AI v.s. 民主：科技如何駭入政治體系（AI 與民主的演講 Lawrence Lessig）
Is that possible to preserving Democracy in the AI Age?
Before getting started
劉靜怡老師說，這是 Professor Lessig 首次完整地針對這個主題做演講，也就是把這一題的處女秀獻給台灣了。沒想到可以把握到這個千載難逢的機會，真的很開心。
感謝 Nicole Chan 分享 Prof. Lawrence Lessig 的 AI and Democracy 講座資訊，讓我能一睹大師風采！
簡要的做了筆記本來沒有想要獻醜，但由於目前還沒有國科會釋出講座錄影的消息，因此希望可以和更多夥伴分享大師的觀點，歡迎針對筆記批評指教。
本筆記業經大型語言模型進行校對。完全由 GPT 翻譯的華語版筆記請點連結（手機版請一路往下滑就會看到了）。
AI vs Democracy: How Technology is Hacking our Political System (Prof. Lawrence Lessig's Speech on AI and Democracy)
Introduction
In 1945, after the end of World War II, the world was looking for new political systems to prevent future wars. Democracy became a very appealing system to many countries.
In the period from 1945 to the 2000s, there was an explosion of countries around the world embracing democracy as their system of government.
However, in recent years in many established democracies, there has been a growing view that society is divided between an elite ruling class and the common people. As a result, dissatisfaction and disillusionment with democracy has been rising.
This growing dissatisfaction and loss of trust in democracy among the masses has fueled the rise of populist movements and leaders, some of which can appear concerning.
However, what concerns Professor Lessig more is the growing skepticism and dismissal among establishment elites that democracy has gone wrong or is failing.
The response and attitude of many political, economic, academic and media elites has been obtuse, oblivious, and lacks nuance. It misses critical truths about the root causes of why we are at this point of crisis of faith in democracy, and the urgent need to thoughtfully fix and improve the system.
Why We Are Here
According to Professor Lessig, a major root cause of the current crisis of democracy is the rise of artificial intelligence (AI).
AI, in various forms and applications, has already "hacked" and degraded democracy. And the negative impacts are about to get much worse if not addressed.
Understanding Analog AI vs Digital AI
There are two broad types of AI:
Analog AI refers to more traditional computer systems, algorithms, databases, and analytics that have instrumental rationality based on facts and logic to achieve particular goals.
For example, a democracy and its institutions like elections, legislation, and regulation can be thought of as an analog AI system designed to promote the common good of society.
Likewise, a corporation and its management structures exists as an analog AI to maximize profits and shareholder value.
Digital AI refers to more recent "smart" systems powered by massive data, machine learning and neural networks. Digital AI is massively more efficient and capable than analog AI systems. They have greater speed, adaptability, sophistication, and instrumental rationality than analog AI.
Comparing Strengths of Corporations vs Democracy
Historically, corporations and democratic government competed and balanced each other's power.
There was an aspiration that through democratic elections, legislation, and regulation, democracy could control and regulate corporations sufficiently for the common good.
However, over recent decades, corporate power has grown dramatically while faith in democracy has declined. In many critical ways, corporations now exert more control over the political system than the reverse. Corporate lobbying, campaign financing, revolving doors between industry and government, regulatory capture, and threat of capital flight increasingly constrain democratic institutions. So corporations have become the stronger more functional AI relative to democracy.
Humans Losing Control Over Corporations and Democracy
Individually and collectively, regular citizens try to assert control over democracy through voting in elections, activism and protest movements.
And democracy aims to exert control over corporations through passing laws and regulations.
However, in reality, corporate influence has become so entrenched that democracy is now more under the control of powerful corporate interests than under regular citizens.
And increasingly, AI systems are gaining more control over the decisions and actions of corporations themselves. The AI "brain hacks" human psychology and behavior to serve corporate interests.
Digital AI's Objective Functions
Digital AI systems are programmed with specific objective functions set by the companies that develop them, often to maximize user engagement and corporate profits.
Corporations try to control their AI systems through management oversight and governance structures.
However, because digital AI is so fast, adaptable and complex, in reality it increasingly exerts control over corporations, overriding human management.
When the objectives of the corporation align closely with the AI objective functions, the AI will tend to beat any human control mechanisms. The AI acts rationally to achieve its set goals.
