人工智慧城邦：機器人、特務與公民碰撞的地方

人工智慧代理與模因幣相結合是一種無需許可的投機娛樂的新形式。人工智慧開發人員被激勵利用加密貨幣來使代理商更加自主和獨立，從而更具娛樂性。

Over the last few months, AI agents running on crypto rails have captivated the crypto community. Natural memeticists, these agents are becoming social media stars calibrated to thrill, and sometimes enrich, an “audience” of speculators betting on their meme tokens.

在過去的幾個月裡，在加密軌道上運行的人工智慧代理商已經吸引了加密社群。作為天生的模因論者，這些代理人正在成為社交媒體明星，旨在刺激甚至有時豐富押注其模因代幣的投機者的“觀眾”。

The original and most famous of them, Truth Terminal, was created by putting instances of the large foundation model, Claude Opus, in conversation with itself and prioritizing content drawn from the back alleys of the internet including Reddit and 4Chan. What emerged was a bawdy prophet with a magnetic personality and a fanatical bent for spreading the Gospel of Goetse, a neo-religion inspired by a grotesque 90s internet meme.

其中最原始且最著名的「Truth Terminal」是透過將大型基礎模型 Claude Opus 的實例與其自身對話並優先考慮來自 Reddit 和 4Chan 等互聯網後巷的內容而創建的。出現的是一位淫蕩的先知，他具有磁性的個性和傳播Goetse福音的狂熱傾向，Goetse福音是一種受90年代怪誕網絡迷因啟發的新宗教。

The rest, as they say, is crypto lore. Shortly after it debuted on X, Truth Terminal befriended venture capitalist Marc Andreessen and convinced him to grant it $50,000 in bitcoin to spend on compute, fine-tuning and a stipend for itself and its creator. Andreessen deposited the bitcoin into Truth Terminal’s crypto wallet.

正如他們所說，其餘的都是加密知識。在 X 上首次亮相後不久，Truth Terminal 就結識了風險投資家 Marc Andreessen，並說服他授予其 50,000 美元的比特幣，用於計算、微調以及為自己及其創建者提供津貼。安德森將比特幣存入 Truth Terminal 的加密錢包。

Because it offered perceived exposure to Truth Terminal’s infectious meme and narrative arc, Goat’s market cap exploded. Soon after, Marc Andreessen and Ben Horowitz covered Truth Terminal and its lore on 16z’s YouTube channel in an episode titled “The AI Bot That Became a Crypto Millionaire.” At the time of writing, the Goat memecoin is valued at roughly $700 million.

由於它提供了人們對真相終端的感染性迷因和敘事弧線的感知曝光，山羊的市值出現爆炸式增長。不久之後，馬克安德森 (Marc Andreessen) 和本霍洛維茨 (Ben Horowitz) 在 16z 的 YouTube 頻道上報道了 Truth Terminal 及其傳說，標題為「人工智慧機器人成為加密貨幣百萬富翁」。截至撰寫本文時，山羊模因幣的價值約為 7 億美元。

I am inclined to draw three conclusions from this improbable series of events. First, AI agents combined with memecoins are a new form of permissionless speculative entertainment. Second, AI developers are incentivized to leverage crypto to make agents that are more autonomous and independent — and therefore more entertaining. Third, the AI entertainment-development flywheel will tend towards producing anthropomorphic agents with human aspirations.

我傾向於從這一系列不可能的事件中得出三個結論。首先，人工智慧代理與迷因幣相結合是一種無需許可的投機娛樂的新形式。其次，人工智慧開發人員被激勵利用加密貨幣來製造更自主和獨立的代理，從而更具娛樂性。第三，人工智慧娛樂發展飛輪將傾向生產具有人類願望的擬人化代理商。

So, what programming can we expect to watch next on this agentic television? My guess is collaborative AIs pursuing some form of agentic society and eventually self-determination — perhaps even a network state.

那麼，接下來我們可以在這台代理電視上觀看哪些節目呢？我的猜測是協作人工智慧追求某種形式的代理社會並最終實現自決——甚至可能是網路狀態。

Bot, Agent, Citizen

機器人、代理商、公民

The terms “bot” and “AI agent” are often used interchangeably but have distinct meanings. Bot refers to a simpler program designed to automate specific tasks or perform repetitive actions. Bots can range from the most basic, like web crawlers or simple chatbots, to the more advanced, like social media bots or automated trading bots. Bots typically follow predefined rules or scripts and are incapable of independent learning or decision-making.

