1. Useful fictions
Scientific models omit friction, treat populations as continuous variables, and represent actual people as ideal agents. If truth meant copying reality item by item, almost no model would qualify. Yet models predict, compress data, and expose relations among variables.
An instrumentalist may treat models as devices for organizing experience without committing to unobservable entities. But where a model supports intervention, connects independent measurements, and explains its own failure conditions, it becomes harder to dismiss its success as merely accidental.
2. From prediction to mechanism
Predictive success can come from curve-fitting. Mechanistic explanation asks why a system changes as it does. Black-box systems may predict accurately without explaining, while simplified mechanism models may predict less precisely but support counterfactual questions.
3. Convergent realism
A restrained realism does not project every successful model structure onto the world. It asks whether experiments, measurements, interventions, and models converge on the same mechanism. Realism is a cumulative judgment across practices, not a certificate granted to one model.
References
- Cartwright, Nancy, 1983, How the Laws of Physics Lie.
- Giere, Ronald, 2004, “How Models Are Used to Represent Reality.”
- Wimsatt, William, 2007, Re-Engineering Philosophy for Limited Beings.