Allan Dafoe

今天很高興為大家介紹這場對談，與談人是唐鳳、Hélène Landemore 和 Ben Garfinkel。Hélène Landemore 是我以前在耶魯大學的同事，她在那裡擔任政治學副教授。她一部四卷本的民主研究已經寫到一半，書中為民主提出正當性論證、釐清民主的意涵，也提出創新民主的方法。從我認識 Hélène 以來，她就一直在思考民主的意義；她正是那種特別珍貴的政治理論家，試著解決真實世界中迫切的問題，而且在這個過程中，讓自己的思考接受混亂的經驗現實檢驗。

Ben Garfinkel 是 AI 治理中心的研究員，也是牛津大學政治與國際關係學系的博士生。Ben 的學術貢獻橫跨許多主題，包括密碼學和 AI 對民主的意涵。我總是很期待讀 Ben 的分析，因為他很有天分，能從龐大又雜亂的題目裡提煉出關鍵問題。

唐鳳是臺灣的數位政委，是臺灣史上最年輕的政務委員，也是全世界第一位公開跨性別身分的部長級官員。唐鳳的貢獻很難一言以蔽之。在不同的時期，唐鳳當過新創公司創業者、開源軟體黑客、政治運動者，還有詩人。唐鳳是那種罕見的人，在真實世界裡、在當代文化與科技的最前沿，以重大的方式創新治理。唐鳳透過一些大事來做這件事，例如臺灣的協作公民科技、對抗假訊息攻勢，以及在全球堪稱典範的防疫政策。但也透過一些簡單的創新，例如留意到有時從 app 和裝置上拿掉某些功能，反而能更好地引導我們的使用方式，這其中的好處。

唐鳳提醒我們，數位科技最大的正面影響，有些並不來自它們本身做了什麼，而是來自它們讓我們能夠做什麼。引用《Wired》的一段話：「當世界撕裂在兩種反烏托邦之間，一邊是部分民主國家後信任時代的資訊混亂，另一邊是威權的、以科技為中介的監控與審查體制，唐鳳正在提出並親身示範一個激進的論點：數位工具可以用來打造更強韌、更開放、更負責的民主。」對我們這些相信改善治理既關乎理論、也同樣關乎經驗學習的人來說，像唐鳳以及臺灣人民所做的這些真實世界政策實驗，正是治理創新最關鍵的輸入。

今天我們會盡量採取對談形式，部分時間會由 Hélène 和 Ben 提出評論和問題。我也鼓勵所有觀眾在下方的功能欄輸入你的問題。我們不能保證每個問題都會得到回答，但我們會看到，並試著把它們融入對談。那麼，唐鳳，我們很期待向你學習。現在把時間交給你。

非常感謝唐鳳。Hélène，你願意分享一下你的想法嗎？

我想你大概應該先開始討論了。

好的。謝謝唐鳳，太精彩了。我想接下來，Hélène，如果你也想再深入追問這些線索，我們可以這麼做。但現在我們把時間交給 Ben，請你分享你的想法和問題。

就流程而言，我們已經到了活動預定結束的時間。但唐鳳提到——

——還有多一點時間。

（笑）我不確定 Ben 和 Hélène 接下來有沒有空，但我覺得剛才的對話非常有趣，所以也許我們可以再繼續聊一下。我先向唐鳳和 Hélène 提一個比較帶懷疑態度的問題。Ben 把直接民主的一些疑慮講得相當到位，我也要指出，接觸西方媒體和社群媒體之後，很難不對各種讓公眾更深入參與政治討論的提案感到擔憂。很容易想像事情會出差錯。但與其在理論層面辯論——那當然也有用——我最終更被經驗證據說服，這也是為什麼我對臺灣已經取得的成就感到這麼興奮。

那我想問各位的是，如果要在美國、英國或其他國家部署我們討論的這類公民科技，哪些做法算是最容易著手、影響最大、又最可能成功，而且確實做得到、做成後又能成為對「可能做到什麼」的嚴格考驗？Hélène，比如說，妳提出的那個論點——我們不知道美國人有沒有能力實踐參與式民主，因為他們根本沒有機會嘗試。

我的懷疑是，也許需要多年、幾十年的成熟與實踐，需要文化的發展、全國共識、國家認同等等一系列條件，才能培養出一個有能力做好這件事的公民社群。但讓我們把它變成一個經驗性的檢驗。什麼樣攸關重大的公民科技推行方案，能讓我們看到美國人是否有能力實踐這種多元的、直接的參與式民主？

嗯，妳認為公民科技能通過的最嚴格考驗。對吧？因為考驗越嚴格，懷疑者就越會被說服，相信這真的行得通。

對。所以一個可能的結論是，如果能列出一份公民科技可以推行的政策領域或議題清單就太好了，然後按重要性或緊迫性評分，再乘以公民科技能成功或能改善審議過程的機率。然後按照這個排序，一路做下去。

