Latest / HumanPrint: How to Use AI Without Losing Yourself / Episode 15: Set the Terms of Your Social Media Presence

Transcript

Episode 15: Set the Terms of Your Social Media Presence · HumanPrint: How to Use AI Without Losing Yourself

0:00
Foreign.
0:05
Welcome to Human Print. I am Christine Whitmarsh,
0:08
and if you are open to collaborating with AI, but you're not
0:12
interested in disappearing into AI, you're in the right place.
0:16
Each week we will look at one place where your authentic human signal is at
0:20
risk for drift and one way to shore it up.
0:23
Oh, and Human Print is brought to you by Abby AI
0:27
Brand bank intelligence that helps authors, thought leaders,
0:31
and enterprise teams train AI on their voice
0:35
and point of view so written content stays unmistakably
0:38
human, consistent and on brand. Learn more at
0:42
abby.ink abbi.ink
0:47
like you write with let's begin.
0:51
Hello everyone, and welcome back to Human Print.
0:56
So in episode 13, if you're tracking episode numbers, very recently
1:00
we talked about knowing your writing process well enough that you
1:04
don't feel obligated to disclose or defend it to strangers. The whole substack,
1:08
LinkedIn, you know, AI slot button thing. And then in episode
1:12
14, we talked about refusing to investigate other writers on behalf
1:16
of technology companies for free because reading someone's finished work
1:20
doesn't give you access to the thinking decisions, tools, conversations, all the other stuff
1:24
the that produced it. So where do we go from here? Because you
1:29
may agree with everything I've said about substack's scanner and LinkedIn's seems
1:33
like AI slop button. I'm calling them game show buzzers at this point,
1:36
and the larger attempt to turn writers and readers into unpaid AI
1:41
investigators while also recognizing that your readers are still
1:44
on those platforms. Your subscribers may be there, your professional network
1:49
may be there, your archives of content, your income,
1:53
conversations, years of work. They might be on one or both of these platforms.
1:57
So yeah, leaving is an option. Staying is also an
2:00
option. But staying doesn't require you to accept every new
2:04
feature, participate in every behavior a platform introduces to
2:08
keep you on there, or treat a technology company's current business strategy
2:12
as a moral instruction for your creative life.
2:16
You can stay and set your own terms.
2:20
This begins with understanding what a social media platform is.
2:24
Substack has spent years presenting itself as a home for writers, and I understand
2:28
why. That language, that branding, that kind of positioning appealed
2:32
to people. Writers have been squeezed by publishers,
2:35
shrinking media budgets, unstable freelance work, social algorithms,
2:39
and this expectation that we should produce an endless supply of free material
2:44
while somehow finding time to write the work we actually care about.
2:48
Then Substack came along and it offered direct publishing,
2:52
email distribution, paid subscriptions, and really a cleaner
2:56
relationship between writers and their readers. For many writers, this has been
3:00
useful, and it still is. But Substack is also a technology
3:03
company building a broader social platform.
3:08
It has added notes and recommendations and direct messages and
3:12
all kinds of bells and whistles, videos, live stream, social discovery,
3:15
other features that extend well beyond sending newsletters.
3:19
And now it has partnered with Pangram to let readers scan
3:24
eligible writing for an estimate. There's a key word again,
3:28
estimate of human versus AI involvement. At some
3:32
point we have to stop responding to each new feature as if it's a surprise.
3:36
This is a company making product decisions. That doesn't
3:40
make the company evil. It does mean, though, that its interests and your interests will
3:43
sometimes overlap and sometimes go the opposite way.
3:47
Your job is to know which situation you're in and act accordingly.
3:51
Now, I've been reading a book, Karen Howe's book. It's called Empire of
3:56
AI, which documents the development of OpenAI and the larger competition
4:00
between AI companies to build increasingly powerful systems
4:04
with AGI, artificial general intelligence serving as the
4:08
central goal around which enormous amounts of money and compute
4:12
and energy, data, labor, all sorts of things are organizing around that AGI
4:17
goal with many companies. So the author's argument in the
4:21
book is broader than one company or one product. She describes an
4:25
industry driven by the belief that reaching AGI first is important
4:28
enough to justify a pace and scale that can push
4:33
the costs onto workers, community artists, researchers, and people whose
4:37
data or labor helps make those systems possible.
