Constructing Science
A BEYOND BORDERS column by David Krakauer, President of the Santa Fe Institute.
We might like to know what sets the speed of scientific discovery. And if we knew this, how to accelerate, cruise, break, and perhaps even stop. In the last couple of years, a great deal has been made of the high speed of progress in machine learning, many suggesting that we have achieved as a cultural mass, a dangerous momentum — Elon Musk, for example, warning that “If you’re not concerned about AI safety, you should be.” This is reminiscent of the fear of speed in the early ages of the automobile. In an article in the Detroit Free Press from 1914, there are dire warnings over the risks of driving at 40 miles an hour: “An automobile… rounded the corner from Labelle Avenue onto Woodward Sunday evening and turned turtle going at least 40 miles an hour.” Admittedly the driver was in all likelihood intoxicated. But perhaps the same could be said of the legions of large language model users and their ghost-in-the-machine-written spam epistles.
The metaphor of the automobile, or machine more generally, is fitting in more ways than one. The science of science has at various points emphasized the idea that fundamental science, unlike statistical science, is slowing down. It could have run out of fuel (new ideas), been hindered by too much friction (institutional inertia), excessive congestion on the roads (too many researchers on a project), a paucity of streets and highways (conservative ideas about where science should go), and bad driving (poor educational foundations). And there is ample empirical evidence for every one of these factors. In a recent meeting at SFI on “Accelerating Science,” many of these topics were discussed and everything from the conservative forces operating within large teams to the economics of risk-aversion were covered.
I would like to take a stab at relating these ideas to what the physicist David Deutsch has called Constructor Theory. According to Deutsch, the theory seeks to renovate physics, or what he has called the “prevailing conception,” and to provide a rigorous, consistent framework for talking about possible and impossible transformations. My sense is that the true value of Constructor Theory is the way it allows us to talk about possibility and less for what it has to say about physics. The theory builds on John von Neumann’s Universal Constructor argument for the origin of life, and adds details from what we know about enzyme kinetics (EK).
Von Neumann pointed out that a non-trivial life-form requires far more than replication; it requires programmable development. His argument was reminiscent of the current AI Paperclip Apocalypse, where an algorithm might determine that its objective function is to saturate the universe through the endless replication of a trivial function. A simple replicator likewise might fill the universe with a single bit of information. Von Neumann understood that complex organisms require a means of replicating information, which provides a blueprint for “constructing” a functional machine. As the evolutionary theorist George Williams wrote, all of biology operates through both codical (replicating) and material (phenotypically interacting) domains. It is not a stretch to extend this idea to scientific knowledge, where our educational systems, funding sources, and risk-averse norms, all emphasize replication of existing ideas over the possible construction of new ones.
Within the Constructor Theoretic framework, there are elementary inputs to a machine, the constructor, and outputs that are new configurations of matter. As von Neumann first noted, simple replication can dispense with the constructor, but the constructor is where almost all the interesting affordances and constraints live. In ontology, gene regulatory networks play the role of constructors, and through phylogeny, natural selection is a constructor.
The process of knowledge creation — that is, epistemic causality — resides in constructor mechanics. And of course the constructor is a machine that obeys the second law, and so its contribution to the generation of knowledge (under reasonable assumptions) implies an equal contribution to the production of waste, possibly even misinformation. Moreover, since the constructor is a kinematic machine, it obeys the combinatorial laws of chemical kinetics. And these imply that the speed of reactions is governed by the unique interactions of elements with constructors, the free energy available for transformation, the population size of the inputs, the time delay during the construction process, and so forth. All of the necessary ingredients for a scientific theory of knowledge production.
The historian Fernand Braudel, in the third volume of his Civilization and Capitalism: The Perspective of the World, writes of the “rhythms of conjuncture,” meaning the variable rates of synchronous events throughout history required to transform the cultural world, and that “human life was subject to fluctuations and swings of periodic movements, which carry on in endless succession.” And Jürgen Osterhammel, in his equally titanic The Transformation of the World, suggests that it was in the nineteenth century that what he calls “asymmetrical efficiency growth” accelerated, fueled by the increased mechanical productivity of labor, new sources of energy from new territories, and sequelae of conflict. The tempo of science is a microcosm of the rhythms of history, and as such, the outcome of intelligible construction processes realized at the global scale.
— David Krakauer
President, Santa Fe Institute
From the Winter 2023–2024 edition of the SFI Parallax newsletter. Subscribe here for the monthly email version, or email “news at santafe.edu” to request quarterly home delivery in print.