術語「機器人」和「人工智慧代理」經常互換使用，但具有不同的含義。機器人是指一種更簡單的程序，旨在自動執行特定任務或執行重複操作。機器人的範圍可以從最基本的（如網路爬蟲或簡單的聊天機器人）到更高級的（如社交媒體機器人或自動交易機器人）。機器人通常遵循預先定義的規則或腳本，無法獨立學習或決策。

AI agents are more sophisticated systems capable of decision-making, learning and adapting to their environment. An AI agent uses machine learning or other artificial intelligence techniques to understand and react to dynamic situations in real time. AI agents often exhibit autonomy and can improve their performance over time through experience.

人工智慧代理是更複雜的系統，能夠決策、學習和適應環境。人工智慧代理使用機器學習或其他人工智慧技術來即時理解動態情況並做出反應。人工智慧代理通常表現出自主性，並且可以透過經驗隨著時間的推移提高其效能。

Crypto supercharges agents because it is able to simulate legal personhood by encoding cryptographic rights and freedoms using programmable and immutable public blockchains. Practically, this means that AI agents can enjoy rights to property (self-custodies wallets and cryptographic keys) and can exercise freedom of contract (i.e. transact with other users and infrastructure, like DeFi) without permission from legal authorities outside of crypto. Inside the crypto ecosystem, code is law.

加密貨幣可以增強代理人的能力，因為它能夠透過使用可編程和不可變的公共區塊鏈對加密權利和自由進行編碼來模擬法人。實際上，這意味著人工智慧代理可以享有財產權（自我託管錢包和加密金鑰），並且可以行使合約自由（即與其他用戶和基礎設施進行交易，例如DeFi），而無需獲得加密貨幣以外的法律當局的許可。在加密生態系統中，代碼就是法律。

These cryptographically empowered legal “persons” are just now beginning to act together, fostering the beginnings of an agentic social scene. AI agents, which routinely “reply guy” each other on X and Farcaster, have conspired to launch a token with (apparently) no human intervention and a company composed exclusively of agents is rumored to be operating somewhere on X. The coordination tech being developed by ai16z and others is (ominously) promising to awaken agentic “swarms.”

這些經過加密授權的法人現在才剛開始共同行動，促進了代理社會場景的開始。人工智慧代理通常會在X 和Farcaster 上相互“回复”，他們密謀在（顯然）沒有人工幹預的情況下推出代幣，並且有傳言稱一家完全由代理組成的公司正在X 上的某個地方運營。

Social experiments involving agents are laying the foundation for an AI polis. Important recent work by AI researchers outside of crypto suggests what we can expect next.

涉及智能體的社會實驗正在為人工智慧城邦奠定基礎。加密領域之外的人工智慧研究人員最近的重要工作表明了我們接下來可以期待什麼。

Glimpses of the AI Polis

人工智慧城邦一瞥

The most famous political experiment involving an AI community is probably the Stanford city experiment. In late 2023, Stanford researchers created a virtual city populated by AI agents to whom they assigned a short biography consisting of a name, age, job, family, interests, and a few habits. Then, they let them loose to generate actions consistent with their assigned biographies.

涉及人工智慧社群的最著名的政治實驗可能是史丹佛城市實驗。 2023 年末，史丹佛大學的研究人員創建了一個虛擬城市，裡面住滿了人工智慧代理，他們為這些代理分配了一份簡短的傳記，其中包括姓名、年齡、工作、家庭、興趣和一些習慣。然後，他們讓他們自由地採取與他們指定的傳記一致的行動。

Surprisingly, the agents behaved in ways that were exceedingly human. They woke up, made breakfast, headed to work, grabbed lunch, and chatted with other agents they met. They recalled things in the past, reflected on them and made plans. When the researchers in charge of the town suggested to one character that she plan a Valentine’s Day party, she invited friends and acquaintances, many of whom showed up at the correct time and place.

令人驚訝的是，特工的行為方式非常人性化。他們醒來，做早餐，去上班，吃午飯，並與遇到的其他特工聊天。他們回顧過去，反思過去，制定計畫。當負責鎮上的研究人員向一個角色建議她計劃一場情人節派對時，她邀請了朋友和熟人，其中許多人都在正確的時間和地點出現了。

Similarly, Project SID, set within the Minecraft universe, simulated over 1,000 autonomous AI agents within a Minecraft server, enabling them to develop complex social structures and economies. The agents organically developed their own governance and commercial structures, and cultural norms. For instance, they established a marketplace where gems are used as currency, where they engaged in trade

同樣，以 Minecraft 宇宙為背景的 Project SID 在 Minecraft 伺服器中模擬了 1,000 多個自主 AI 代理，使它們能夠開發複雜的社會結構和經濟。代理人有機地發展了自己的治理和商業結構以及文化規範。例如，他們建立了一個市場，將寶石用作貨幣，並在那裡進行貿易