An Intelligent Groove

Fixation’s slow but the funk’s begun

Most of what I write on the subjects of science and philosophy are heavily augmented by AI, but to describe them as AI-written is to miss the point entirely. AI writing is for things like MIDNIGHT’S WAR, in which one lays out the characters and the outline, then describes what is going to happen in a chapter and the AI fills in all the blanks, after which one edits the text.

That doesn’t work at all for science. AI-augmented writing is where the AI serves primarily as a tireless team of researchers and a collection of oppositional AIs serve as a critical Red Team, stress-testing every result and paper produced in a back-and-forth process that is designed to systematically identify and remove every weakness, flaw, and error from the final product. Which, of course, looks absolutely nothing like the single-AI product of AI-writing.

This is more what it looks like:

The neo-Darwinian synthesis has never been subjected to a comprehensive empirical audit in which every proposed evolutionary mechanism operates simultaneously under observation. Defenders of the theory have constructed what we term the post-Darwinian polythesis: a shifting constellation of supplementary mechanisms — neutral drift, parallel fixation, supermutation, hitchhiking, epistasis, clonal interference, compensatory mutation — each invoked ad hoc to answer a specific mathematical objection, but never tested collectively. The Long-Term Evolution Experiment (LTEE) with Escherichia coli, now in its fourth decade, is the only experiment in existence where every one of these mechanisms operates simultaneously under controlled conditions. We analyze the total fixation throughput of the LTEE using two independent public datasets: Tenaillon et al. 2016 (clone-pair sequencing through 50,000 generations, 264 genomes) and Good et al. 2017 (metagenomic frequency trajectories through 60,000 generations). The result for non-mutator populations is 1,322 generations per fixation at 60,000 generations — cross-validated at 893 generations per fixation by clone-pair analysis at 50,000 generations. Mutator populations with 100-fold elevated mutation rates achieve only an 8.5- to 17-fold increase in fixation throughput depending on methodology, not 100-fold, because fixation is bottlenecked by sweep dynamics rather than mutation supply. Applied to human-chimpanzee divergence (205 million required fixations on the human lineage across 252,000 available generations), the shortfall is 1,075,000-fold for non-mutators and 104,873-fold across all twelve populations including hypermutators. Three independent lines of genomic evidence confirm that no supermutational phase occurred on the human lineage. The post-Darwinian polythesis fails comprehensively under the most favorable experimental conditions ever constructed.

That was as of one week ago. The more accurate number, as it turns out when you do what the scientists didn’t do and go to the high-resolution data set for all 273,000 generations across the 12 populations, then run the numbers, is 1,587 generations per fixation.

But most people, including most scientists, are neither psychologically nor intellectually capable of meaningfully processing information presented that way, so there is a third way to utilize AI in the transmission of information, which can be usefully combined with a fourth way. And that’s how the most recent Vibe Patrol music video on the subject of the average rates of genetic fixation came to pass. Which means you can safely disregard all of the above, get down with the Vibe Patrol gang instead, and still end up in the right place.

Congratulations. You now know more about the actual state of evolutionary science than pretty much every single evolutionary biologist and population geneticist on the planet. Please collect your complimentary PhD in Evolutionary Biology on the way out.