Solenne Lazare

大家好。很高興看到滿滿一屋子的人，一起來談 2026 年維基媒體與 AI 的現況。今天，我們會透過 Wikimania 2026 的三個價值來看這件事：自由、公平、可靠。

這場座談的來賓陣容非常精彩：Netha Hussain，醫師，也是維基人。Jimmy Wales，維基百科創辦人——應該不需要我再多介紹了。Wikimedia LLC／Wikimedia Enterprise 的 Lane Becker。維基媒體基金會機器學習與資料工程資深總監 Chris Albon。維基媒體瑞士分會創新計畫負責人 Ilario Valdelli。還有遠距連線的唐鳳（Audrey Tang）——數位領域的先行者、臺灣前數位部長，也透過仁工智慧（Civic AI）等倡議，帶頭推動開放源碼治理。謝謝各位的參與。

好。我是 Solenne Lazare，在維基媒體德國分會負責鏈結開放資料產品——Wikidata、Wikibase Suite、Wikibase Cloud——的產品策略。今天同場還有 Shani Evenstein Sigalov，就請她自己來介紹。

好。接下來請每一位 只用一句話 回答：和三年前相比，AI 已經為我們這個運動帶來的最大單一轉變是什麼？

各位……Lane，我喜歡你談機會。那我們就先從機會談起——運動之內、志工工作裡的機會。

Chris——你在維基媒體基金會掌管機器學習與資料。我們天天聽到 AI，它當然正在改變很多事，但其中有些面向是炒作——大家什麼都想塞一點 AI 進去。能不能舉個例子：你在哪裡看到 AI 真正、實在地改變了志工或維基媒體組織的工作方式？又有哪裡，基本上只是炒作？

非常感謝這個視角。

Chris，再一句就好。抱歉。

沒關係，講得非常好。這一題就先到你們兩位，我們進下一個主題。

技術談了，工具也談了，讓我們回到人身上。接下來，由 Shani 帶我們進入和「人」有關的問題。

好。你會對一位 2026 年開始編輯的十九歲新手說什麼？我想這正好接上你剛才說的。一句話——當然，只能一句。Shani 盯時間，盯得比我還緊。

Lane，你先開始好嗎？

批判思考，放在今天真是切中要害。

沒關係。

唐鳳呢？也許從比較外部的視角？

好，謝謝唐鳳，非常、非常精彩。Chris，我想回頭接唐鳳提到的那條 Kami 家規：智慧體讀了公地，就欠公地一次編輯。技術上，維基媒體真的有辦法強制執行這種規則嗎？那會是什麼樣子？

好，謝謝。我還想補一點：當我們談基礎設施、談為了防範 AI 或與 AI 共處而訂規則時，這些規則有時也會對邊緣或代表性不足的社群產生副作用。舉個我們自己看過的例子：我們在維基媒體德國分會，正在做 Wikidata 的治理，提草案、和社群討論，過程中就發現，有些原本提議的規則，可能對邊緣社群造成非預期的衝擊。我覺得這一點，值得放進今天的討論裡。

Netha——你認為代表性不足與被邊緣化的社群，實際上如何被 AI 影響——維基媒體之內、之外都可以，如果你有例子的話？偏見、治理，任何你覺得合適的角度都歡迎。

謝謝。唐鳳——你常說，大家在談奇點的同時，多元其實早已在場；你剛剛也提到了。但我們在世界上看到的是：AI 的發展，高度集中在少數幾個全球基礎模型手上；同時，許多語言與知識社群拿不到同等的投資，一部分原因是它們不代表同樣的商業分量、同樣的市場份額。你是否看得到這樣的未來：有許多在地治理的 AI 系統，各自向自己的社群負責？你認為「多元」會是確保小語種與小型知識社群不被拋下的最佳方式嗎？

好。AI 公司顯然愈來愈依賴我們的內容，而你的視角獨一無二：我們老是在談大科技公司做了什麼、怎麼利用我們，但真正坐在談判桌上談這些協議的人，是你。2026 年，和這些大型玩家之間的健康關係，長什麼樣子？在你看來，像唐鳳提出的那種規則，能不能變成合約條款——回存和出處歸屬落在協議裡，可能會是什麼樣子？還有一件我真正想知道、我猜現場很多人也想知道的事：在這些談判裡，我們到底握有多少權力？

我想，這正好接到我要給 Jimmy 的下一題。在一個變動的世界裡，可以想像我們得改動自己的原則，因為我們也許得採取比較防禦性的姿態。你怎麼看？生成式 AI 有沒有哪一點，真的要求我們改變原則？還是說，這些原則在今天比以往任何時候都更重要？

好。Shani，要開放現場了嗎？

謝謝 Lane。Chris 是實際打造工具的人，你要補充嗎？

Netha，你要替這一輪收尾嗎？因為 Chris 等於把球丟回給我們志工了。你想說幾句嗎？

誰想接這一題？

我想謝謝各位的到來，也謝謝你們一路這麼專注。當然，更要謝謝所有與談人。我認為今天的討論水準很高、處處是細節，我自己也從各位身上學到很多。謝謝。Shani，閉幕前還有什麼想說的嗎？

Hello, everyone. I’m really excited to see such a full room for our panel about the state of Wikimedia and AI in 2026. We’re going to look at this through the prism of the values of Wikimania 2026: freedom, equity, and reliability.

