無需信任地執行任意程式碼,並足以共享給全世界使用,世界電腦的夢想深深地紮根於去中心化網路。在以太坊之後,許多基礎設施項目都做出了嘗試,Arweave 即將推出的 AO 網路也是這些嘗試之一。
對於一個「世界電腦」來說,可以粗略地分為資料的運算、存取與儲存三個部分,Arweave 過去一直在充當「世界硬碟」的角色,AO 網路(Actor Oriented)則引入了通用計算的能力,並提供了智能合約。
目前的主流去中心化運算平台分為兩類,即智慧合約平台與通用運算平台;智慧合約平台以Ethereum 為代表,網路共享全局的狀態內存,對改變狀態的運算過程進行共識,因為共識要求大量的重複運算,因此高成本下僅用於處理高價值業務;通用計算網絡不對運算過程本身進行共識,而是根據業務驗證計算結果,處理請求順序,不存在共享的狀態內存,這降低了成本,允許網絡擴展到更多領域的計算,這類以Akash 等算力網絡為代表。
當然,還有一些專案基於虛擬機器安全的安全假設,將通用運算與智慧合約整合。即共識只處理交易的順序,並對計算結果進行驗證,多個狀態變化計算在網路節點中並行處理,計算的環境虛擬機保證了確定性結果,因此只要交易順序一致,最終狀態也將一致。
這類網路由於不共享狀態內存,擴容成本很低,多個任務可以並行計算且互不影響。這類專案往往基於 Actor 程式設計模型,代表就是 ICP,而 AO 也屬於此類。 Actor 下每一個運算單元被視為單獨的智慧獨立處理事務,運算單元之間透過通訊互動(Actor 是傳統 Web2 服務中非常常見的架構)。 AO 標準化了 Actor 的訊息傳遞,實作了一個去中心化的運算網路。
與傳統被動觸發的智能合約(如Ethereum/Solana 智能合約)不同,具備通用計算Actor 下的AO 可以透過一致固定時間循環觸發的「cron」方式,來實現智能合約的主動運行,例如一個持續監控套利空間的交易程序。
可快速擴充的去中心化運算能力,Arweave 的超大資料儲存能力,Actor 的程式設計模型,與主動觸發交易的能力,這讓 AO 網路非常適合託管 AI Agent。 AO 也支持將 AI 大模型引入區塊鏈的智能合約中運行。
上文介紹過AO 與智慧合約網路的區別,AO 不對計算過程進行共識,對交易順序進行共識,並默認虛擬機的運作結果是確定性的,從而實現最終狀態的一致性。
AO 也具備一定的彈性,網路以模組化方式設計。網路中存在三種基本單元,調度單元 SU、計算單元 CU 與信差單元 MU。
一個交易被發出,作為通訊層的信使單元接受交易,驗證簽名,轉發給調度單元;調度單元可以看做AO 與AR 鏈的連接點,幫助網絡對交易順序進行排序,並上傳至AR 鏈完成共識,目前的共識方式是POA(權限證明);對交易順序的共識完成後,任務被分配給計算單元,CU 負責處理具體計算,結果返回MU 轉發給用戶。
CU 集合可以看做是一個去中心化的算力網絡,在完整的經濟學規劃下,CU 節點需要質押一定資產,透過計算性能、價格等因素競爭,提供算力賺取收益,如果出現計算錯誤的情況,會被罰沒資產。這是一個標準的經濟學保障。
AO 做為通用運算平台,與 Ethereum 等智慧合約平台的差異顯而易見。與 AR 同為「世界硬碟」的 Filecoin 也推出了自己的智慧合約平台 FVM,但這是一種等效於 EVM 的狀態共識機架構,且在體驗上不如 Ethereum 等傳統智慧合約平台。
與 Akash、io.net 等去中心化運算網路不同,AO 依然保留了智慧合約能力,AO 最終也在 AR 儲存上維護了一個全域的狀態。
In fact, the most similar architecture to AO is ICP. ICP created the earliest paradigm of asynchronous computing blockchain networks, and AO continues the design of ICP to a large extent, such as only sorting transaction order, trusting virtual machine deterministic calculations, asynchronous processing of Actor models, etc.
The biggest difference is that ICP is based on container maintenance status, that is, each smart contract container can only maintain its own private status, or set conditions for status reading; while AO has a shared status layer, That is, AR, anyone can restore the entire network status through transaction sequence and status proof. This increases the decentralization ability of the network to a certain extent, but also loses the possibility of realizing special privacy services in ICP (for example, customers have hidden arbitrage paths) needs).
In terms of economy and design, in order to ensure network performance, ICP has made higher hardware requirements for participating nodes, which has resulted in a higher threshold, while AO operates in a relatively fair launch and no-access manner. , you can participate in competitive mining by staking. The ICP network chooses a large stack implementation, sacrificing flexibility for performance, while AO uses a modular design, with MU, CU, and SU separated. Users can also choose their own virtual machine implementation, which also reduces the cost for some developers. Cost of entry.
Of course, AO may also have the same system shortcomings as ICP, such as the lack of atomicity in cross-contract transactions under the Actor asynchronous model, which will lead to difficulties in the development of DeFi applications. The idea of AgentFi seems difficult to realize in a short time ; New computing models that break away from the traditional smart contract paradigm also put forward higher requirements for developers. However, the wasm virtual machine under the AO architecture has a maximum management limit of 4GB, which also makes some complex models unable to be used on AO. From this point of view, AO’s choice of AI Agent is actually to maximize its strengths and avoid weaknesses. Interestingly, ICP also announced in early 2024 that it will focus on the AI field.
Of course, compared to ICP’s total market value of US$5 billion, AR’s current total market value is US$2.2 billion, which is still a big gap. In the context of the rapid development of AI, AO may still have great potential.
以上是技術解讀:AO如何打造適用於AI Agent的去中心化運算網路?的詳細內容。更多資訊請關注PHP中文網其他相關文章!