I’d extend your frame one layer down — from epistemic reliability to systemic energy flow. Scientific unreliability isn’t just noise in the data; it’s the steady-state output of institutions optimizing for exergy throughput under asymmetric incentives.
Once you model pharma, academia, and regulation as dissipative control systems rather than neutral truth engines, patterns like thalidomide, specialty-drug patent-extension combos, or DARPA’s dual-use biomedical programs stop looking like isolated moral lapses and start reading as structural feedback loops. The profit function rewards transient efficacy signals, not long-horizon stability, so the weighting between benefit, chronic side effect, and negative counterfactual is systematically skewed.
In short: the bias is thermodynamic, not merely statistical — energy follows the incentive gradient, even when it flows through Bayesian priors. Even clinical trials select for deception — participants routinely downplay side effects or conceal medical history to stay eligible, turning the evidence base itself into a performative energy market rather than a truth engine. The concept of “evergreening” (making small changes, new formulations, combinations, or regulatory strategies to extend patent life) is well known in pharma criticism literature (though not always legally or scientifically indefensible). The case of thalidomide → Revlimid is one of those.
The tobacco case Emil cites was detectable only because of its massive effect size, ubiquity, and temporal duration; many pharmaceuticals have more modest effect sizes, shorter timeframes, or heterogeneous populations, making distortions from deception or structural bias more influential.
I’d extend your frame one layer down — from epistemic reliability to systemic energy flow. Scientific unreliability isn’t just noise in the data; it’s the steady-state output of institutions optimizing for exergy throughput under asymmetric incentives.
Once you model pharma, academia, and regulation as dissipative control systems rather than neutral truth engines, patterns like thalidomide, specialty-drug patent-extension combos, or DARPA’s dual-use biomedical programs stop looking like isolated moral lapses and start reading as structural feedback loops. The profit function rewards transient efficacy signals, not long-horizon stability, so the weighting between benefit, chronic side effect, and negative counterfactual is systematically skewed.
In short: the bias is thermodynamic, not merely statistical — energy follows the incentive gradient, even when it flows through Bayesian priors. Even clinical trials select for deception — participants routinely downplay side effects or conceal medical history to stay eligible, turning the evidence base itself into a performative energy market rather than a truth engine. The concept of “evergreening” (making small changes, new formulations, combinations, or regulatory strategies to extend patent life) is well known in pharma criticism literature (though not always legally or scientifically indefensible). The case of thalidomide → Revlimid is one of those.
The tobacco case Emil cites was detectable only because of its massive effect size, ubiquity, and temporal duration; many pharmaceuticals have more modest effect sizes, shorter timeframes, or heterogeneous populations, making distortions from deception or structural bias more influential.
https://ijb.utoronto.ca/news/inside-canadas-exploitative-clinical-trial-industry-where-study-participants-say-theyre-incentivized-to-lie-even-about-medications-side-effects/ https://www.astralcodexten.com/p/your-review-participation-in-phase/comments
Normal is that which functions according to it's design.