Gov/en/Portal:R&D/Innovations:AI Disclaimer: Difference between revisions

← Older edit Newer edit →

VisualWikitext

@@ Line 1: / Line 1: @@
 {{KidsIntro|When a computer helps write something on WikiDeal, the page says so clearly. That way everyone knows a person checked it, just like a teacher checks homework.}}
+{{ExpertIntro|The AI Disclaimer is a transparency rule under exploration: any AI-assisted content would be visibly flagged, sourced where relevant, and treated as provisional until a human contributor reviews it. It extends Wikipedia's sourcing culture to machine-drafted material and supports community governance of content quality.}}
-{{NotApproved}}
-== AI Disclaimer ==
-''Innovation — WikiDeal R&D''
 {| class="wikitable"
 |-
-| Origin || 🔵🟢 Wikimedia Foundation + Théo Bondolfi
+| Origin || 🔵🟢 Wikimedia Foundation + WikiDeal concept
 |-
-| Status || Prototype 1 — In testing
+| Status || Prototype 1, In testing
 |}
-=== What is it? ===
+Readers of any collaborative platform face a growing question: was this text written by a person or by a machine? Without a clear signal, trust erodes for everyone, not because anyone acts in bad faith, but because the ecosystem lacks a shared convention. The AI Disclaimer aims to give users that signal, so a parent reading a babysitting contract template, or a supporter reviewing a funding page, can immediately see what has been human-validated and what has not.
-The AI Disclaimer is a mandatory transparency rule: every page or element produced with the help of artificial intelligence must be openly disclosed. The idea mirrors Wikipedia's citation and sourcing culture — readers always know where a statement comes from. On WikiDeal, any AI-generated element must be declared, sourced where relevant, and marked as requiring human validation before it is treated as reliable.
-=== How it works on WikiDeal ===
+'''On this page:'''
-Each AI-assisted contribution carries a visible marker so readers can tell machine-drafted material from human-authored material. Flagged elements remain provisional until a human contributor reviews and endorses them. The exact display and review workflow will be documented progressively as the platform evolves.
+* [[#the-need|The need: knowing who wrote what]]
+* [[#how-it-works|How it is intended to work]]
+* [[#roles|Who does what]]
+* [[#why-it-matters|Why it matters for the ecosystem]]
+* [[#state-of-the-art|State of the art]]
+__NOTOC__
+<span id="the-need"></span>
+== The need: knowing who wrote what ==
+Imagine a user, Sofia, comparing two contract templates on WikiDeal. One was drafted with AI assistance and not yet reviewed; the other was checked line by line by experienced contributors. Without a disclaimer, both look equally authoritative. The AI Disclaimer intends to make that difference visible at a glance, in the same spirit as [[wikipedia:Wikipedia:Citing_sources|Wikipedia's citation culture]]: readers should always know where a statement comes from.
+<span id="how-it-works"></span>
+== How it is intended to work ==
+Each AI-assisted contribution would carry a visible marker so readers can tell machine-drafted material from human-authored material. Flagged elements would remain provisional until a human contributor reviews and endorses them. The exact display and review workflow is being documented progressively as the platform evolves.
+<span id="roles"></span>
+== Who does what ==
+* '''Authors (human or AI-assisted)''' declare AI involvement when publishing.
+* '''Reviewers''' (experienced community members) check flagged content and endorse it when it meets quality standards.
+* '''Readers''' see the marker and can decide how much weight to give unreviewed material, or join the review effort themselves.
+<span id="why-it-matters"></span>
+== Why it matters for the ecosystem ==
+This rule supports transparency and citizen participation, two of WikiDeal's guiding principles: anyone can see the status of a page, and anyone can help validate it. It also keeps the platform flexible: AI assistance can speed up drafting without lowering the trust bar, because human validation stays the final step.
 ----
+<span id="state-of-the-art"></span>
+== State of the art ==
+Mandatory AI transparency markers align with an active research field on algorithmic accountability and AI documentation standards. The references below are its core texts.
+* Nicholas Diakopoulos (2016), ''Accountability in Algorithmic Decision Making'', Communications of the ACM, 59(2): [https://doi.org/10.1145/2844110 doi:10.1145/2844110].
+* Margaret Mitchell et al. (2019), ''Model Cards for Model Reporting'', Proceedings of FAT* 2019: [https://doi.org/10.1145/3287560.3287596 doi:10.1145/3287560.3287596] · [https://arxiv.org/abs/1810.03993 open access version].
+* Timnit Gebru et al. (2021), ''Datasheets for Datasets'', Communications of the ACM, 64(12): [https://doi.org/10.1145/3458723 doi:10.1145/3458723] · [https://arxiv.org/abs/1803.09010 open access version].
+* Luciano Floridi et al. (2018), ''AI4People: An Ethical Framework for a Good AI Society'', Minds and Machines, 28: [https://doi.org/10.1007/s11023-018-9482-5 doi:10.1007/s11023-018-9482-5].
+* European Union (2024), ''Artificial Intelligence Act'', Regulation (EU) 2024/1689, including transparency obligations for AI-generated content: [https://en.wikipedia.org/wiki/Artificial_Intelligence_Act overview on Wikipedia].
 '''See also:''' [[Gov/en/Portal:R&D/Innovations:Main|All innovations]] · [[Gov/en/Portal:R&D/Main|R&D Portal]]
 [[Category:Migration June 2026]]
-<!-- visible -->
+[[Category:Innovation]]