正式驗證是計算機科學的理論領域之一。

從歷史上看，這一領域是晦澀難懂的，但是AI的最新進展可能會將其帶到前面和中心。

In the realm of computer science, few areas are as theoretical and hold as high a threshold for practical application as formal verification. It essentially takes the tools of mathematical logic and applies them to verifying whether statements are correct.

在計算機科學領域，很少有理論上的領域像正式驗證那樣具有實際應用的高閾值。從本質上講，它採用了數學邏輯的工具，並將其應用於驗證語句是否正確。

This field has remained largely in the academic sphere, but recent advances in AI may finally bring it front and center.

該領域主要保持在學術領域，但AI的最新進展最終可能使它成為領先和中心。

I spoke with Clark Barrett, a professor of computer science at Stanford, who tells of a software bug that once led to the explosion of a rocket. The software ran an instance that forced it to convert a floating-point number into an integer. This caused the program to crash and the rocket to explode. A formal verification of the code would have avoided that problem.

我與斯坦福大學計算機科學教授克拉克·巴雷特（Clark Ba​​rrett）進行了交談，他講述了一個曾經導致火箭爆炸的軟件錯誤。該軟件運行了一個實例，該實例迫使其將浮點數轉換為整數。這導致程序崩潰，火箭爆炸。對代碼的正式驗證將避免這個問題。

Compiling is the weakest form of verification. A stronger form would be to run a battery of test cases. To see this more clearly, consider a function that divides two numbers. Without doing any internal checks, that function could run on any numerical inputs. If your test cases excluded 0, your function would still compile. But the edge case of 0 in the denominator would cause the program to crash. Only a formal verification would catch this because it’s not sufficient just to evaluate the functions on the different inputs, but rather to assess the function on its underlying logic.

編譯是驗證的最弱形式。更強大的形式是運行一系列測試用例。要更清楚地看到這一點，請考慮一個劃分兩個數字的函數。無需進行任何內部檢查，該功能就可以在任何數值輸入上運行。如果您的測試用例不包括0，則您的功能仍將編譯。但是分母中0的邊緣情況會導致程序崩潰。只有正式驗證才能捕捉到這一點，因為它不足以評估不同輸入的功能，而是要評估其基礎邏輯的功能。

The bar for formal verification is high, and the tools are obscure and hard to use. Outside of the Mars rover, they have not had wide acceptance. But the one possible exception today is cloud services. Cloud providers allow customers to enter their own query logic when using their services. An error in the query logic, such as inadvertently typing “or,” instead of “and” can have existential consequences, giving everyone access instead of no one. As such, companies like AWS are now recruiting computer scientists in formal verification by the hundreds.

正式驗證的標準很高，工具晦澀難懂。在火星漫遊者之外，他們沒有得到廣泛的接受。但是今天可能的一個例外是雲服務。雲提供商允許客戶在使用服務時輸入自己的查詢邏輯。查詢邏輯中的一個錯誤，例如無意中鍵入“或”而不是“和”可以帶來存在的後果，可以使每個人訪問而不是沒有人。因此，像AWS這樣的公司現在正在招募數百人進行正式驗證的計算機科學家。

The big use case will be formally verifying code written by AI. As AI tools improve, more code will be written by AI, and we need fast and cheap ways to verify this code beyond simply compiling it. That’s where formal verification could have its Super Bowl moment. There is now a big research effort underway to deploy these formal verification tools at scale to AI-generated code.

大用例將是由AI編寫的代碼正式驗證的。隨著AI工具的改善，AI將編寫更多代碼，我們需要快速，便宜的方法來驗證此代碼，而不是簡單地編譯該代碼。這就是正式驗證可能擁有超級碗時刻的地方。現在，正在進行大量的研究工作，將這些正式驗證工具按大規模部署到AI生成的代碼。

This could have an enormous impact, making software bugs a thing of the past. Not only would software be written faster with AI, but it would be better too.

這可能會產生巨大的影響，使軟件錯誤成為過去。不僅可以使用AI編寫軟件，而且還會更好。

What about Bitcoin?

那比特幣呢？

Once these formal verification tools arrive, I’m eager to see how Bitcoin would fare. But the early answer here is that Bitcoin should fare well because it uses several strict forms of logic that give it its high security. For example, full nodes of the network check signatures (through SigOps) when verifying transactions. If the signature fails, the transaction will never enter the mempool, nor be included in a block. Similarly, miners win a block only if their hash of the block header lies below the difficulty target. And a transaction is valid only if the inputs exceeds its outputs.

這些正式驗證工具到達後，我渴望看到比特幣的票價。但是這裡的早期答案是，比特幣應該很好，因為它使用了幾種嚴格的邏輯形式，使其具有很高的安全性。例如，在驗證交易時，網絡的完整節點檢查簽名（通過sigops）。如果簽名失敗，則交易將永遠不會進入MEMPOOL，也不會包含在塊中。同樣，礦工只有當他們的塊標頭的哈希在難度目標以下時才能贏得一個障礙。僅當輸入超過其輸出時，交易才有效。

In other words, the logic in Bitcoin is fully deterministic. There is no uncertainty about the rules of the protocol. And because of this, there is little room for software bugs, evidenced by the lack of hacks over the last 15 years.

換句話說，比特幣中的邏輯是完全確定性的。協議規則沒有不確定性。因此，軟件錯誤幾乎沒有空間，這證明了過去15年中缺乏黑客攻擊。

That said, Bitcoin is still an example of social computing. You could say that it is technically vulnerable to collusion if, for example, every single miner in the world agreed to fork the chain. That could happen in theory. But that's where economics comes in: It would not be in the miner's interest to do so.

也就是說，比特幣仍然是社交計算的一個例子。您可以說，例如，如果世界上的每個礦工都同意為鏈條分叉，那麼它在技術上很容易受到勾結。從理論上講，這可能發生。但這就是經濟學的來源：這樣做並不符合礦工的利益。