Realism vs. Instrumentalism in Prediction: Do Convergent Systems Describe Reality or Just Work?

BY NICOLE LAU

When eight prediction systems converge on "Biden wins 2020 election," what does this mean? Does it mean Biden really will win (realism)? Or just that these tools are useful for making decisions, regardless of underlying reality (instrumentalism)? This ancient debate—do theories describe reality or merely serve as useful instruments?—takes on new urgency in prediction.

This article explores realism vs. instrumentalism in the context of convergent prediction, examining what convergence tells us about truth, reality, and the nature of knowledge.

The Realism-Instrumentalism Debate

Scientific Realism

Core claim: Scientific theories aim to describe reality as it actually is. When theories are successful, they're (approximately) true.

Prediction version: Convergent predictions describe real future states. High CI means multiple methods are tracking the same reality.

Example: Weather models converge on hurricane path → Hurricane will actually follow that path (models describe real atmospheric dynamics)

Instrumentalism

Core claim: Scientific theories are tools for prediction and control, not descriptions of reality. Success means utility, not truth.

Prediction version: Convergent predictions are useful tools for decision-making. High CI means multiple tools work well, not that they describe reality.

Example: Weather models converge → Use this for evacuation planning (models are calculation devices, agnostic about "reality")

Key Arguments for Realism

1. No-Miracles Argument (Putnam)

Argument: The success of science would be a miracle if scientific theories weren't at least approximately true.

Why: If theories were completely false, why would they make accurate predictions?

Prediction version: Convergence of independent systems would be miraculous coincidence if they weren't tracking reality.

Example: 8 independent election models converge on Biden win. If models weren't describing real voter preferences, economic conditions, etc., convergence would be incredible luck.

2. Entity Realism (Hacking)

Argument: If we can manipulate something, it's real. "If you can spray them, they're real."

Example: We manipulate electrons in experiments → Electrons are real (not just theoretical constructs)

Prediction version: If predictions enable successful intervention, they describe reality.

Example: Hurricane prediction enables successful evacuation → Prediction describes real hurricane path

3. Convergent Realism

Argument: When independent methods converge, they're likely tracking the same reality (not coincidence).

Why: Independent methods have different assumptions, biases. Convergence despite differences indicates shared reality.

Prediction version: High CI indicates multiple systems tracking same future reality.

Key Arguments for Instrumentalism

1. Underdetermination

Argument: Multiple incompatible theories can fit the same data. Can't determine which is "true."

Example: Ptolemaic (geocentric) vs Copernican (heliocentric) both fit astronomical data initially.

Prediction version: Multiple prediction models can converge on same outcome for different reasons. Can't say which describes "reality."

Instrumentalist conclusion: Focus on utility (which model is most useful), not truth (which is "real").

2. Pessimistic Meta-Induction (Laudan)

Argument: History of science shows past theories were false (phlogiston, caloric, ether). Current theories will likely be false too.

Example: Newtonian physics was "successful" for 200 years, then replaced by Einstein. Newton wasn't "true."

Prediction version: Today's prediction models will be replaced by better models. Don't claim they describe "reality."

Instrumentalist conclusion: Theories are useful tools, not true descriptions. Use them, don't believe them.

3. Constructive Empiricism (van Fraassen)

Argument: Science aims for empirical adequacy (saving the phenomena), not truth about unobservables.

Example: Atomic theory is empirically adequate (predicts observations), but we can be agnostic about whether atoms "really exist."

Prediction version: Prediction models aim to fit observable data, not describe unobservable "reality."

Middle Ground: Structural Realism

Worrall's Structural Realism

Claim: Mathematical structure of theories is preserved across theory change, even if entities change.

Example: Fresnel's wave theory of light → Maxwell's electromagnetic theory. Entities changed (ether → fields), but mathematical structure (wave equations) preserved.

Prediction version: Convergence indicates shared mathematical structure, agnostic about underlying entities.

Example: Election models converge on "Biden 52%, Trump 48%" (mathematical structure), agnostic about whether "voter preferences" are "real" entities.

Why Structural Realism Fits Convergence

Convergence preserves structure:

• Different models (polls, markets, experts) have different ontologies (entities, mechanisms)
• But they converge on same mathematical predictions (probabilities, trends)
• Convergence indicates shared structure, not necessarily shared entities

Convergence Through Realist and Instrumentalist Lenses

Realist Interpretation of High CI

CI = 0.85: 85% of independent methods agree → Strong evidence they're tracking same reality

Why: If methods were just useful fictions, convergence would be coincidence

Implication: Prediction likely describes real future state

Instrumentalist Interpretation of High CI

CI = 0.85: 85% of useful tools give same answer → High confidence in utility for decision-making

Why: Multiple tools working well is pragmatic warrant, not metaphysical claim

Implication: Prediction is reliable tool, agnostic about "reality"

Structural Realist Interpretation

CI = 0.85: 85% of methods converge on same mathematical structure (probabilities, trends)

Why: Structure is preserved even if underlying entities/mechanisms differ

Implication: Prediction captures real structure, agnostic about entities

Practical Implications

Weather Forecasting

Realist: Models describe actual atmospheric dynamics (pressure, temperature, humidity). Convergence means models tracking real weather system.

