion that happens invisibly, accountable to
no one, optimised for the platform’s legal exposure rather than the reader’s
epistemic wellbeing.
Not warning labels as currently
implemented — the academic literature on these is mixed at best, and they’ve
been so thoroughly gamed that they’ve lost most of their signal value.
Something behavioural. Something that
doesn’t feel like punishment.
What if posting worked like using an
elevator?
You press the button. The floor indicator
shows you’ve arrived. But you still wait a natural beat for the door to open.
Nobody feels detained by the elevator door. It’s just a pause, a built-in
rhythm. You could force the door — in extremis — but most of the time you don’t
bother.
That pause, applied to posting, is where
an AI could offer its view. The post is not blocked. The poster is not warned.
They simply encounter, for a moment or two, an opinion — here is what I think
about what you’re about to share — before the door opens.
The spectrum matters. An instant green
for straightforward posts. A gentle amber nudge for contested claims (this
appears disputed — still posting?). A slower amber for things the AI is
still working through (I’m still checking a few things — want to wait?).
A full red with detailed reasoning for the genuinely problematic.
And if you’re posting in a tearing hurry?
The opinion attaches retroactively. The post goes out immediately. The view
catches up within seconds, and the poster receives a notification: your post
from two minutes ago has been reviewed — here’s what came up. They now face
a visible choice: delete, respond, or brazen it out. That visible decision is
itself new social information.
HMV: The Naming Breakthrough
The conversation spent some time on what
to call this, and I think the name matters more than it might seem.
“Flag” is wrong. It has the energy of a
traffic stop. A challan. It implies that authority has noted your
transgression. The word itself makes people defensive before they’ve read a
word of the content.
The word we landed on: HMV. Here’s My
View.
The AI is not a gatekeeper. It’s not a
referee or a regulator. It is an entity with an opinion — one that travels with
the post, visible to all readers, given equal standing alongside the post
itself. The user is not being commanded or stopped. They’re choosing whether to
listen to a voice that has looked at something they haven’t.
The shift this creates in emotional
register is significant. A flag feels like a checkpoint. An HMV feels like a
well-read colleague leaning across a table: “you know, I’d double-check this
bit before you send it around.” Same information. Entirely different
relationship. And people don’t get defensive with opinions the way they do with
verdicts — which is the entire UX philosophy of the thing, compressed into
three letters.
There’s also an accountability dimension.
Current content moderation is a black box. When a post is suppressed or
labelled, there is no legible reasoning. An HMV that says I’m flagging this
because claim X lacks any sourcing and directly contradicts peer-reviewed
finding Y is auditable. Users can argue with it. The reasoning can be
improved over time. The AI’s view becomes a starting point for a conversation
rather than the conversation’s end.
Buckets, Mood Maps, and the Absurd
Feed
Posts that receive an HMV self-sort into
browsable categories — buckets — that readers opt into at their discretion. The
reader who wants a clean feed gets one. The reader who wants to browse the
annotated feed gets that. The reader who is, by 9pm, completely done with news
channels and just wants to look at flagged posts and laugh at them gets a
dedicated, curated experience for exactly that purpose.
This last use case — and I want to dwell
on it for a moment — is currently underserved almost completely. People do this
already. They screenshot absurd posts and share them in WhatsApp groups. They
quote-tweet them with a raised eyebrow. They have entire group chats dedicated
to the daily harvest of dubious content. What they don’t have is a
platform-native, well-curated version of this experience. Making it native and
first-class is not a small thing.
The trending bucket dashboard becomes
something else entirely: a real-time social mood map. If “Absurd” is trending
at 11am on a Tuesday, something is happening in the world. If “Unverified but
spreading fast” spikes every evening between 7 and 9pm, that tells you when
people are tired and credulous and vulnerable to sharing things they’d reject
with fresh eyes. Epidemiologists and historians and platform researchers would
find this data extraordinary. It is currently captured nowhere.
There is also a sub-bucket worth naming: Contested
Flags — the bucket for posts where the AI’s HMV and the users’ collective
response diverge significantly. Posts where the AI said this looks
problematic and half the users disagreed, or vice versa. That divergence is
the most interesting data point of all. It is where genuine argument happens,
where context the AI lacked gets surfaced, where the system learns.