好的。那我們看看還有沒有最後的問題。不然的話，我們大概該讓這些貢獻了這麼多智慧的大腦們休息了。我滑一下得票最高的問題。我想大部分都已經部分回答過了。有一個…… 噢，唐鳳，我看到你回覆了。對。那把投影片分享出來就太好了，當然以開放為原則（笑），這是一定的。

好。也許我就…… 這些問題算是比較具體的，但我還是提一下。一個是 Shivangi Rajora 問的，談到某種原則之間的張力——決策要快，或者說流程要快，同時流程又要具包容性；我們就把這題提出來。然後還有一個相關的、比較具體的問題，之前已經稍微談過了，就是不同流程之間的取捨，以及選擇偏差和代表性的問題。如果你們任何一位還想補充的話，那就太好了。

其實我還有一個問題，是還沒有真正被問到的。我們是 AI 治理中心（GovAI）。我們經常特別關注 AI 技術。所以，你知道的，我不知道在座的各位對此有多少想法，但你們怎麼看機器語言理解可以為公民科技做些什麼？或者，你知道的，你們當中有人提到用 AI 協助引導審議。

那這是不是，你知道的，是被過度炒作了？我們不太可能在短期內擁有真正有意義的工具？還是你們認為在未來幾年，會有在語言理解方面足夠強的 AI 工具問世，可以真正放大這方面的效益？

太好了。那麼，我覺得這是一個很好的結尾。我想再次感謝我們的與談人。感謝唐鳳、Hélène、Ben，帶來這麼精彩的對話。真的是很令人興奮的想法。我確實認為，這正是從理論和實證兩方面探索民主的前沿。我非常期待看到你們這些創新的實證證據持續累積，讓我們可以，你知道的，看見民主的未來有哪些可能。謝謝。

歡迎大家。我是 Allan Dafoe，AI 治理中心（GovAI）的主任，這個系列講座就是由我們主辦的。我們設在牛津大學的 Future of Humanity Institute（FHI）。我們研究 AI 及相關技術進展帶來的機會與挑戰，好為政策提供建議，讓先進 AI 的好處最大化、風險最小化。我們名稱裡的關鍵詞「治理」，一方面描述性地指涉 AI 的開發與部署是怎麼做決定的，另一方面也指一種規範性的期許：希望這些決定出自有效、公平而且具正當性的制度。如果想進一步了解我們的工作，可以上 governance.ai。

I’m delighted today to introduce our conversation featuring Audrey Tang, Hélène Landemore, and Ben Garfinkel. Hélène Landemore is my former colleague from Yale University, where she is an associate professor of political science. She is halfway through a four-volume examination of democracy, in which she provides a justification for democracy, clarifies its meaning, and suggests ways to innovate on it. Hélène has been thinking about the meaning of democracy for as long as I’ve known her, and she is that especially valuable kind of political theorist who is trying to solve pressing real-world problems, and in so doing, exposes her thinking to the mess of empirical reality.

Ben Garfinkel is a research fellow at the Center for the Governance of AI and a DPhil student at Oxford’s Department of Politics and International Relations. Ben’s intellectual contributions span many topic areas, including the implications for democracy of cryptography and of AI. I’m always eager to read Ben’s analyses, as he has a gift for distilling out the key issues from large, messy topics.

Audrey Tang is Taiwan’s digital minister, Taiwan’s youngest minister in the country’s history, and also the world’s first openly transgender minister. Audrey’s contributions are hard to summarize. Audrey, at various points, has been a startup entrepreneur, open source software hacker, political activist, and poet. Audrey is one of those rare people who is innovating in governance in a major way in the real world at the cutting edge of contemporary culture and technology. Audrey is doing this through big things, such as through collaborative civic technologies in Taiwan, combating disinformation campaigns, and globally exemplary policies for addressing COVID. But also through simple innovations, such as being attentive to the virtues of sometimes removing features from apps and devices to better steer our use of them.

Audrey reminds us that some of the greatest positive impacts from digital technologies come not from what they do, but from what they enable us to do. To quote in part from Wired, “While the world is torn between twin dystopias of post-trust information chaos amongst some democracies and the other dystopia of authoritarian, technologically mediated surveillance and censorship regime, Audrey is making and demonstrating the radical argument that digital tools can be used to build stronger, more open, more accountable democracies.” For those of us who believe that improved governance is as much about empirical learning as it is about theory, real-world policy experiments, such as Audrey’s, and those of the people of Taiwan, are the critical input to governance innovation.

