Re: Concern about AI-generated contributions in VCWG discussions

Thank you for sharing this article. It’s the best argument for human
reputation being more important than identity that I have seen so far.

It also explains, quite clearly, why insisting on disclosure of AI use is
not just futile but also deeply damaging to humanity.

Which brings us to the issue of “proof of humanity”, which the article does
not explicitly discuss.

I want your reputation to be contextual and irrevocably tied to your
Sybil-resistant biometric identity. That doesn’t mean I want your
biometrics, because that would allow me to track you across contexts à la
Chinese social credit scoring.

I don’t want to be asked if I used an LLM to author something I said. If
you ask, I will be offended and may leave the conversation - if I can. As
an employee, I don’t expect to be restricted in how I use whatever AI I
choose to be responsible for.

From my perspective as “invited expert on privacy and community interest”
to the VC and DID standards, the lack of human reputation as an essential
use-case is disappointing. My disappointment is not important, because I am
not the intended audience for the standards, so all I can do is to remind
the community of this issue every once in a while.

However, the problem for this community may turn out to be that LLMs
highlight the use-case of human agency, responsibility and reputation to
such an extent that our work is ignored.

Adrian


On Sun, Feb 15, 2026 at 2:04 PM Steven Rowat <steven_rowat@sunshine.net>
wrote:

> On 2026-02-14 12:52 pm, Daniel Hardman wrote:
>
> Good experiment, Moses. I'll be very curious to see the results.
>
> FWIW, I have been trying to hold myself to the following standard:
> https://dhh1128.github.io/papers/ai-coca.html
>
> The part that I'm not sure about is: "I will acknowledge the AI’s
> contribution according to the reasonable expectations of my audience." Are
> "the reasonable expectations of my audience" shifting?
>
> There's an analysis appearing today in the Atlantic today, "Words Without
> Consequence" that bears directly on this, written by "Deb Roy
> <https://www.theatlantic.com/author/deb-roy/>...a professor of Media Arts
> and Sciences at MIT, where he directs the MIT Center for Constructive
> Communication."
>
> https://www.theatlantic.com/technology/2026/02/words-without-consequence/685974/
>
> I found it deeply reasoned, and convincing. The key point is that the
> attribution link to a given person allows moral responsibility, current AI
> practice has cut this, and we risk major destruction if this is not
> rectified.
>
> Of course the range of attribution needs to be a continuum, as your
> example of the song lyric shows. But it still needs to exist, and we need
> new norms for ensuring that it does.
>
> My current suggestion: just as norms for a scientific paper's statement of
> a fact are to reference a given source for that fact, I believe the use of
> AI should at very least entail the specific AI model and application area:
> "Claude Code xx.03, Cloud Version was used to generate this text", or
> "ChatGPT 4.5.xx running on a local machine I own generated the original of
> this paper based on documents I fed it. Then I substantially revised that
> result."... etc.
>
> Does this seem plausible or viable?
>
> Steven Rowat
>