Fireside Chat, Interview
Who Will Own the Internet? a16z’s Chris Dixon on AI and Crypto
- Technology Convergence Thesis: AI, crypto, and new hardware (robotics, self-driving cars, VR/AR) are viewed as a reinforcing "trifecta" of technology waves that complement each other, similar to the previous convergence of cloud, mobile, and social.
- AI Architecture Control: There is a significant trend toward closed-source AI models, driven by defensive business reasons rather than pure safety; this centralization threatens to concentrate power in a small set of companies, prompting investments in open, decentralized AI infrastructure.
- Decentralized Compute Models: New crypto-based services (e.g., Jensen) are creating crowd-sourced compute networks that function like an "Airbnb for compute," allowing startups to access distributed processing power and establishing economic ledgers for supply and demand.
- IP and Royalty Enforcement: Projects like Story Protocol utilize blockchain to register intellectual property, enabling creators to set automated, enforceable licensing terms (e.g., mandatory revenue sharing on downstream derivatives) that mirror legal agreements and allow money to "waterfall" back to original creators.
- The "New Covenant" Breach: The current AI model breaks the historical "economic covenant" of the internet where platforms provided traffic to creators in exchange for indexing their content; AI chatbots now provide answers directly, diverting traffic and revenue away from the original content creators.
- Human Verification Systems: Initiatives like Worldcoin (co-founded by Sam Altman) aim to prove human identity via cryptographic methods (e.g., iris scanning, passport verification) to prevent AI-generated spam and replace CAPTCHAs, creating a verifiable layer for human-to-machine interaction.
- Physical Infrastructure Incentives: Decentralized Physical Infrastructure Networks (DePIN), exemplified by Helium, use crypto tokens to bootstrap community-owned networks (telecom, energy, climate modeling) that undercut traditional incumbents by leveraging crowd-sourced hardware.
- Skeuomorphic vs. Native AI Stages: The speaker defines a three-phase rollout:
- Skeuomorphic Phase: Replacing existing jobs/tasks with AI (e.g., customer service bots) to do old things cheaper; this phase is estimated to last roughly 20 years.
- Native Phase: Creating entirely new art forms, media, and behaviors that did not exist before (analogous to the transition from photography to film).
- Second-Order Effects: Unforeseen societal shifts and political movements resulting from the widespread adoption of native AI applications.
- Adoption Bottlenecks: The transition from skeuomorphic to native AI is constrained not by technical capability but by human creativity, organizational culture (e.g., Hollywood union resistance), and heavy regulatory frameworks in sectors like healthcare and finance.
- Regulatory and Legal Uncertainty: Current copyright lawsuits and state legislation (e.g., California AI bills) are addressing the fundamental question of whether AI training constitutes "copying" or "learning"; the speaker predicts a eventual compromise via congressional legislation rather than purely market-driven outcomes.
- Market Concentration Risk: Current trends indicate an internet where economic value and control are flowing to a centralized "middle" (approx. 5 major companies), contrasting with the original 1990s vision of value flowing to the "edges" (individual creators and small businesses).
- Strategic Investment Focus: The firm prioritizes "little tech" (small, decentralized startups) and open-source AI to prevent the internet from becoming a static, broadcast-style ecosystem controlled by incumbents, aiming to re-establish architectures where value and governance are distributed.
- Consumer Network Effects Limitation: Unlike social platforms, current generative AI tools lack inherent network effects; user data does not significantly improve the model for others, leading to potential price competition and vulnerability to incumbents copying novel features.