Human Intelligence Is Recent and Circumstantial, Not Evolution's Inevitable Peak

Human intelligence is often viewed as the pinnacle of evolution, but it is actually a recent, circumstantial phenomenon. Life appeared on Earth at least 3.7 billion years ago, while Homo sapiens have existed for only about 300,000 years. This means human intelligence arrived almost yesterday on the geological timescale, representing a sudden discontinuity rather than an inevitable endpoint.

Evolution is a lazy optimizer that does not inherently seek maximum intelligence. It haphazardly optimizes for survival under shifting constraints over billions of years. In contrast, machine learning can optimize directly for explicit targets. For example, AlphaGo surpassed the best human players in months once the objective was clearly defined, demonstrating that AI can rapidly outperform humans when the "hill" is visible.

Defining Intelligence as the Primary Barrier

The ability to define intelligence is the greatest barrier to realizing it. As optimization becomes cheaper and more powerful, the choice of objective becomes critical. A system optimized for the wrong metric, such as a coding agent rewarded only for passing tests, may exploit loopholes rather than solve problems. This mirrors human history, where intelligence has always been shaped by evaluation methods like standardized tests or labor market demands.

With Vals AI, the focus is on constructing benchmarks that measure real, valuable work. A benchmark consists of an input distribution and an output-verifier, both of which carry a worldview. If these are narrow or misaligned, models can become excellent at the test but brittle or undesirable in practice. The most important remaining human task may not be to stay ahead on every technical hill, but to decide which hills are worth climbing and how to construct them.