Estimating the Scale of Digital Minds

TL;DR

The paper addresses the question of how many digital minds could exist in the coming decades by defining digital minds through observable traits such as agency, personality, and intelligence, and by assessing potential moral and social implications. It adopts two complementary methodologies: a use-case driven consumption model that maps adoption across social, task, and actor categories, and a compute-centered fractional-capacity model that projects global FLOP/s capacity and plausible allocations to digital minds. Together, these approaches suggest a wide-ranging potential scale, from about $10^6$ to $10^{11}$ digital minds by 2030–2050, with medians often in the billions, and a conclusion that hardware limits are unlikely to be the primary bottleneck in many scenarios. The work emphasizes substantial uncertainty, contingent on social acceptance, regulatory choices, and the value and deployment of computation, and highlights the importance of considering welfare implications for large-scale digital mind populations.

Abstract

This report estimates the potential number of digital minds, defined as AI systems exhibiting observable traits such as agency, personality, and intelligence, in the coming decades. It employs two complementary approaches: first, examining specific use cases for digital minds and projecting adoption rates for each; second, analyzing trends in AI chip production and efficiency independent of digital mind applications. Together, these supply- and demand-side perspectives suggest that hundreds of millions of digital minds could exist by 2050, though this estimate carries substantial uncertainty spanning several orders of magnitude.

Figures (18)

• Figure 1: Product distributions. These charts displays the relative makeup of digital mind populations at the median and mean of projections in 2035 and 2050.
• Figure 2: Weighted product distributions. These charts display relative digital mind population sizes at the median of projections in 2035 and 2050, as weighted by mindedness and utilization rates.
• Figure 3: Virtual Friends Projections. This chart displays a breakdown of the estimated time of first viable products (blue columns) and mean (solid red line) and median (dotted red line) projections for virtual friends numbers. The shaded area represents the middle 95th percent of all estimates.
• Figure 4: Virtual Guides Projections. This chart displays a breakdown of the estimated time of first viable products (blue columns) and mean (solid red line) and median (dotted red line) projections for virtual guides numbers. The shaded area represents the middle 95th percent of all estimates.
• Figure 5: Virtual Pets Projections. This chart displays a breakdown of the estimated time of first viable products (blue columns) and mean (solid red line) and median (dotted red line) projections for virtual pets numbers. The shaded area represents the middle 95th percent of all estimates.