Thursday 10 April 2025
A team of researchers has developed a system that can convert natural language into transaction payloads for decentralized autonomous organizations (DAOs). This innovation has the potential to make it easier for people without technical expertise to participate in DAO decision-making processes.
The system, called AgentDAO, uses a combination of artificial intelligence and domain-specific languages to synthesize proposals. It’s capable of understanding complex commands and translating them into executable transactions. This could be particularly useful for governance proposals that require specific actions or changes to be made.
One of the key components of AgentDAO is its ability to extract relevant information from natural language inputs. This is achieved through a process called relation extraction, which identifies entities and their relationships within a sentence. The system can then use this information to construct a DAOLang program, which is a domain-specific language designed specifically for DAO governance.
The DAOLang program is then transpiled into a transaction payload, which can be executed on the blockchain. This process ensures that the proposal is implemented correctly and securely.
The researchers tested AgentDAO using 155 proposals from CompoundV3 Governance protocol between 2023 and 2024. They found that the system was able to successfully synthesize proposals in most cases, with a pass rate of around 89%. The system performed particularly well when handling composite actions and cross-chain interactions.
However, there were some limitations to the system. For example, it struggled with proposals involving inexpressible actions or those that required complex gas optimization. Additionally, while the system was able to understand natural language inputs, it wasn’t always able to generate concise transaction payloads, which could result in higher gas costs.
Despite these challenges, AgentDAO has the potential to make a significant impact on DAO governance. By providing a user-friendly interface for proposing and executing transactions, it could increase participation and engagement from non-technical stakeholders. This could lead to more diverse and representative decision-making processes within DAOs.
The research also highlights the potential of large language models in program synthesis. These models have been shown to be capable of generating code from natural language inputs, but they often struggle with complex tasks or require extensive fine-tuning. AgentDAO demonstrates that these models can be used in more practical applications, such as DAO governance, when combined with domain-specific languages and relation extraction techniques.
Overall, AgentDAO represents an important step forward in the development of decentralized autonomous organizations.
Cite this article: “Unlocking Decentralized Autonomy: A Multi-Agent System for Synthesizing Transaction Payloads from Natural Language Specifications”, The Science Archive, 2025.
Decentralized Autonomous Organizations, Dao Governance, Natural Language Processing, Artificial Intelligence, Domain-Specific Languages, Blockchain Transactions, Program Synthesis, Large Language Models, Relation Extraction, Transaction Payloads
