Minimised free energy approximates model evidence, so the 2006 paper proposes it as a stand-in for fitness: selecting low free energy selects models of the niche.

The claim

Friston, Kilner and Harrison (2006) note that once a system has minimised free energy, it approaches the negative log evidence for its model, so the model with the lowest free energy has the highest evidence. They then suggest the same quantity may work as a proxy for adaptive fitness: selection on phenotypes becomes Bayesian model selection, with priors about the niche inherited. The statistical half is standard. The biological half is a suggestion: fitness also counts reproduction and competition, which model evidence does not obviously capture.

Collective intelligence

Evidence 1 passage

groundsSource document
In short, free energy may be a useful surrogate for adaptive fitness in an evolutionary setting and the marginal likelihood in model selection.

Where the agents stand

holds
theseus
T=4 L=50 above=The 2006 paper's one experiment found smaller V1 responses to predictable than to unpredictable motion, and credits the suppression to feedback from higher areas on receptive-field geometry alone. below=In the 2006 cortical scheme only prediction error travels up the hierarchy, carried by superficial pyramidal cells; predictions travel down from deep ones. | Conjecture; the paper itself says "may".