4:40
Again, you don't have to agree with every part of her analysis to recognize the
4:44
competitive pressure that she's talking about. These companies are not
4:48
developing AI as a side project. They are competing over
4:51
models and users and data and infrastructure, the infamous
4:55
the data centers and talent distribution. And really, they're competing
5:00
for control over how AI enters our lives.
5:04
Social media and publishing platforms are part of that environment because
5:08
they possess something that AI companies need and and technology companies
5:11
value, which is large quantities of human communication
5:15
along with information about who created it, how other people responded,
5:20
and what happened next. It's all behavioral data.
5:23
So that brings us back to Substack. One theory I raised in the
5:28
most recent episode is that readers scan and creators disclosures
5:33
and error reports, all that kind of stuff. AI swap flags.
5:37
They can produce useful classification and behavioral data like tagging.
5:41
Substack says that neither Substack nor Pangram uses publisher content to
5:46
train gen AI models. That's their policy. I have no evidence that they're doing something
5:50
crazy. But training generative AI models, that's one specific
5:54
use of data. A platform can still learn from,
5:58
which posts people scan, which results they challenge,
6:02
what readers perceive as AI, there's the button which
6:06
writing gets reported and whether a label changes the level
6:10
of engagement on a post. LinkedIn has described its
6:14
new user reports as feedback that can help its
6:17
system identify and reduce the visibility of low quality or inauthentic material.
6:22
So the seems like AI button, which continues to drive me crazy because that sounds
6:25
so vague. And again, that doesn't prove some sort of hidden scheme,
6:29
but it does tell us what your clicks and reports and disclosures and reactions
6:33
and, you know, how they might have product value. And this is where
6:37
my next theory comes in. I just have so many lately and again, clearly labeling
6:41
it as a theory. What if platforms are becoming more concerned
6:46
with identifying distinctively human content because
6:50
that content is becoming more valuable as data?
6:54
The Internet is filling as we speak with machine generated material
6:58
that has been trained on earlier material, like human material, then published back into
7:02
the same environment at a volume that like no group of humans could physically
7:07
produce. So as more synthetic text, you know, the AI data circulates.
7:12
Platforms, model developers, research, reader researchers, readers,
7:15
advertisers, they all have various reasons to distinguish original human
7:20
expression from recycled or generated language. Original work
7:24
contains new stories, personal stories, new combinations of those current
7:28
experiences, specialized knowledge, domain knowledge,
7:32
individual choices, all that unpredictable material produced by actual people,
7:36
actual humans living actual human lives. And yeah, those things have
7:41
value, perhaps even monetary value. So when a platform asks writers
7:46
to label their process, when it allows readers to scan their work
7:50
or it collects reports about content that seems like AI
7:54
slop, I think it's reasonable to ask who benefits from the classification
7:59
and what may eventually be built from it. Again, I can't prove
8:03
that Substack introduced Pangram because it wants cleaner human content for future data
8:07
sets. Substack's stated reason is that readers value transparency
8:11
and they want to know whether the work they support was written by a person.
8:13
That's what they say. But stated reasons do
8:17
not eliminate other incentives. And technology companies can pursue trust,
8:22
market positioning, product improvement, audience growth, and future data value at
8:26
the same time that they're saying that a platform has introduced a
8:30
system that estimates human and AI involvement in writing. Two platforms.
8:34
It allows readers to initiate that judgment. It allows
8:39
writers to publish statements about how they create their work.
8:42
And if a writer disables detection, the reader receives a message that
8:46
detection is unavailable and the platform then receives reports.
8:49
And when writers believe the detector was wrong,
8:52
you don't need access to a private corporate strategy meeting to
8:56
decide how you're going to respond. You need your own terms. So the first boundary
9:00
is around your creative process.
9:03
Decide how you use AI, what you use it for,
9:08
which parts of your work require direct human control.
9:12
Write those decisions down so you can revisit them. Check in on them. Accountability.
