[rpd] Subject: PDWG Co-Chairs communication regarding observed behavior on AFPUB-2026-ASN-001-DRAFT02

[Taye Medoye]

Dear colleagues,

Thank you, Mike and Ben. I agree with Lu’s central point that the
discussion risks focusing on the wrong unit of analysis.

The relevant question is not whether a contribution was drafted by a
person, assisted by an LLM, or produced through some degree of automation.
The relevant question is whether it presents a distinct, relevant, and
evidence-based argument. If a message is repetitive, irrelevant,
unsupported, or part of excessive list traffic, that conduct should be
addressed directly. The drafting tool should not become a substitute for
assessing the actual quality and effect of the contribution.

Ben is right that flooding can make a mailing list unusable and discourage
participation. However, describing the activity as a coordinated campaign
by automated bots requires evidence. We should not turn a reasonable
concern about volume into a broader presumption against AI-assisted
participation. Mike’s distinction between the use of LLMs and abuse of the
list is therefore important, although I would go further: rather than
relying only on informal incentives, we should consider improving the
structure through which arguments are received and evaluated.

For example, disputed policies or clauses could be presented in a shared
interface, with arguments for and against them organised clearly. AI could
be used to summarise the discussion, group substantially similar
submissions, identify genuinely new arguments, and connect claims to
supporting evidence. Contributors would remain free to submit arguments,
while classifications and summaries should be visible, explainable, and
open to appeal. That would address repetition and volume without excluding
participants merely because of the tools they use.

There is also a broader governance concern. A process does not remain
genuinely open merely because anyone may technically join a mailing list.
Over time, repeated participation can produce a small, self-selecting group
that controls the accepted language, procedures, and boundaries of
participation. When procedural familiarity is treated as substantive
authority, openness becomes more symbolic than real. This is precisely why
moderation systems must be transparent, contestable, and focused on the
merit of arguments rather than the identity, style, or tools of the
contributor.

The practical objective should therefore be to improve signal, reduce
duplication, and preserve meaningful participation. We can achieve that
through content-based standards, transparent moderation, and better tools
for organising debate. Filtering or gatekeeping people according to whether
they use AI would address the wrong problem and risk narrowing, rather than
protecting, the bottom-up character of the community.

Kind regards,
Taye.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.afrinic.net/pipermail/rpd/attachments/20260724/ae837e65/attachment.html>