AI and cryptocurrencies are two revolutionary technologies with different development paths: AI emerged top-down from prestigious universities, while cryptocurrencies rose bottom-up from underground forums.
During the recent SuperAI Conference in Singapore (June 5-6, 2024), former CTO of Coinbase, Balaji Srinivasan, presented an intriguing topic: “AI and Crypto, Beyond Cliches: AI Makes Everything Easy to Fake. Crypto Makes It Hard Again.” Here are the 10 main messages:
1. AI is probabilistic, Crypto is deterministic:
Modern AI is largely statistical while crypto is procedural. Compare diffusion models and LLMs to cryptography.
2. AI is digital abundance, Crypto is digital scarcity:
Generative AI augments (or substitutes) human writers, researchers, artists, lawyers, and doctors, making these services abundant. Crypto, by contrast, digitizes currency, stocks, loans, and assets, representing scarcity in a natively digital fashion.
3. AI makes things easy to fake, Crypto makes it hard again:
How do we know whether a given generated piece of text or image is “real”? We need some kind of cryptographic verification.
4. AI generates, Crypto authenticates:
Modify the IPFS/ENS stack to determine whether a piece of content was provably publicly authored by a given ENS name.
5. AI busts captchas, Crypto rebuilds them:
AI can now bust just about any CAPTCHA. So, we may need to retreat to cryptographic scarcity, paying or signing in with a high-reputation crypto address.
6. AI proselytizes, Crypto verifies:
In generated text, we don’t just want the output to have references; we want these to be on-chain references.
7. AI breaks the public web, Crypto builds Web3 of trust:
Much content on the world wide web is becoming fake or generated. But with “Web3 of trust,” we can start restoring signal for backlinks.
8. AI training is centralized, Crypto can decentralize it:
We have centralized training + centralized deployment (like OpenAI), and centralized training + decentralized deployment (like Llama 3). Ideally, we want decentralized training and deployment. Crypto crowdfunding could enable this.
9. AI evaluation is centralized, Crypto can decentralize it:
AI model evaluation is moving to something that can be done on individual beefy machines. Imagine decentralized evaluation of different versions of a model, similar to hosting a node.
10. AI creates new authorities, Crypto decentralizes them:
It’ll be a “polytheistic” rather than “monotheistic” model of AI. Every sufficiently large community crowdfunds an AI that serves as the central guiding force of their community. What would George Washington do, what would Lee Kuan Yew do?
Full presentation (20min) is available on YouTube: Watch here