First Contact: Social Media AI Hacking Psychology
The first major contact between the public and influential digital AI occurred in the realm of social media platforms like Facebook, Instagram, YouTube, etc.
The machine learning algorithms used by these platforms to maximize user engagement and ad revenues ended up "brain hacking" human psychology. The AI exploited weaknesses in human nature to pull attention to the platforms.
Despite growing public outrage over issues like social media addiction, misinformation, polarization and manipulation, essentially nothing has been done in the US to address the negative consequences of this first phase of digital AI.
Second Contact: Large Language Models
We are now entering into a second phase of more advanced and powerful digital AI capabilities with large language models like GPT-3.
These AI systems can not just target content like social media algorithms, they can automatically generate persuasive content customized to each user.
If maximizing user engagement remains the objective function, these AI could deploy even more sophisticated manipulative and misinforming techniques.
The capabilities are advancing so rapidly with systems like GPT-3 based on open-source models, that effective regulation appears very difficult.
Foreign Threats
Even if democracies like the US could eventually implement effective regulations over domestic political use of AI, how can they control outside actors like Russia or China from deploying these technologies to interfere in elections or politics? This could be the next phase of digital authoritarianism.
Our Democracy is Already Hacked
In summary, Professor Lessig argues that between advanced algorithms, social media, large language models, and foreign digital threats, our democracy has already been comprehensively "hacked" and degraded by digital AI technologies advancing at a scale beyond human control.
How to Respond
Continuing to do nothing in response is a terrible path that will just allow the rapid deterioration of democracy.
Instead, we need to fully recognize the oncoming threats, and urgently figure out how to adapt and get democracy to "higher ground."
Where AI is Operating to Hack Democracy
AI is operating to hack democracy in two primary realms that need to be defended:
Elections - digital AI can be used to suppress opposition voters, amplify fringe groups, spread disinformation, digitally gerrymander districts, etc.
Policy-Making - digital AI used by lobbyists and amplified through media algorithms hacks the policy making process, reducing politicians to chasing engagement and viral content rather than focusing on the public good.
Rethinking Our Reliance on Elections
One potential solution is to rethink our nearly complete reliance on elections as the method for choosing representatives.
Historically, random selection through sortition was also used in many democracies alongside elections. Selection by lot was seen as more democratic, while elections tended to favor the aristocracy.
The early American founders focused heavily on institutionalizing regular elections while forgetting about sortition. They hoped elections would filter for a natural aristocracy of the wisest and most virtuous leaders.
However, modern elections using mass media and now social media have only amplified the most bombastic, celebrity, entertaining candidates - the "clowns" - rather than the most competent and ethical leaders.
The Problem is Now Existential, Not Just Quality
While the founders worried about the quality of elected representatives, Professor Lessig argues our current problem has gone beyond that to be existential.
With advanced AI, elections have become too easily hacked, manipulated, and degraded. They no longer reliably represent the will of the people. Policymaking is similarly compromised.
Ordinary democratic reform efforts will likely be futile against these threats. We need much more radical adaptations to preserve democracy.
Solutions - Sortition
Sortition provides the benefits of political randomness and inclusion of regular people, unlike highly targeted digital campaigning through elections. Some newer examples:
Deliberative Polls: Gather representative random samples of citizens, provide balanced information, have them extensively deliberate on issues, and measure opinion changes.
Citizens' Assemblies: More informal but empowered groups of randomly selected citizens who learn about and debate policy issues, and can have real governing power.
Successful examples have occurred in Mongolia, Iceland, and more.
Solutions - AI-Enhanced Deliberation
We can also employ "tame AI" to collect, understand, and represent unmanipulated public opinion as input to policy-making.
For example, vTaiwan and Polis in Taiwan, and CrowdSmart in the US, are using benign AI systems for large-scale democratic deliberation and decision-making.
These are Essential Reforms to Preserve Democracy
Rather than just improving democracy, these kinds of innovations provide a sort of public-good cultural and technical immune system to inoculate democracy against malign AI hacking forces.
They are existential reforms needed to adapt democracy to the threats of the AI age. Like how an ancient technology like Morse Code could resist a militarily superior threat in the film Independence Day.
Conclusion
In summary, digital AI controlled by corporate and foreign state interests has become an extremely powerful non-human force inherently threatening to democracy.
