François Chollet on Why Scientific Progress is Linear, Not Exponential, and the Limits of AI Explosion

The core argument presented is that scientific progress, and by extension, the progress of problem-solving systems like artificial intelligence, is fundamentally linear rather than exponential, despite an exponentially increasing consumption of resources. This perspective directly challenges the prevalent narrative of an impending "intelligence explosion" or "singularity." The speaker emphasizes that what is often misconstrued as exponential growth in science (e.g., the number of published papers or patents filed) is merely an indicator of escalating resource input, such as the number of researchers and computational power, rather than a true measure of output or the significance of new knowledge generated.

A crucial distinction is drawn between resource consumption and the actual output or significance of scientific breakthroughs. The speaker references research by Michel Nielsen, which attempted to quantify scientific progress by measuring the "temporal density of significance" of discoveries over the past 100-150 years. This study reportedly found a consistently flat graph across various disciplines, indicating linear progress. This linearity is attributed to the phenomenon that as a field matures, the initial "low-hanging fruit" are quickly picked, making subsequent discoveries exponentially more difficult and resource-intensive to achieve the same level of impact or significance.

The concept of "exponential friction" is introduced to explain this linearity: as progress is made and one bottleneck is overcome, another part of the system inevitably becomes the new limiting factor, preventing unbounded acceleration. This principle applies broadly, from physical systems encountering air resistance to the institutional challenges of science, such as the vast amount of prior knowledge researchers must ingest, the increasing overhead in communication and synchronization among a growing number of scientists, and the exponentially rising cost of experimental equipment. The speaker suggests that the scientific community dynamically adjusts its resource investment to maintain this linear rate of progress, akin to a market for discoveries.

Finally, the discussion extends to the broader implications for AI, positing that AI development will also be subject to these same systemic constraints, leading to linear rather than exponential progress in terms of significant breakthroughs. The speaker characterizes the "singularity" narrative as more of a belief system than a scientific argument, noting that challenging it often elicits strong pushback because it is deeply intertwined with the identity and worldview of many AI enthusiasts. The overall implication is that while AI will continue to advance, it will likely do so within the bounds of inherent systemic friction, rather than leading to an unbridled, exponential intelligence explosion that renders human intelligence obsolete.

scientific progress is not actually exploding if you look at science what you see is the picture of a system that is consuming an exponentially increasing amount of resources it's having a linear output in terms of scientific progress

what you should look at is the output is progress in terms of the knowledge that sense generates in terms of the the scope and significance of the problems that we solve

if the output of Sciences institution were exponential you will expect the temporal density of significance to go up exponentially... and what actually happens if you if you plot this temporal density of significance measured in this way is that you see very much a flat graph

as you make progress you know in a given field or in a given substance it becomes exponentially more difficult to make further progress

the resource consumption of science is exponential but the output in terms of progress in terms of significance is linear

exponential promise triggers exponential friction so that if you tweak one part of a system suddenly some other part becomes a bottleneck

there is an amount of knowledge you have to ingest which is huge so there's a very large overhead to even start to contribute

AI is not just a subfield of computer science it's more like a belief system like this belief that the world is headed towards an event the singularity past which you know I will become we go exponential very much and the world will be transformed and humans will become obsolete

if you go against this narrative because because it is not really a scientific argument but more of a belief system it is part of the identity of many people if you go against this narrative it's like you're attacking the identity of people who believe in it