Generative AI has grabbed the limelight from Web3 in recent months. A little less than a year ago, several venture capital firms funded Web3 startups and crypto projects. Now, Generative AI is the apple of their eye. As the dust settles and a new winner emerges, some see a meaningful overlap between the two nascent industries. For starters, AI is capable of sifting through data, organising it, and presenting it in a useful way. This aspect of AI makes it industry agnostic in some sense.
“Intersections of AI and Web3 reveal instances where large language models (LLM) render on-chain data” that are humanly readable, Deep Gandhi, Investment Analyst with Hashed Emergent said. “Users can now simply input queries to understand on-chain data that previously seemed cryptic to most.”
While on-chain data is visible to everyone and contains all the details about the transactions recorded, like amount, time, wallet address and the fees, it isn’t necessarily interpretable by all. But LLMs are able to translate this complex blockchain data into simple natural language, making the technology far more accessible.
Belgium-based SettleMint, a low-code blockchain programming firm, added an AI assistant to its platform earlier this month to help Web3 developers write better smart contracts. Singapore-based Web3 startup Bunzz released an AI tool to help develop smart contracts with a click of button.
(For top technology news of the day, subscribe to our tech newsletter Today’s Cache)
Gandhi sees Generative AI streamlining processes in Web3 that “could attract numerous Web2 developers to the Web3 industry as they can now easily code customised smart contracts.”
AI coding assistants like GitHub Copilot have made it clear that AI inevitably makes life easy for coders. “It simply frees resources to focus their time and energy on better things. A team that may have needed three developers, may now only need two, and the additional resources can be deployed in UI/UX improvements. The productivity and cost improvements are key, especially to early-stage startups, as they can increase their runway and their chances to survive and succeed,” Ian Wittkopp, Head of Investments at Sino Global Capital said.
From a cybersecurity standpoint, AI helps in making “transactions more efficient during tokenisation thereby promoting more democratisation,” Sunny Vaghela, co-founder of Web3 cybersecurity firm, Zyber365, said. “With AI, there are more intelligent ways to secure data while sharing and maintaining it. AI will help make Web3 more open.”
Mutual benefits
If AI can smoothen the rough edges of blockchain, there are some qualities it can definitely take - namely the truth-seeking verifiability of the technology and its decentralised nature.
As the Internet is flooded with text, images, and audios, there are few ways to accurately gauge how authentic they are. “In this context, we’ll need mechanisms to verify whether something is human-generated or AI-created. In Web3, transactions are authenticated using the user’s private key to prove their legitimacy. Similarly, content across various modalities can be authenticated using the creator’s private key to prove its genuineness where one can verify signatures against the creator’s public wallet address,” Gandhi explained.
“Web3 tools can add a lot of value to Generative AI,” Raghav Agarwal, Investment Analyst with LongHash Ventures said. “Decentralizing the training process of models prevents AI hegemony in the hands of a few Big Tech companies.”
LongHash has long been exploring the meeting point of Web3 and AI. The fund invested early in ChainML, a community of independent actors that provide compute capacity for training and running AI models in a decentralized manner that can be accessed directly through smart contracts.
Gandhi suggests another way Web3 tools can help authenticate AI models using zero-knowledge proof (ZKP). Blockchain has a cryptographic technique called ZKP that allows one party to prove the validity of a statement to another party without revealing any specific information about that statement itself.
“We anticipate ZKP to let model providers make their model/weights private using ZKPs at the same time ensuring authenticity. ZKP provide unique advantages for model inferencing such as enabling users to share private data with model providers without exposing the actual data content,” he said.
Using this method, credit-scoring firms, for instance, can share information on users with lenders without revealing their personal financial information. And healthcare-related firms can share information about patients without disclosing specific medical reports. ZKP can assuage security concerns arising from use of data in LLMs.