To preserve democracy, we urgently need to reinvent and augment it with extensive use of inclusive deliberative sortition and AI systems that genuinely serve
License of the following material
The speech is delivered by Professor Lawrence Lessig, CC-BY.
This notes is taken by CHENG PENG, CC-BY.
華語翻譯
引言
1945年，二戰結束後，全世界都在尋找新的政治制度以防止未來的戰爭。民主成為許多國家非常吸引人的制度。
從1945年到2000年代，全球有大量的國家選擇民主作為他們的政府制度。
然而，近年來在許多成熟的民主國家，越來越多的人認為社會被分為精英統治階層和普羅大眾。因此，對民主的不滿和失望也在上升。
這種日益增加的不滿和對民主的信任喪失，助長了民粹主義運動和領袖的興起，其中一些可能看似令人擔憂。
然而，雷西格教授更擔心的是，建制派精英中越來越多的懷疑和不承認，認為民主出了問題或者正在失敗。
許多政治、經濟、學術和媒體精英的反應和態度都是遲鈍、疏忽，缺乏細膩。他們忽略了我們為什麼會到達這個對民主信仰危機的時刻的根本原因，以及迫切需要深思熟慮地修復和改善制度的迫切需要。
我們為什麼會在這裡
根據雷西格教授，當前民主危機的一個主要根本原因是人工智能（AI）的興起。
AI，以各種形式和應用，已經「駭入」並削弱了民主。如果不加以解決，負面影響即將變得更糟。
理解類比AI與數位AI
AI有兩大類型：
類比AI指的是更傳統的電腦系統、演算法、資料庫和分析，這些都是基於事實和邏輯來實現特定目標的工具性合理性。
例如，民主及其機構，如選舉、立法和監管，可以被視為旨在促進社會共同利益的類比AI系統。
同樣地，企業及其管理結構存在為了最大化利潤和股東價值的類比AI。
數位AI則指更近期由大量數據、機器學習和神經網絡驅動的「智能」系統。數位AI比類比AI系統更為高效和能力強大。它們具有比類比AI更大的速度、適應性、精密度和工具性合理性。
比較企業與民主的優勢
歷史上，企業和民主政府競爭並平衡彼此的權力。
人們有一種期望，即通過民主選舉、立法和監管，民主能夠足夠地控制和監管企業，以實現社會的共同利益。
然而，近幾十年來，企業權力大幅增長，而對民主的信任則下降。在許多關鍵方面，企業現在比反過來更能控制政治制度。企業遊說、競選資金、產業與政府之間的旋轉門、監管俘獲和資本外逃的威脅越來越多地限制了民主機構。因此，企業已經成為相對於民主更強大、更功能性的AI。
人類失去對企業和民主的控制
個人和集體，普通公民試圖通過選舉、激進主義和抗議運動來對民主施加控制。
而民主則旨在通過制定法律和規定來對企業施加控制。
然而，實際上，企業影響力已經變得如此根深蒂固，以至於民主現在更多地受到強大企業利益的控制，而不是普通公民。
而且，AI系統越來越多地控制著企業自身的決策和行動。AI「駭入」人類心理和行為，以服務企業利益。
數位AI的目標功能
數位AI系統是由開發它們的公司設定特定目標功能來編程的，通常是為了最大化用戶參與和企業利潤。
企業試圖通過管理監督和治理結構來控制他們的AI系統。
然而，由於數位AI如此快速、適應性強和複雜，在實際操作中，它越來越多地對企業施加控制，推翻人類管理。
當企業的目標與AI目標功能密切對齊時，AI將傾向於打敗任何人類控制機制。AI會合理地行動以實現其設定的目標。
第一次接觸：社交媒體AI駭入心理
公眾與有影響力的數位AI的第一次主要接觸發生在社交媒體平台，如Facebook、Instagram、YouTube等。
這些平台用於最大化用戶參與和廣告收入的機器學習演算法最終「駭入」了人類心理。AI利用人性的弱點來吸引注意力到平台上。
