Living with Exponential Change (with Azeem Azhar)

Episode Summary

Executive Summary: Azeem Azhar argues that technology is entering an age of compounding exponential improvement across computing, energy, manufacturing, and biology, but society’s norms and institutions are lagging behind. He explains Moore’s Law, Wright’s Law, and rapid adoption dynamics, highlights solar, batteries, and AI as continuing accelerants, and worries that smartphones and immersive devices may outpace human-scale social life unless new norms and governance emerge.

Main Topics: Exponential technology and the acceleration thesis (Priority: 5/5): Azhar frames the modern era as one in which multiple technologies improve exponentially, not just computing, creating faster diffusion, lower costs, and broad economic transformation. Moore’s Law and Wright’s Law (Priority: 5/5): He contrasts Moore’s Law (transistor density/performance growth) with Wright’s Law (cost declines through cumulative production and learning), arguing the latter is often more predictive across technologies. Solar, batteries, and the energy transition (Priority: 5/5): The conversation emphasizes steep cost declines in solar panels and batteries, modularity, decentralization, and implications for energy security, local agency, and market expansion. AI, compute demand, and adoption (Priority: 4/5): Azhar discusses the explosive growth in compute needs driven by AI and broader digital demand, while noting that AI is still finding practical applications in workflows, development, and science. Human capital, development, and productivity (Priority: 4/5): He argues AI can bootstrap human capital, especially in developing countries, by giving people access to expertise, support, and agency where institutions are weak. Social norms, smartphones, and the ‘exponential gap’ (Priority: 5/5): Roberts and Azhar debate whether society can adapt to devices that increasingly demand attention, potentially reducing conviviality and human-scale interaction. Governance, regulation, and institutional adaptation (Priority: 4/5): Azhar says existing legal, regulatory, and cultural frameworks are not keeping pace with rapid technological change; he favors a mix of norms, design limits, and possible regulatory pressure.

Key Arguments: Technology costs can fall exponentially across many domains, not only semiconductors, and this drives rapid adoption and economic change. Moore’s Law worked partly because of a shared industry norm and market incentives, not because it was a physical law like gravity. Wright’s Law captures learning-by-doing: cumulative production lowers unit cost through accumulated knowledge and process improvement. Solar power and batteries are becoming cheap enough to change energy systems from centralized, volatile commodity dependence to more predictable, decentralized models. AI/LLMs are best understood as information compression tools that reduce search and coordination costs across language-mediated work. AI could be especially transformative in developing countries by increasing human capital, access to knowledge, and personal agency. The social challenge is not just speed of information but the volume and persistence of attention capture, especially via smartphones and immersive devices. Current institutions, norms, and regulation are not adapting quickly enough to the pace of technological change; new norms and possibly policy pressure will be needed. Some technologies, like smartphones and social platforms, are optimized for engagement in ways that can undermine human conviviality and self-control. Technological progress should be evaluated not only by efficiency gains but also by whether it preserves human-scale life and meaningful social connection.

Data Points: Computer count in 1945: less than 100 - Azhar cited the small number of computers in the world at the dawn of modern computing. Computers today: more than 25 billion - He used this figure to illustrate the economy’s enormous appetite for computing. Moore’s Law growth: roughly twice the density every couple of years - Description of transistor scaling in semiconductor chips. Wright’s Law learning rate: about 15% decline per doubling of cumulative production - Historic example from aircraft airframes and learning-by-doing. TikTok user growth: faster than Facebook to about a billion users - Used to show acceleration in adoption of digital platforms. ChatGPT user growth: 100 million users within a matter of a few days - Example of the speed of adoption for generative AI. Solar panel cost trend: 15% to 19% compound decline since 1970 - Azhar described long-run solar price decreases. Chinese solar price change: halved the price of solar panels or a component within a year - Evidence of continuing rapid cost declines in solar. New solar additions since 2010: 61% compounded growth - Global annual net new solar additions have grown rapidly. Global electricity generating capacity in 2022: about 9 terawatts - Used as a benchmark for current world capacity across sources. Bloomberg forecast for solar by end of decade: 7 terawatts of new generating capacity over seven years - Forecast cited to underscore solar’s expected scale. Battery price decline: from over $1,000/kWh to approaching $100/kWh - Short-duration battery storage became much cheaper over roughly a decade. Estimated compute growth since 1971: about 65% per annum compounded - Azhar’s rough global compute estimate starting from the Intel 4004 era. AI training compute scale: 10^25 floating-point operations - Order-of-magnitude estimate for frontier model training. Large-capex by big tech: $50 billion a year plus - Scale of current server/compute investment by major cloud firms. Smartphone adoption gap: 1 billion+ people without smartphones; 2 billion without modern smartphones - Used to argue that compute demand and adoption are far from saturated. National energy access example: $5,000 - Approximate cost for a homeowner to add solar panels and connect to the grid today, contrasted with billions needed for traditional generation.

Pivotal Quotes: "We are at this moment where it's not just the digital products. It is the big, heavy physical ones that are also being deployed in our economies at rates that we haven't really seen before." — Azeem Azhar: On the broadening scope of exponential deployment beyond software into physical technologies like EVs and energy systems. "The game is not for us to try to keep up with it, but it's for us to work out how to govern it and harness it so that we can live at a human scale and a human speed." — Azeem Azhar: His central framing for coping with technology that may outrun human norms and institutions. "I like that. I'm not, you know, I'm not quite sure what it means, but I understand it." — Russ Roberts: Roberts responding to Azhar’s ‘human speed vs silicon speed’ framing and expressing both sympathy and uncertainty.

Implications: Listeners should expect cheaper energy, more compute, and more AI—but also more attention pressure and social friction. The main challenge is building norms, products, and policies that preserve human connection while capturing the gains from exponential technologies.