In mid-March, Stanford University held a conference on Blockchain and AI, gathering professors, startup CEOs, and venture capitalists (VCs). The event's main focus was the convergence of two significant technologies: blockchain and AI. However, the conference might have benefitted from a stronger focus on Bitcoin and AI, considering Bitcoin’s leading market position and the new innovations emerging in Bitcoin Layer 2 solutions. A key issue at the event was that blockchain and AI have largely developed as separate fields—each with distinct investors, entrepreneurs, researchers, and communities. Although the idea of merging these domains was ambitious, many speakers remained centered on their own specialties, struggling to draw clear connections between blockchain and AI. Arguably, calling it the Blockchain OR AI Conference would have been more accurate. For instance, a venture capitalist gave a broad overview of the AI sector, highlighting remarkable progress in image, audio, and code generation. Meanwhile, a DeepMind researcher addressed adversarial machine learning, where small tweaks to input data can drastically change an AI’s output. A notable example involved altering just a few pixels in a cat image, causing the AI to misidentify it as guacamole. On the blockchain front, conversations revolved around various protocols, yet much of the technology remains highly experimental or, in some instances, entirely theoretical. Integrations between blockchain and AI are still nascent, with real-world applications yet to materialize. Proof of Computation One of the more illuminating talks came from Dan Boneh, an applied cryptographer at Stanford, who spoke about SNARKs (succinct non-interactive arguments of knowledge) and zero-knowledge proofs. These address a core cryptographic challenge: efficiently proving knowledge of a computation. This principle is well established in blockchain and cryptography.

For example, factoring a large number into primes is computationally hard, but verifying the factorization by multiplication is easy. Similarly, finding a block header whose hash meets a target difficulty is costly, whereas verifying this proof is inexpensive. This gap between computation and verification is essential in blockchain systems, where nodes verify each other’s work continuously. In Bitcoin, nodes validate signatures and miners’ proof of work. SNARKs expand this concept, allowing cryptographic proofs that can be verified without exposing sensitive information. As AI agents become more autonomous, verifying computations while preserving privacy will be a significant challenge. Many users hesitate to upload sensitive data to platforms like OpenAI due to security concerns. This creates a strong need for privacy-preserving verification—a method that lets users prove an AI model performed a computation correctly without disclosing the underlying data. Such technology could unlock AI use cases in sensitive fields like healthcare, defense, and finance, where data protection is critical. It’s expected this will evolve into a multi-billion-dollar industry over the coming decade. Notably, this idea originates from blockchain networks implementing cryptographic techniques. As Boneh noted, the concept of one machine efficiently verifying another’s expensive computation arose from Bitcoin, but it may find a major second application in AI. Looking ahead, I hope future conferences will place more emphasis on Bitcoin’s contributions to these areas. For example, BitVM builds on zero-knowledge proof concepts to bridge Bitcoin and new Layer 2 protocols—potentially enabling AI agents to interact directly with Bitcoin’s ecosystem.