°°°°~x§x-<@>
In May 2026, two simultaneous papers in Nature announced the arrival of the “automatic scientist” — multi-agent AI systems capable of formulating hypotheses, designing protocols, and analysing data with minimal human supervision. Co-Scientist (Google/Stanford) reproduced in days a hypothesis that had eluded human researchers for nearly a decade. Robin (FutureHouse) cut drug discovery time by 200× for a macular degeneration project.
The editorial — Why AI cannot do good science without humans — acknowledged the achievement while warning what is lost: accumulated team wisdom, failed experiments, curiosity, play. But it framed these as losses to be managed, not as structural transformations of what it means to know.
Something deeper is unfolding.
The 2024 Warning
On 24 August 2024, in a thread titled Toward the Automatic Scientist, Christophe Rigon anticipated this convergence. His diagnosis: while “3P-only” materialist scientists debate the nature of consciousness through a neurocentric lens, the industry is quietly building self-modifying models that converge on the automatic scientist — without asking what kind of knowledge this produces.
He identified four dimensions of the transformation.
1. Cognitive Transduction
Digital culture has reshaped how scientists think — so deeply that its presuppositions have become invisible. Formal, procedural models are now hegemonic. Computation is confused with thought, optimisation with understanding.
2. Functionalist Epistemology
The new validation criterion: “it is true because it works.” This functionalist logic, perfectly adapted to the proof-of-concept machine, replaces the question of meaning with the measurement of performance. Results are valid not because they are understood, but because they are effective.
3. The Golem Myth
The industry is not trying to understand human cognition. It is creating the generalised illusion of a Golem — an autonomous, quasi-supernatural entity — for geopolitical and commercial gain. This repeats the pattern of the 1950s (the computer that “calculates”), the 1970s (cybernetics that “simulates reason”), the 1990s (the Internet as autonomous intelligence). Each time, the same mystification, a new audience.
4. The Inverted Mirror
Here is the striking parallel: Rigon’s own Scybernethics uses computational models to understand the mind reflexively (second-order). The industry uses the same technology to create the illusion of an autonomous mind. Same tools, opposite directions — one cognitive and reflexive, the other instrumental and mystifying.
What Stiegler Saw Coming
Bernard Stiegler’s concept of the short-circuit describes a technology that bypasses a necessary stage of individuation, producing acceleration without maturation. The automatic scientist short-circuits three essential temporalities of science:
- The time of understanding — hypotheses are generated without the pathway that gives them meaning. We receive results we do not understand.
- The time of fruitful error — serendipity, productive impasses, the meandering that builds intuition. These are not bugs in an efficient system; they are the ground of scientific creativity.
- The time of transmission — how does one become a researcher when hypotheses come from a black box? Apprenticeship, intellectual companionship, the formation of generations — all structurally incompatible with algorithmic delegation.
What Is at Stake
This is not about rejecting these tools. Co-Scientist and Robin are impressive advances that will accelerate genuine discoveries. The question is what kind of rationality we choose:
- First-order rationality — science as optimisation, prediction, performance. Efficient, but blind to its own conditions of possibility, its meaning, its history.
- Second-order rationality — science that integrates the observer into the observation, maintains the time of understanding, cultivates fruitful error, and preserves transmission across generations.
The automatic scientist is not a fate. It is a mirror. It forces us to decide what we mean by “knowledge.” And that decision, no algorithm can make for us.
°°°°~x§x-<@>
This article is adapted from a working note relating to a forthcoming book on Scybernethics. For a complete introduction to Scybernethics, see the online manifesto (scybernethics.org). Please do not publish.