The Vertigo of Scaling: Why We Can No Longer Ignore AI's Trajectory

In the world of artificial intelligence, hundreds of billions of dollars are currently being poured into computing facilities. This isn't speculative venture capital chasing a fleeting trend—it is foundational infrastructure built on a highly predictable formula: scaling up compute and data reliably yields more capable AI systems.

This relentless scaling is inducing a sense of "vertigo" among industry insiders, a phenomenon recently articulated by AI analyst Jack Clark during a lecture at Oxford University’s Institute for Ethics in AI. Using a forward-looking thought experiment, Clark highlighted the sheer velocity of AI advancement, pointing to the Epoch Capabilities Index (ECI). This metric, which aggregates AI performance across more than 40 distinct benchmarks, displays a growth curve that is practically vertical.

We have all seen the individual milestones make headlines. In early 2023, AI systems successfully passed the bar exam. By 2024, they were securing silver medals at the International Math Olympiad, and researchers are already mapping the trajectory toward gold medals and systems capable of co-authoring novel mathematical proofs. While it is tempting to view these as isolated breakthroughs, they are actually symptoms of a unified, rapidly accelerating engine of intelligence.

Because the financial and computational resources required for the next generation of models are already being deployed, a significant degree of future AI progress is essentially "locked in." This reality forces a stark societal binary: we must either proactively navigate tomorrow or passively withdraw from today.

Choosing to ignore the implications of this technology relegates individuals and institutions to a state of perpetual reactivity. On the other hand, confronting the future requires us to grapple with deeply uncomfortable possibilities. AI is on a path to becoming more capable than the aggregate of human intelligence. The ultimate frontier is recursive self-improvement—a scenario where AI systems become sophisticated enough to design and train their own successors, removing human engineers from the bottleneck of innovation.

Building such technology is arguably one of the most powerful ways humanity can expand the frontiers of science and tackle global challenges. Yet, assuming this endeavor comes without immense, even existential, risk is dangerously naive. Acknowledging AI's dual capacity—to either elevate our species or pose unprecedented threats—is the first necessary step toward responsible stewardship. The technology is rapidly outgrowing its creators; our societal frameworks must now race to catch up.

Key Points

• The Epoch Capabilities Index reveals an accelerating, exponential growth curve in AI performance across multiple domains.
• Massive financial investments in compute and data make continued AI advancements highly predictable and essentially 'locked in.'
• Society faces a definitive choice: actively shape the AI future or passively retreat into reactivity.
• The ultimate governance challenge lies in managing the transition toward AI systems capable of recursive self-improvement.

Why It Matters

As AI transitions from individual breakthroughs to a systemic force capable of self-improvement, proactive societal governance is essential to harness its benefits and mitigate extreme risks.

Sources:

• Import AI 458: Reckoning with the future; and a singularity story — Import AI (Jack Clark)