And the unanimously-flagged bucket —
posts where the HMV fired and not a single user disagreed — is quietly the most
revealing. When nobody bothers to defend a post, that is a different category
of information than a contested flag. It is the water in the vat, undeniable.
Why Platforms Discarded This
At some point in this conversation came
the observation that the sharpest engineers and product minds at every major
platform have almost certainly thought of versions of this concept. And
discarded them.
The reason is not mysterious. Every
friction point in the posting flow — even two seconds — shows up in the
engagement metrics as a drop. Someone in a quarterly review presents the data: we
introduced the pause and daily post volume fell 3%. The room goes quiet.
The feature is removed. The metric being optimised is volume. Quality has no
column in the dashboard.
The slightly deeper reason is that the
current mess is not visibly costing the platforms anything they can measure.
Trashcans are cheap to run. Sorted libraries cost money and require judgment
calls. You only invest in sorting when unsorted starts losing you something,
and so far — in quarterly revenue terms — it hasn’t.
The irony that the concept is designed to
expose: platforms already do a version of this. Shadow banning, algorithmic
suppression, content warnings, reduced distribution — they make the same
underlying judgment about content that the HMV would make, but they make it
invisibly, unaccountably, and optimised for their legal team rather than their
users. The HMV is the same function, done openly. What’s actually missing is
not the technology. It’s the will to be transparent about it.
Who Should Build It
This is not, I think, something that gets
retrofitted onto existing platforms. The architecture of those platforms — the
engagement metrics, the advertiser relationships, the quarterly targets, the
institutional addiction to volume — is too load-bearing to rebuild around a
different principle.
What the concept requires is an AI-native
entrant who builds legibility in from the start, the way certain founding
principles get encoded into a product before it scales. An entrant without the
engagement-volume addiction. One willing to accept that some users will not
come for more — they’ll come for better.
The open question, and it’s a real one,
is whether a new entrant’s incentive structure would eventually corrupt in the
same direction. Probably yes. Which is the actual satirical target here — not
the absurd posts, not the users who share them, but the institution that saw
the tool, ran the numbers, and chose volume. That institution has a lot of
screen time coming to it.
My 2 Cents: The Idea That Closes the
Loop
The last idea in the conversation arrived
almost as an afterthought and I think it’s the most interesting one.
People now spend hours daily interacting
with AI — at work, in personal projects, in conversations exactly like the one
that produced this piece. Those interactions contain thinking. Judgments.
Half-formed ideas that occasionally, under the right pressure, become something
more.
What if AI interfaces had a “My 2
Cents” button — a feature that reviewed your day’s or week’s AI
conversations and surfaced what seemed, from the outside, worth sharing with
the world? Not a full transcript. Not a dump of logs. A distillation, curated
by the AI, of the moments in your own thinking that might land somewhere useful
— then routed to the communities or feeds where it was most likely to find an
audience.
AI as curator of your own insights. Not
generating content for you, but surfacing what you already thought.
The meta-point, which I am aware of and
which is part of why I’m writing this: that is more or less what happened here.
I was rambling. I was not trying to build anything. An AI held up a mirror to
the conversation and said, approximately: this seems worth keeping. This
piece is that mirror.
The Ideas, Collected
For anyone who prefers a summary:
The Elevator Pause — a 2-second non-blocking delay at point of posting. Behavioural, not
punitive. The door opens; you just wait for it.
HMV: Here’s My View — an AI opinion that travels with the post. Not a flag, not a verdict.
An equal voice, visible to all, with its reasoning shown.
Self-sorting Buckets — flagged content routes itself into categories. Readers opt in.
Trending buckets become a social mood map.
Contested Flags — the sub-bucket where AI and users disagree. The most interesting
data on the platform.
My 2 Cents —
a button on AI interfaces that distils your week’s interactions into what’s
worth sharing. The loop closes.
None of this requires new technology.
It requires a different set of
priorities, which is — as it turns out — the hardest possible thing to build.
This piece grew out of an unplanned
conversation with Claude. The milk vat story is traditional. The amusement park
story came from somewhere on the internet and may or may not be accurate —
which is, fittingly, exactly the point.
LinkedIn Newsletter Article
No comments:
Post a Comment