9:16
Especially because your boundaries may change as the tools change as they will,
9:20
and as you learn what helps and what harms your writing, that may change
9:24
as well. A process statement created for your own governance.
9:28
It's useful. A process statement produced because a platform has made you feel accused like
9:33
a witch. That's something else. You get to
9:37
decide whether public disclosure serves your readers, your subject, or your own purposes.
9:41
You do not owe a detailed production report every time you publish a
9:45
paragraph on a social media platform. So that's the
9:48
first boundary. The second boundary is around the scanning itself. Now you can decline to
9:52
scan your own work for reassurance, you can decline to scan other
9:56
writers. Definitely encourage that. And you can refuse to
10:00
treat a probabilistic estimate, a statistical estimate, as a ruling
10:04
on authorship, honesty, intelligence or skill.
10:08
Pangram's current technical report describes strong overall
10:11
test performance, including a low false positive, improved detection of mixed
10:15
human AI text. And yeah, those are reported model performance results, results across test settings.
10:20
Again, they they don't reveal the complete creative process behind one person's
10:24
piece and they don't determine whether that person exercised meaningful authorship
10:28
when they wrote that piece. This distinction might seem technical, but the
10:33
consequence is a human one. A model estimate is revealed in
10:37
text. People turn the estimate into a whole story, a whole narrative about that
10:41
writer. Got to keep those things separate.
10:44
That's the most important part. You got to keep those things separate. Estimates and judgments
10:48
and you know, these technical things that are bleeding into opinions of other
10:52
writers. This is, we're eating each other as writers and this is something that bothers
10:56
me. So, meaning going, you know,
11:00
yeah, I'm on track here. The third boundary is around
11:04
your work itself. Keep a complete archive somewhere
11:08
that you control. Your drafts, your published posts, audio files,
11:12
transcripts, research images, comments you want to preserve and
11:16
any other material with long term value should not exist solely inside a platform
11:20
account. We've known this for years. We've been hearing for years that like the whole
11:23
idea of like social media, like online real estate and social media real estate you
11:27
don't own. So you need to own your own, whether your own website, your own
11:30
blog, your own email newsletter list. This is something I've always been a fan of
11:33
the idea of like, you know, if all the Internet things the social media disappeared
11:37
tomorrow, would your followers still be able to find you? So that's what that
11:41
point is about. Because, yeah, platforms change, features disappear, companies are
11:45
acquired all sorts of things. Their accounts are suspended by mistake,
11:48
never know what the algorithms are going to do. Export options can change.
11:53
So you know a body of work that took you years to create it shouldn't
11:56
depend on your continued access to one company's website.
12:00
One social media platform. Your archive is a part of your
12:04
writing practice. Treat it that way. And the fourth
12:08
boundary we're talking about here, that's access to your audience.
12:11
Substack's email model gives writers a more direct connection to
12:15
subscribers than a standard social feed does, and that's part of it. That's been part
12:19
of its appeal. But direct access doesn't mean independent
12:23
when the account, payment system, discovery tools, and publication infrastructure
12:27
still sit inside someone else's company,
12:30
keywords there export the subscriber information
12:33
the platform permits you to export, store it securely, understand what's included,
12:38
and know what would be involved if you ever chose to leave.
12:41
Don't wait until you're angry or locked out or facing some policy
12:45
change by the tech companies to learn how your own audience relationship
12:49
works. Those relationships are yours, not the tech companies. Make sure you protect them
12:53
and create at least one reliable way again for readers to find you outside
12:57
the platform. My I've had a it's called the Writer Wednesday email newsletter list.
13:02
I've had it forever. I think people know that's like the main place you can
13:04
find me if everything else falls apart. So that could
13:09
be your own website again, your mailing list, podcast feed, professional contact address,
13:12
you know, texting network, you know, WhatsApp, all those kind of things.
13:15
The exact method will depend on your work. The principle is simple.
13:19
No single company should be the only door between you and the people who
13:23
choose to hear from you. So the fifth boundary is around
13:27
unpaid platform labor. Here we go. Here's where they've enlisted
13:31
us as the AI police. Do not train yourself to respond to
13:35
every feature the way the platform wants you to respond. So this means you
13:39
don't have to scan, you don't have to report, you don't have to label other
13:43
writers. You don't have to debate every accusation. You don't
13:47
have to produce extra content explaining why your content deserves to be considered
13:51
human. And you don't have to reshape your natural writing voice because you're
13:55
worried that a detector or suspicious reader might misclassify it.
