About Me

The Industrial Revolution changed the relationship between human capability and the human body. Before industrialization, much mechanical work came from people, animals, wind, and water. Coal and steam engines changed that. In Britain, installed steam power grew from about 35,000 horsepower in 1800 to 9.66 million by 1907. Productive capacity was becoming less tied to biological capacity. ^[1]

A similar shift later happened with computation. Intel's 4004 processor contained about 2,300 transistors in 1971. By 2021, average processor transistor counts had reached about 58.2 billion. More and more computation moved from human minds into machines. [2][3]

AI changed the role of those machines again. Modern systems can learn from data and perform tasks that once required substantial human reasoning. By 2026, advanced models were performing strongly across mathematics, science, multimodal reasoning, and computer use. On OSWorld, AI agent performance rose from about 12 percent to 66.3 percent. [4]

AI has also entered science itself. In 2025, around 80,150 papers in the natural sciences were related to AI. Some systems can generate hypotheses, write code, run experiments, analyze results, and interact with laboratory equipment. The gap with human researchers remains substantial. On one research benchmark, the best AI agent scored 38.8 percent, compared with 83.5 percent for PhD experts. [5]

This creates a new possibility. Earlier technologies amplified human capability, while humans still designed the next generation. AI is beginning to participate in that process itself. Technology produces AI. AI contributes to better technology. Better technology supports stronger AI.

The idea has been discussed for decades. I. J. Good described an intelligence explosion in which increasingly capable machines could help design their successors. Stuart Russell, Stephen Hawking, and Ilya Sutskever have discussed related possibilities. The idea is still speculative, but parts of the mechanism are now measurable. METR found rapid growth in the length of tasks frontier AI agents could complete autonomously, while warning that the trend should not be treated as a guaranteed forecast. [6][7]

If this feedback loop continues to strengthen, intelligence may become much less scarce. The next constraint does not disappear. It moves.

A design still has to become a physical system. That requires energy at the right place and time, materials in usable forms, manufacturing capacity, cooling, infrastructure, and sometimes years of construction. The limiting factor is rarely the absolute existence of energy or matter. It is how quickly, cheaply, reliably, and sustainably they can be turned into useful physical capability.

The scale of that problem is already visible. Global material extraction reached about 106.6 billion tonnes in 2024, more than three times its 1970 level. UNEP projects that, under current trends, resource extraction could rise another 60 percent by 2060. Global final energy consumption exceeded 450 exajoules in 2024, with industry accounting for nearly 40 percent. Many industrial processes require temperatures above 500 degrees Celsius. [8][9]

AI itself makes the physical bottleneck unusually clear. Data centres consumed about 415 TWh of electricity in 2024, and the IEA projects roughly 945 TWh by 2030. Generation is only part of the problem. More than 2,500 GW of generation, storage, and large loads are waiting in grid connection queues worldwide. A data centre can sometimes be built in one to three years, while major grid infrastructure can take five to fifteen. [10][11]

That mismatch is what interests me.

I am interested in the engineering systems that determine how fast an idea can become physical reality. Energy generation matters, but so do power conversion, heat removal, materials, manufacturing throughput, robotics, control, and the infrastructure connecting them. None of these is the answer by itself. They are coupled constraints on the same physical system.

My working name for this direction is Matter and Power. Matter is what we transform. Power determines how quickly energy can be delivered and converted to transform it. Engineering determines how effectively that can happen under real constraints. I do not use the phrase to claim that matter and power are the only limits on civilization. It is a way of describing the physical side of the problem I want to work on.

How can we increase the rate at which civilization can turn knowledge into useful physical systems?

If AI accelerates discovery, physical engineering determines how much of that discovery can leave the computer and enter the world. Better AI can help design better physical systems. Better physical systems can support more computation, experimentation, and intelligence.

One loop expands what we can discover. The other expands what we can build. That intersection is where I want my work to be.