Ben Horowitz: RSI, Crypto as AI Money, & Classified Physics

Ben Horowitz believes artificial intelligence will discover the next theory of relativity within two years. This isn't venture capital hyperbole or marketing speak from the a16z co-founder , it's a concrete prediction about how recursive self-improvement in AI systems will accelerate scientific discovery to an unprecedented degree. The implications stretch far beyond physics labs into fundamental questions about how societies should regulate transformative technology.

Horowitz argues we've already entered the era of recursive self-improvement, where AI systems improve themselves through experimentation rather than human-directed programming. "You're discovering relativity. You know? It's just litanies of experiments with different, you know, different trials and then taking the one that worked and redeploying it, and now you have a smarter AI," Horowitz explains. This isn't gradual enhancement , it's a feedback loop where each improvement accelerates the next round of discoveries.

The timeline matters because it challenges the conventional wisdom about AI safety regulation. When asked directly about predicting "discovery by an AI of something as significant as relativity," Horowitz responds: "I think next two years." This timeframe puts AI-driven scientific breakthroughs ahead of most regulatory frameworks currently under discussion in Washington. If major physics discoveries happen faster than policy cycles, the question becomes whether regulation helps or hinders humanity's ability to solve existential problems.

Horowitz's experience briefing the Biden administration on AI policy reveals the fundamental tension between security concerns and scientific progress. Government officials suggested classifying AI advances similar to how nuclear weapons technology was restricted during the Cold War. But Horowitz told them they would essentially be "regulating math" , an approach he considers both technically impossible and historically ineffective. "The Russians did get, like, the bomb, including the exact the exact trigger mechanism, which was the most proprietary thing," he points out about nuclear classification efforts.

The classification precedent matters because it demonstrates how security-focused restrictions can fail while slowing beneficial applications. Horowitz sees AI as "probably the best chance we have" at addressing the 150,000 daily deaths worldwide from various causes. From his perspective, regulatory delays don't just slow economic growth , they postpone solutions to humanity's most pressing challenges. The risk calculation flips when transformative capability arrives faster than governance structures can adapt.

This connects to Horowitz's broader thesis about crypto serving as the natural currency for AI agents. As AI systems become more autonomous and capable of conducting transactions, they'll need payment systems that operate at machine speed without human intermediaries. Traditional banking infrastructure wasn't designed for sub-second, high-frequency transactions between artificial agents. Crypto protocols can handle this natively.

The economic implications extend beyond payment systems. Horowitz sees potential for "triple digit GDP growth in the next five years" driven by AI productivity gains. This isn't incremental improvement but a fundamental shift in how value gets created. When AI systems can accelerate scientific discovery, automate complex reasoning, and operate continuously without human limitations, traditional economic models break down.

The key insight from Horowitz is that regulatory frameworks designed for gradual technological change become counterproductive when dealing with exponential improvement curves. AI development operates on venture capital timelines , measured in months and quarters , while government policy operates on electoral cycles measured in years. This mismatch creates a coordination problem where well-intentioned regulation might prevent beneficial outcomes without meaningfully reducing risks.

For listeners, the practical implication is to watch for signs that Horowitz's timeline is correct. If AI systems begin making significant scientific discoveries within the next 24 months, it validates his broader argument about recursive self-improvement. It also suggests that policy debates focused on pausing development may be solving for yesterday's problems rather than tomorrow's opportunities.

The real question isn't whether AI will transform society , Horowitz considers that inevitable , but whether regulatory approaches will help societies capture benefits while managing risks. His prediction about relativity-level discoveries serves as a concrete benchmark for measuring whether we're moving faster than our institutions can adapt.