Today, we will aim for a conversational format, in part where Hélène and Ben will offer comments and pose questions. I also want to encourage all audience members to type your questions in the feature below. We can’t promise that your questions will be answered, but we will see them and try to integrate them into the conversation. So with that, Audrey, we look forward to learning from you. The floor is yours.

Thank you so much, Audrey. Hélène, would you like to share your thoughts?

I think you should probably start discussing things.

Yes. Thanks, Audrey. That was fantastic. I suspect as we continue, Hélène, if you wanna also pull on some of these threads further, we can do that. But now we’ll turn it over to Ben for your thoughts and questions.

So just in terms of process, we are at the nominal ending of our event. But Audrey has mentioned that —

— has a bit more time.

(laughter) So I don’t know about Ben and Hélène’s availability, but I thought that conversation was very interesting, so maybe we can continue for a bit. And I’ll start by posing maybe a skeptical question to Audrey and Hélène. So, you know, I thought Ben laid out, you know, some of the concerns with direct democracy pretty effectively, and I’ll just flag, you know, it’s hard maybe, exposed to Western media and social media, to not be concerned about the, you know, any or various proposals of enabling the public to be more engaged in political discourse. It’s easy to imagine ways that that can go awry. But rather than debate it on theory, which is also useful, I am ultimately more persuaded by empirics, and so this is why I’m so excited about the achievements that have happened in Taiwan.

And then my question for, I guess, the panel is, what are the sort of lowest hanging fruit, the highest impact, but most likely to succeed deployments of civic technology of the kinds we’ve been discussing in, say, the United States or the UK or other countries that, you know, could be achievable and upon being achieved would represent a hard test of what’s possible? And Hélène, for example, to your kind of argument that we don’t know how capable, say, Americans are of a participatory democracy because they’re not given a chance.

You know, my skepticism is it may be that it takes years, decades of maturation and practice and sort of development of the culture and national consensus and national identity and sort of a range of things to have a civic community that’s able to do this well. But let’s turn this into an empirical test. What would be a sort of a high stakes rollout of civic technology where we could see how capable, say, Americans are of this kind of pluralistic direct participatory democracy?

Well, the hardest test that you think civic technology can pass. Right? ’Cause the harder the test, the more persuaded skeptics will be that this can actually work.

Yeah. So one maybe takeaway is, it’d be great to have a list of policy domains or areas where civic technologies could be rolled out, and you can score them by the importance or urgency, and then times the probability that the civic technologies will succeed or improve the deliberative process. And then, you know, sort by that and work your way down.

Okay. Well, let’s see if there’s any last questions. But otherwise we should probably excuse our many brains that have contributed here. So just scrolling through some of the top voted questions. I think most of these have been partly addressed. One... Oh, Audrey, I see you replied. Yes. So it’d be great to get the slides shared, which of course with openness as a principle (laughter) that would be the case.

Okay. Maybe I’ll just... These are sort of relatively narrow questions, but I’ll pose these. One is from Shivangi Rajora, which asks about the tension between the principle of something — decisions being fast or processes being fast and the processes being inclusive, so we’ll put that one. And then there’s a related sort of narrow question, which has already been slightly addressed, but the trade-off between, I guess, different processes and selection biases and representativeness. So to the extent that there’s a bit more you — either of you would like to say on that, that’d be great.

So I actually have one more question that hasn’t really been asked. So, you know, we are the Center for the Governance of AI. We’re often looking especially towards AI technologies. So I, you know, I don’t know how much anyone on this seminar has thoughts on it, but what are your views about what machine language understanding could do to civic technologies, or, you know, some of you mentioned AI facilitators for deliberation.

So is this, you know, is this sort of over-hyped, that it’s unlikely that we’re going to have meaningfully useful tools anytime soon, or do you see in the coming years the rollout of AI tools that are sufficiently competent with language understanding that they can really amplify the benefits here?

Great. Well, with that, I think it’s a excellent place to conclude. I just wanna thank our panelists again. Thank you Audrey, Hélène, Ben, for this wonderful conversation. Really exciting ideas. I do think this is the frontier of democracy that’s being theoretically and empirically explored, and I’m extremely excited to see the continued empirical evidence of your innovations rolling in, so we can, you know, see what’s possible for the future of democracy. Thank you.

Welcome. I’m Allan Dafoe, the director of the Center for the Governance of AI, which is organizing this talk series. We are based at the Future of Humanity Institute at the University of Oxford. We research the opportunities and challenges brought by advances in AI and related technologies, so as to advise policy to maximize the benefits and minimize the risks from advanced AI. Governance, this key term in our name, refers both descriptively to the ways that the decisions are made about the development and deployment of AI, but also the normative aspiration that those decisions emerge from institutions that are effective, equitable, and legitimate. If you want to learn more about our work, you can go to governance.ai.