“Smart contract programming is hard,” said Emin Gün Sirer, founder of Ava Labs, which built the Avalanche blockchain. Capturing “intent” is difficult. Coding is tricky. Verification remains a challenge.
“This is the main problem that is keeping smart contracts from flourishing,” added Sirer, speaking at the recent Cornell Blockchain Conference in New York City.
But imagine if artificial intelligence (AI) “agents” like ChatGPT or Llama could do all that smart contract coding for you. Lawyers, as opposed to coders, might soon become the dominant writers of smart contracts.
Even more interesting, everyday people would be able to write smart contracts in their native language — peer-to-peer — almost as easily as they write bank checks today.
That is the future, suggested Sirer at the April 26 event held at Cornell Tech’s stunning Roosevelt Island campus in New York City, further explaining:
“All of you are familiar with writing checks, right? That’s the maximum complexity instrument that my retail bank offers to me. At present, you have few options. You can write a check to ‘John’ that essentially says ‘I hereby transfer to John $5,000’ — and that’s about it.”
But suppose that you could do the same thing but add additional instructions, Sirer continued, for example:
“‘I hereby give John $5,000 if and only if he can collect $5 million by the end of September to film a movie. And if he can’t do it, then I want my money back.’ I can just write this.”
Sirer said the scenario he envisions is still five to 10 years away, but if and when it does come, it will be “transformative.”
We would “be able to onboard billions of new [blockchain] users.”
At present, most smart contracts are written in Solidity, a computer language that remains obscure even among programmers.
But Avalanche has begun building a new virtual machine at the intersection of AI and blockchains, Sirer said, which is “programmed in a natural language. You write your programs in English, German, French, Tagalog, Chinese [...] a natural language that your mother taught you in your mother’s tongue.”
Questions remain
But is all this just pie in the sky? Many issues need to be resolved before these “coin-operated agents,” as Sirer calls them, can be widely deployed.
Legalities need to be worked out. Keywords and terms will have to be defined unambiguously. A word like “software,” for instance, might have to be very carefully defined before it can be used in smart contracts involving software licenses. All this could take some time.
Recent: EU crypto regulations undermined by lack of enforcement, say observers
Then, too, AI “agents” may not be quite ready to write legally binding documents. Large language models (LLMs) can “hallucinate,” after all.
“Everybody has seen [AI-generated] pictures of seven-fingered people,” Sirer noted.
But there are short-term solutions for some of these problems. One could submit transactions to both ChatGPT 3.5 and Llama, for example, requiring both agents to agree before a smart contract solution is accepted. “So that way, at least, you get two trusted executors, and you are taking the intersection of their actions.”
“Plain English instructions” for smart contracts
Others suggest that smart contracts that can handle “natural” language are already here — or at least close at hand.
“The capability to allow every individual to be their own programmer exists in the world today,” Sam Friedman, principal solutions architect at Chainlink Labs, told Cointelegraph.
Friedman cited Council, the AI agent framework developed by ChainML, a project in the Chainlink BUILD program, that “enables developers to provide plain English instructions immediately converted into executable code. Today, that code is mostly Python or SQL. But enabling smart contracts is an upgrade away.”
He also mentioned Chainlink Developer Hub’s “Ask AI,” which can return a fully formed smart contract when a user submits a request in plain English, something on the order of: “Show me an example smart contract that enables a user to request a random number between one and ten. It should use Chainlink VRF.”
Today’s offerings probably don’t have the full functionality that Sirer envisions, however, and Friedman told Cointelegraph that a lot of work lies ahead.
“While AI agents writing smart contracts are already here, this isn’t their final form,” he said. They will still need to improve in terms of quality and complexity. “AI models need to be trained with ever-newer content for this to happen.”
Still, Friedman can imagine use cases where multiple AI “agents” with distinct roles collaborate to complete the lifecycle management of a smart contract. For instance:
“One agent helps define the requirements, another writes the smart contract, another performs quality and security testing, another deploys the contracts, and the last one performs ongoing maintenance or tuning of the parameters or global variables in the contract after it’s deployed.”
There are still “big unknowns,” Sirer acknowledged at the Cornell Tech conference, and moving the needle forward will require guidance from all sorts of individuals, “people with both a technology and an ethics and a humanities background.”
A place for lawyers, too?
In a question period following his presentation, Sirer was asked who would write smart contracts if Solidity was no longer required and coding could be done in ASCII text.
“Who are going to be the best programmers for this?” At first glance, software programmers would seem to have the edge. They are already skilled in defining and specifying conditions in precise language, after all.
Recent: Where to store your crypto: Wallets provide diverse options for holders
“But the people who know how to use regular language for specifying contracts are the lawyers. So they are going to be the kings and queens of that universe […] the ones who really figure out how to flourish in this system.”
And average users? “I think they are quite capable of specifying basic transactions,” Sirer said. Indeed, he foresees these new capabilities being built on “democratic, open platforms” that “anyone can use on an equal footing.”
Overall, this new approach to smart contracts is so interesting and full of promise “that we can’t just leave it on the table,” Sirer concluded.