13:59
Recent research into online accusations of AI use found
14:04
that the prose features distinguishing AI from human text did not did
14:09
not reliably predict which human comments people accused of being AI
14:13
generated. The accusations increasingly served as social gatekeeping around
14:18
perceived authenticity rather than accurate identification
14:22
of AI use. In other words, writing badly on
14:25
purpose will not save you. Removing every EM dash will not
14:30
save you. Making your sentences clumsy or adding random
14:34
mistakes, stripping the polish and cleanness from your work,
14:37
your deliberately polished work. That's not going to save you either,
14:41
because the suspicion is not operating as a careful writing assessment.
14:45
Write in your voice. Let other people manage
14:49
their own suspicions. The final boundary is the one that determines whether staying
14:53
still makes sense. And decide what would cause you to leave
14:57
one of these platforms. That's your decision. Your platform relationship should have conditions.
15:01
Maybe you leave if the company prevents you from exporting your data.
15:05
Or maybe you leave if claims rights over your work that you're unwilling
15:09
to grant. Or you leave if the platform begins punishing your content
15:13
based on AI detection results, or if it interferes with your income,
15:17
or if it restricts your access to your reader relationships,
15:20
or if it creates a process you consider fundamentally incompatible
15:24
with your personal writer values. Again, your line might be different, but know
15:29
where it is before the company reaches it for you.
15:31
Otherwise, each new policy becomes a separate argument, each new
15:36
feature takes a little more ground. And you keep adjusting because leaving now feels
15:40
harder than it did six months ago. So yeah, you're allowed to stay.
15:43
You're allowed to benefit from the distribution and the community and the payment
15:47
tools, all the things that a platform provides for you. And you're
15:51
allowed to remain clear minded about the exchange.
15:54
You provide writing attention, data,
15:56
activity, subscription revenue, professional information, and reasons for
16:00
other people to visit. In exchange. The platform provides infrastructure,
16:05
reach and tools. That's a business relationship. That's an exchange.
16:09
Use what serves your work. Protect what belongs to
16:13
you. Refuse the parts that ask you to surrender judgment,
16:16
investigate and judge your peers, or modify your creative
16:21
process for the comfort of a machine.
16:24
So here's this week's human print homework.
16:27
Write down your platform terms. You know I was going to say that, right?
16:30
Choose the primary platform where you want to publish your work and build professional
16:34
relationships. Then answer questions like this. What does this platform provide
16:39
that is genuinely useful to me? What work,
16:43
information, attention, revenue? What do I provide in return for that?
16:46
Which parts of my creative process will always remain private?
16:51
What do I need to copy, export? You know, the data, you know,
16:55
what do I need to protect somewhere where I can control it? What specific change
16:59
would cause me to reduce my participation? Or leave and then take one
17:03
action this week? Download your archive. Just got social media platform. Pretty easy
17:07
to do on substack, I believe. Export the subscriber or contact information that
17:12
you've allowed, you've been allowed to retain. Update your website.
17:16
Create a second way for readers to reach you. Write your private AI
17:20
used boundaries or stop using the scan and report features.
17:23
Stay if staying serves you, but set the terms before
17:27
the platform sets all of them for you. So any of those actions, I just
17:30
listed, like five actions. Just do one of those things this week just because.
17:35
So set the terms. Your terms. It's like your terms
17:39
of use as a writer, what you will permit and what you will not permit
17:43
before the platform does it for you.
17:47
All right, that's it for today's episode of Human Print.
17:51
Until next time, use the platform if it helps. Keep your process,
17:55
your work and your decision in your own hands.
18:06
And that's it for this episode of Human Print. If you want a
18:10
practical framework for protecting your voice amid the AI noise, please follow
18:14
the show. And before you go, take a minute and ask yourself, as the AI
18:18
noise grows louder and louder, which areas of yourself do
18:22
you need to protect the most? I'll see you next time,