Emin Gün Sirer, the founder of Avalanche blockchain, believes that smart contract programming is a challenging task due to the difficulty in capturing “intent” and the complexities involved in coding and verification. He highlighted this issue at the recent Cornell Blockchain Conference in New York City, stating that it is the main obstacle preventing the widespread adoption of smart contracts.
However, Sirer proposed an intriguing solution: using artificial intelligence (AI) agents like ChatGPT or Llama to perform smart contract coding. In this scenario, lawyers would become the primary writers of smart contracts instead of coders. Furthermore, ordinary individuals would be able to write smart contracts in their native language, making it as simple as writing a bank check.
Sirer explained that this future vision is still five to 10 years away but emphasized its transformative potential. It could allow billions of new users to join the blockchain ecosystem. Currently, most smart contracts are written in the computer language Solidity, which is unfamiliar to many programmers.
To address this limitation, Avalanche is developing a new virtual machine that combines AI and blockchains. This machine is programmed in natural languages such as English, German, French, Tagalog, and Chinese. However, numerous challenges need to be overcome before AI agents can be widely deployed for smart contract writing.
Legal issues must be resolved, and keywords and terms must be unambiguously defined. Additionally, AI agents may not be ready to generate legally binding documents, as they can sometimes produce inaccurate or false information. Nonetheless, there are potential solutions like requiring multiple AI agents to agree on a smart contract solution before it is accepted.
Some experts argue that smart contracts capable of handling natural language already exist or are close to being developed. For example, the AI agent framework Council, developed by ChainML, enables developers to provide plain English instructions that are immediately converted into executable code. Similarly, Chainlink Developer Hub’s “Ask AI” can generate fully formed smart contracts based on plain English requests.
Despite these advancements, there is still work to be done to improve the quality and complexity of AI-generated smart contracts. Sam Friedman, a principal solutions architect at Chainlink Labs, believes that AI models need to be continuously trained on new content to enhance their capabilities.
Nonetheless, Friedman envisages a future where multiple AI agents collaborate to manage the lifecycle of a smart contract. This collaboration could involve agents with different roles, such as creating, verifying, and executing the contract.
Sirer acknowledged that there are still many unknowns in this field, and progress will require input from individuals with diverse backgrounds in technology, ethics, and humanities. He also emphasized that lawyers would play a significant role in writing smart contracts since they are adept at using regular language to specify contracts.
In conclusion, Sirer believes that this new approach to smart contracts holds tremendous potential and should not be ignored. He envisions a future where democratic and open platforms enable anyone to use these capabilities on an equal footing.