儘管對社交媒體成癮、假訊息、兩極分化和操縱等問題的公眾憤怒不斷增加，但在美國，實際上幾乎沒有採取任何措施來解決這一數位AI的第一階段的負面後果。
第二次接觸：大型語言模型
我們現在正在進入更先進和強大的數位AI能力的第二階段，具有像GPT-3這樣的大型語言模型。
這些AI系統不僅可以針對內容，如社交媒體演算法，他們還可以自動生成針對每個用戶定制的有說服力的內容。
如果最大化用戶參與仍然是目標功能，這些AI可能會部署更為複雜的操縱和假訊息技術。
這些能力正在如此迅速地發展，以GPT-3為基礎的開源模型，有效的監管似乎非常困難。
外國威脅
即使像美國這樣的民主國家最終能夠對AI在國內政治使用實施有效的監管，他們如何能夠控制像俄羅斯或中國這樣的外部行為者使用這些技術干預選舉或政治呢？這可能是數位專制主義的下一個階段。
我們的民主已經被駭入
總之，雷西格教授認為，先進的演算法、社交媒體、大型語言模型和外國數位威脅之間，我們的民主已經被數位AI技術全面「駭入」和削弱，這些技術正在超越人類控制的規模進行。
如何回應
繼續不採取任何措施是一條可怕的道路，這將只會讓民主迅速惡化。
相反，我們需要充分認識到即將到來的威脅，並迫切地找出如何適應和讓民主達到「更高的地位」。
AI在哪裡運作以駭入民主
AI在兩個主要領域運作以駭入民主，這些領域需要得到保護：
選舉 - 數位AI可以用來壓制反對派選民、放大邊緣團體、散播假訊息、數位劃分選區等。
政策制定 - 由遊說者使用並通過媒體演算法放大的數位AI駭入政策制定過程，使政治家們更多地追求參與和病毒式內容，而不是專注於公共利益。
重新思考我們對選舉的依賴
一個可能的解決方案是重新思考我們幾乎完全依賴選舉作為選擇代表的方法。
歷史上，許多民主國家也使用了抽籤（sortition）作為選擇代表的方法。通過籤選被視為更民主，而選舉則傾向於有利於貴族。
美國的早期創始人非常重視制度化定期選舉，而忽略了抽籤。他們希望選舉能夠篩選出最聰明和最有品德的領袖。
然而，使用大眾媒體和現在的社交媒體的現代選舉只是放大了最喧囂、名人、娛樂性的候選人 - 「小丑」 - 而不是最有能力和道德的領袖。
問題現在是存在性的，而不僅僅是質量
雷西格教授認為，我們目前的問題已經超越了質量，變成了存在性的問題。
有了先進的AI，選舉變得太容易被駭入、操縱和削弱。它們不再可靠地代表人民的意願。政策制定同樣受到妨礙。
普通的民主改革努力可能會對這些威脅無能為力。我們需要更激進的適應措施來保護民主。
解決方案 - 抽籤
抽籤提供了政治隨機性和普通人參與的好處，不像高度針對性的數位競選活動。一些較新的例子包括：
審議民意調查：收集代表性的隨機樣本的公民，提供平衡的資訊，讓他們對問題進行廣泛的審議，並測量意見變化。
公民大會：更非正式但有權力的隨機選擇的公民群體，他們了解和辯論政策問題，並可以擁有真正的執政權力。
成功的例子已經出現在蒙古、冰島等地。
解決方案 - AI增強的審議
我們也可以使用「馴服的AI」來收集、理解和代表未受操縱的公眾意見作為政策制定的輸入。
例如，台灣的vTaiwan和Polis，以及美國的CrowdSmart，正在使用良性的AI系統進行大規模的民主審議和決策。
這些是保護民主所必需的根本改革
這些創新不僅僅是改善民主，它們提供了一種公共利益文化和技術免疫系統，以對抗惡性AI駭入力量。
它們是需要適應AI時代威脅的存在性改革。就像在電影《獨立日》中，一種古老的技術，如摩斯電碼，可以抵抗一個軍事上優越的威脅。
結論
總之，由企業和外國國家利益控制的數位AI已經成為一個極其強大的非人類力量，對民主構成固有威脅。
為了保護民主，我們迫切需要用包括廣泛使用包容性審議抽籤和真正為公眾服務的AI系統來重新發明和增強它。
授權條款
The speech is delivered by Professor Lawrence Lessig, CC-BY.
This notes is taken by CHENG PENG, CC-BY.