Former Coinbase CTO and advisor to Chainlink Labs Balaji Srinivasan is consistently one of the most anticipated speakers at SmartCon. In 2021, he delivered a crowd-favorite presentation on how an onchain “ledger of record” could replace the world’s flawed fiat information model with a cryptoinformation model. In 2022, he expanded this concept to explain how oracles could facilitate decentralized governments based on his bestselling book, The Network State.
This year, Srinivasan addressed how blockchain, crypto, and oracles can combat the rapidly growing problem of AI-generated misinformation with a talk titled “AI Makes Everything Easy to Fake – Crypto Makes it Hard Again.”
“One way of thinking about it is: if AI breaks a public web, crypto builds a web3 of trust,” he said.
He began by outlining the severity of the problem, which is quickly outpacing human intervention. A popular technique for detecting AI-generated images of public figures, for example, is to examine the fingers, one of AI’s biggest anatomical obstacles. “That’ll eventually get fixed,” Srinivasan said, pointing to how quickly AI systems are improving.
The solution, in his view, is to tackle the root of the problem with a system for verifying an image’s source with a cryptographic signature from an Ethereum Name Service (ENS) address or equivalent onchain identity.
In Srinivasan’s vision of a web3 of trust, such onchain identities would serve as ledgers of record for every piece of content generated by a particular address so that “you can start to figure out how reliably human or how real that entity is.”
“We have a worldwide web where right now the kind of weight that’s being put on it in terms of determining whether content is real or not is too much,” he said, highlighting the growing volume of misinformation ranked highly in popular search engines. “You actually want a decentralized web3 of trust, which anybody can look at, anybody can index.”
The Chainlink Network has securely delivered more than 10 billion data points onchain to enable over $8.8 trillion in transaction value throughout decentralized finance (DeFi). Amid the expansion and diversification of Chainlink’s web3 services, Srinivasan anticipates the types of datasets that can be verified onchain will extend beyond financial transactions to include social interactions that deter the spread of misinformation.
Beyond this, he said crypto can prevent AI from busting captchas with systems that require onchain payment, onchain payment history, staked crypto, or a combination of the above so that fraud requires more resources.
“Of course it could still be a robot logging in with Ethereum, but you could charge something for it,” he explained. “And that scarcity is really what we’re getting at when you want to try to make spam costly.”
With the majority of AI systems being trained on centralized models, Srinivasan sees the opportunity to crowdfund decentralized training models in the vein of popular crypto projects.
“Huge amounts of money can be harnessed online to crowdfund things with crypto and we can potentially use that to train AI,” he explained. “I think it’s something to think about in terms of freeing these things from centralized actors.”
Instead of one “monotheistic AGI superintelligence,” Srinivasan imagines decentralized AI training could foster a more realistic “polytheistic AGI” model, concluding, “Different societies, different communities would have their own AGI, and they could query it, and they could ask it for citations, and it could even give onchain citations.”
Watch Balaji Srinivasan’s full presentation from SmartCon 2023.