Instrumentalist: Models are calculation tools. Convergence means tools work well for planning (evacuations, flights). Don't need to believe in "real" atmospheric dynamics.

Structural realist: Models capture mathematical structure of weather (differential equations). Agnostic about whether "pressure" is "real" entity.

Economic Prediction

Realist: Models describe real economic mechanisms (supply/demand, rational agents). Convergence indicates models tracking real economy.

Instrumentalist: Models are tools for policy. Convergence means tools useful for decisions. Don't need to believe "rational agents" are real.

Structural realist: Models capture mathematical relationships (elasticities, equilibria). Agnostic about "utility functions" being real.

Medical Diagnosis

Realist: Convergence of tests (blood work, imaging, symptoms) indicates real disease. Tests describe actual biological state.

Instrumentalist: Convergence means treatment will work. Tests are diagnostic tools. Agnostic about "disease" as real entity vs useful category.

Structural realist: Tests capture structural patterns (biomarkers, correlations). Agnostic about disease ontology.

Does It Matter?

Pragmatic Equivalence

For prediction: Realism and instrumentalism often give same practical advice

• Both say: High CI → Rely on prediction
• Both say: Low CI → Remain uncertain

Difference is metaphysical, not practical

When It Matters

1. Theory development:

• Realist: Seek true theories (describe reality)
• Instrumentalist: Seek useful theories (solve problems)
• Can lead to different research priorities

2. Explanation:

• Realist: Convergence explains success (methods track reality)
• Instrumentalist: Convergence is success (methods work)
• Different understanding of "why" predictions succeed

3. Confidence:

• Realist: High CI → High confidence in truth
• Instrumentalist: High CI → High confidence in utility
• Subtle difference in what we're confident about

Historical Case Studies

Phlogiston → Oxygen

Instrumentalist lesson: Phlogiston theory worked (explained combustion, calcination) but was false. Shows theories are tools, not truths.

Realist response: Oxygen theory is closer to truth (progressive). Phlogiston had some truth (something is released in combustion—oxygen is absorbed, but inverse relationship).

Newton → Einstein

Instrumentalist lesson: Newton worked for 200 years, then replaced. Shows even successful theories aren't "true."

Realist response: Einstein is refinement, not replacement. Newtonian physics is limiting case (approximately true at low speeds). Progressive convergence on truth.

Structural realist: F = ma preserved in Einstein (limiting case). Mathematical structure continuous.

Atomic Theory

Realist: Atoms are real (we manipulate them—Hacking). Convergence of chemistry, physics, microscopy confirms reality.

Instrumentalist: Atomic theory is useful tool. Agnostic about whether atoms "really exist" (could be useful fiction).

Structural realist: Atomic theory captures real mathematical structure (quantum mechanics). Agnostic about atoms as "little billiard balls."

Convergence and the Debate

Convergence Favors Realism (Slightly)

Why:

• Convergence of independent methods is unlikely coincidence (no-miracles argument)
• Best explanation: Methods tracking same reality
• Instrumentalism can explain convergence (multiple tools work), but realism explains it better (they work because they track reality)

But Instrumentalism Remains Viable

Why:

• Convergence could indicate shared utility, not shared reality
• Underdetermination: Multiple interpretations of convergence possible
• Pragmatic: For prediction, utility is what matters (truth is bonus)

Conclusion

The realism-instrumentalism debate in prediction:

Realism: Convergent predictions describe reality. High CI indicates multiple methods tracking same future state. Success explained by approximate truth.

Instrumentalism: Convergent predictions are useful tools. High CI indicates multiple tools work well. Success is utility, not truth.

Structural realism (middle ground): Convergence indicates shared mathematical structure. Agnostic about underlying entities.

Key arguments:

• For realism: No-miracles (convergence unlikely if not tracking reality), entity realism (manipulation indicates reality), convergent realism (independent methods converge)
• For instrumentalism: Underdetermination (multiple theories fit data), pessimistic meta-induction (past theories false), constructive empiricism (aim for adequacy not truth)

Practical implications: Often equivalent (both say use high-CI predictions), but differ on explanation, theory development, and what we're confident about.

Historical cases: Phlogiston, Newton, atoms—both sides claim support, debate continues.

Convergence slightly favors realism: Best explanation for independent convergence is shared reality. But instrumentalism remains viable (convergence could be shared utility).

For prediction: Use convergence regardless of metaphysical stance. Realism and instrumentalism agree on practice, differ on interpretation.

Next: Multi-Perspectivism—beyond relativism and absolutism in prediction.

As you navigate the delicate dance between the seen and the unseen, remember that the tools we use to glimpse the future are bridges between worlds, not perfect maps of what is to come. To deepen your own practice of discernment, consider exploring the 40 manifestation rituals intention to reality to anchor your intentions in grounded action, or the 13 new moon rituals lunar beginnings for aligning with the cycles of creation. For a more introspective journey into how your inner world shapes your outer predictions, the tarot journaling prompts 100 questions for self discovery can guide you toward the truths that systems alone cannot reveal.