Scientific discourse is often accompanied by slogans that are both memorable and misleading: “truth does not exist,” “all models are wrong,” or “scientific facts are socially constructed.” Each of these expressions captures a legitimate insight. None of them, however, should be accepted without qualification.

This article examines the philosophical confusions that arise when such slogans are taken literally or extended beyond their proper scope. It defends a fallibilist and realist conception of science: scientific knowledge is always open to revision, models are selective and imperfect representations, and the production of knowledge is inseparable from social institutions and collective practices. Yet none of this implies that truth is inaccessible, that all models are simply false, or that scientific facts are reducible to social constructions.

Does scientific truth exist?

Science rarely provides absolute certainty. Its conclusions are revisable, and even well-established theories may eventually be corrected, restricted, or incorporated into more general frameworks.

But fallibility does not imply that scientific claims cannot be true. Many questions admit answers that are strongly constrained by evidence: blood circulates through the body; species evolve over generations; humans and chimpanzees share common ancestors; the universe is expanding.

A residual possibility of error does not place such conclusions on the same epistemic level as unsupported alternatives. The legitimate insight behind the slogan “truth does not exist” is therefore better expressed as follows:

In science, absolute certainty does not exist.

Are all models wrong?

Scientific models simplify, idealise, and selectively represent reality. No model reproduces every feature of the system it represents. In this limited sense, George Box’s famous claim that “all models are wrong, but some are useful” expresses an important methodological point.

The slogan becomes misleading, however, when wrong is treated as synonymous with false, and when the imperfect character of models is used to deny their capacity to provide genuine knowledge.

There is no universal agreement in philosophy of science about whether models themselves are the kinds of entities that can literally be true or false. Some philosophers reserve truth and falsity primarily for propositions, treating models as representational structures through which claims are formulated. Others describe models themselves as approximately true, partially true, empirically adequate, accurate, robust, or fit for a particular purpose.

Whatever position one adopts, the connection between models and truth does not disappear. Models are constructed from assumptions, principles, equations, initial conditions, parameter values, and claims about how their elements correspond to a target system. These components may be strictly true, approximately true, accurate within a stated margin of error, or deliberately false idealisations. Models also produce conclusions and predictions that can themselves be assessed as true, false, or approximately true.

Some assumptions are knowingly false but harmless for the question under investigation. The ideal-gas model, for example, treats molecules as occupying negligible volume and as exerting no forces on one another. These assumptions are not strictly true, but under suitable conditions they do not prevent the model from yielding highly accurate relations between pressure, volume, and temperature.

Other assumptions are approximately true, or sufficiently accurate for a specific purpose. A model of planetary motion may treat the Sun as stationary or neglect the gravitational influence of very distant bodies. These claims are not exact, but their deviations can be quantified and shown to fall below the precision required for the calculation. An apparently approximate claim can often be reformulated as a strictly true one by explicitly specifying its domain of validity or its margin of error.

Still other propositions incorporated into models are very probably strictly true. A model of the Solar System may assume that the Earth has mass, that it exerts gravitational effects on other bodies, or that its orbit is not perfectly circular. These are not merely convenient fictions, but well-supported claims about the system being represented.

A model may therefore be inaccurate in one respect and accurate in another, reliable within one parameter range and inapplicable outside it. Its scientific value depends on what it represents, with what degree of accuracy, within which domain, and for which explanatory or predictive purpose.

The legitimate insight behind “all models are wrong” can therefore be reformulated more carefully:

All scientific models simplify and idealise. Their value depends on which features of reality they represent accurately, within which domain, and for which purpose.

Are scientific facts socially constructed?

Scientific knowledge is undeniably produced through social processes. Research depends on institutions, laboratories, instruments, funding structures, disciplinary norms, peer review, criticism, replication, and the circulation of information within scientific communities.

Scientific classifications and representations also involve human decisions. Researchers choose which phenomena to investigate, which variables to measure, which distinctions to introduce, and which standards of evidence to adopt.

But it does not follow that the realities investigated by science are themselves created by these practices. The social construction of a category, a measurement procedure, or a scientific representation must be distinguished from the causal production of the phenomenon being studied.

The classification of a virus, for example, depends on human concepts and institutional practices. The virus itself, its molecular structure, and its capacity to infect organisms do not depend on our decision to classify it. Similarly, the definition of a unit of measurement is conventional, but the relations measured using that unit are not therefore arbitrary.

The social character of science can help explain both its successes and its failures. Collective criticism, specialised expertise, replication, and institutionalised disagreement make scientific inquiry more reliable than isolated individual judgement. Conversely, financial interests, prestige, political pressures, and disciplinary conformity can introduce distortions.

The fact that scientific knowledge has social conditions of production therefore does not imply that its content is reducible to those conditions. A more defensible formulation is:

Scientific knowledge is socially produced, but the realities it investigates are not thereby socially constructed.

A fallibilist scientific realism

The position defended in this article combines realism with fallibilism.

Scientific theories and models do not provide exhaustive, infallible copies of reality. They are selective, historically situated, and open to revision. Scientific progress may involve conceptual transformations, changes of representation, and the abandonment of previous assumptions.

Nevertheless, mature sciences have produced a vast body of reliable knowledge about observable facts, historical processes, causal relations, and structures that exist independently of our descriptions. The replacement of one theory by another does not generally erase every achievement of the earlier theory. Successful theories are often recovered as approximations within the domains in which they worked.

Fallibilism should therefore encourage epistemic modesty, not indiscriminate scepticism. Recognising the possibility of error does not require treating all positions as equally plausible. Nor does acknowledging the role of models and social institutions require abandoning the concepts of truth, objectivity, or reality.

The complete article develops these arguments in greater detail through examples drawn from physics, biology, cartography, and the history and sociology of science.

[Read or download the complete article in PDF]