For this amazing panel we have great guests. We have Netha Hussain , a physician and a Wikimedian. Jimmy Wales , the founder of Wikipedia — I don’t think I should introduce him any more. We have Lane Becker of Wikimedia LLC / Wikimedia Enterprise. Chris Albon , Senior Director of Machine Learning and Data Engineering at the Wikimedia Foundation. Ilario Valdelli , Innovation Programme Manager for Wikimedia Switzerland. And we also have, remote, Audrey Tang , a digital pioneer — Taiwan’s former Digital Minister, and someone leading advocacy for open-source governance with initiatives such as Civic AI. Thank you all for being here.

Yes. I’m Solenne Lazare. I’m the Head of Product Strategy for Linked Open Data products — products such as Wikidata, Wikibase Suite, and Wikibase Cloud at Wikimedia Deutschland. And we have Shani Evenstein Sigalov with us, whom I will leave to introduce herself.

Yeah. So all of you: you have one sentence only to answer this question. Compared with just three years ago, what is the single biggest shift that AI has already brought to our movement?

You all… Lane, I like that you talk about opportunity, and I think I want to first start by talking about the opportunity — opportunities within our work for the movement, for the volunteers.

So Chris — you run machine learning and data at the Wikimedia Foundation. We hear a lot about AI, and of course it’s a transforming thing, but there are some aspects of it that are hype. Everyone wants to put AI in everything. Could you give me a piece of example of a place where you see AI is genuinely and in a valuable way transforming the way volunteers or the Wikimedia organizations are working — and maybe one place where it’s basically hype?

Thank you so much for this perspective.

One more sentence, Chris. Sorry.

No worries. It was so very good. We’re going to keep it between the two of you so we can move on to the next topic.

I think we discussed technology, we discussed tools. Let’s bring it back to the people. So now Shani will introduce us to the people’s questions.

Yeah. What would you tell a 19-year-old starting to edit in 2026? And I think that’s the perfect bridge with what you were saying. One sentence — in one sentence, of course. Shani is even harder on the clock than I am.

Lane, do you want to start?

I think critical thinking is an excellent point today.

It’s okay.

And Audrey, from maybe more the external perspective?

Okay. Thank you, Audrey. That was really, really interesting. Chris, I would like to maybe bounce back on the rules that Audrey mentioned the Kami has: that if an agent reads the Commons, it owes the Commons an edit. Technically, could Wikimedia actually enforce such a rule? And what would that look like?

Yes, thank you. I also feel what’s interesting to look at is: when we’re talking about infrastructure and creating rules to protect against the AI or to work with the AI, sometimes these rules can also have side effects for marginalized or underrepresented communities. And maybe as one note: we’ve seen it as Wikimedia Deutschland. We are working on governance for Wikidata, and doing proposals and talking with the community. We also realized that some of the rules we were proposing could have unwanted impact on marginalized communities. So that was something that I found important to address in this debate, in this discussion.

Netha — how do you think underrepresented and marginalized communities are actually affected in practice by AI, both within Wikimedia and beyond, if you have examples? And feel free to discuss biases, governance, whatever feels appropriate.

Thank you. Audrey — you often say that while people talk about the singularity, as you also mentioned before, the plurality is already here. But what we’re seeing in the world is that AI development really concentrates on a few global foundation models. And at the same time, there are many languages and knowledge communities that don’t necessarily benefit from the same investments, also because they don’t represent the same commercial weight, the same market share. Do you see a future in which we have many locally governed AI systems, maybe each accountable to its own community? And do you think plurality would be the best way to ensure that smaller languages and knowledge communities are not left behind?

Yes. Clearly, AI companies increasingly rely on our content, and you have a unique perspective here because we keep talking about what big tech does or how it uses us. But you are the one at the deal table having these negotiations. What does a healthy relationship with these big actors look like in 2026? And in your opinion, can some rules like the one Audrey proposed become a contract term — what could even write-back and attribution look like in this deal? And I think what I really want to know, and I guess other people in the room: how much power do we actually have in these discussions?

And I think that makes the perfect transition to my next question to Jimmy. In a changing world, we could imagine changing our principles because we might have to adopt maybe a more defensive approach. What do you think? Is there anything about generative AI that would require us to change our principles? Or are all these principles today more important than ever?

Yeah. Shani, shall we take it to the room?

Thank you, Lane. Chris is the person who is building the tools. Do you want to complement that?

Netha, do you want to close this round on this? Because Chris, in essence, put it on us, the volunteers. So do you want to say something?