Science Explains a Great Deal, But Not Everything

Science wins by narrowing the question

At several points in modern history, educated people have declared a problem permanently beyond science. The history of those declarations is not flattering. Vitalists insisted that life depended on a special force absent from chemistry; Friedrich Wöhler’s synthesis of urea in 1828 did not settle the matter on its own, but it helped puncture the confidence. In the 20th century, many biologists treated protein folding as a grand intractable puzzle: given an amino-acid sequence, how could one predict the final three-dimensional structure among astronomically many possibilities? By 2020, DeepMind’s AlphaFold system had turned a 50-year problem into something much more manageable for large classes of proteins, with performance at CASP14 that startled structural biologists. The science was not finished, but a frontier had moved.

What changed was not magic and not mere data accumulation. It was the disciplined habit of turning vague astonishment into a tractable question. Science does not answer everything at once; it slices. It asks what variables matter, what can be measured, what regularities hold, what models survive contact with observation. Much of its power comes from this austerity. Newton did not explain why there is a universe or what a force is in some ultimate metaphysical sense. He gave equations that linked masses, distances and motion with unnerving accuracy. Darwin did not settle every puzzle about heredity, development or consciousness. He showed how natural selection could explain adaptation without design.

That record matters because sweeping claims about the limits of science are often premature. Some things look mysterious because instruments are crude, data are thin or the mathematics is missing. A century ago, continental drift seemed speculative. Before 1965, the cosmic microwave background had not been measured. Before Rosalind Franklin’s X-ray diffraction images and the 1953 Watson-Crick model, heredity had no molecular basis. There is no wisdom in treating the current edge of knowledge as a permanent wall.

Some limits are practical and still real

Yet not every scientific limit is profound. Some are limits of scale, noise, computation and control. They matter because they constrain what can be known in practice even when the underlying laws are well understood.

Weather forecasting is the standard case. The atmosphere obeys physical laws, but it is a chaotic system: tiny differences in initial conditions grow rapidly. Edward Lorenz showed in the 1960s that deterministic equations could still produce forecast divergence. Operational meteorology has improved enormously through better satellites, denser observations and larger numerical models, but there remains a rough ceiling of about 10 days for detailed weather prediction at useful resolution. The exact horizon depends on region, variable and season, and ensemble methods can extend probabilistic skill further, but no amount of cleverness abolishes the basic problem. The Lyapunov time, which characterises how fast nearby trajectories separate, sets a limit on the value of increasingly precise initial data.

This is not ignorance in the old sense. It is knowledge of a limit. One can explain why a forecast fails after a certain interval and still be unable to rescue the forecast. The same distinction appears elsewhere. Turbulence in fluids, earthquake timing on specific faults, and many-body problems in condensed matter can be governed by equations that are known while their exact solutions remain computationally or practically out of reach.

Protein folding offers the opposite lesson. For decades the problem looked impossible because the search space was too vast and the relevant constraints too subtle. AlphaFold did not prove that all proteins are now explained, nor did it replace experiment. It performs less well on some complexes, disordered proteins and dynamic conformational changes. But it showed that an apparently forbidding problem could yield once data, computation and model design reached the right level. Practical limits can be severe; they are also contingent. Confusing them with absolute barriers is a recurrent mistake.

Other limits are built into logic and physics

A harder class of limits does not merely reflect current weakness. It arises from the architecture of formal systems and from the structure of the universe itself.

Gödel’s incompleteness theorems, published in 1931, showed that any sufficiently powerful formal system capable of elementary arithmetic contains true statements that cannot be proved within that system, assuming the system is consistent. This result is often abused. It does not show that science is futile, nor that human minds float above logic. It does show that there is no universal formal machine that can derive every mathematical truth from a fixed finite set of axioms. Since mathematics is the language of much of theoretical science, this matters. It places a principled limit on what can be captured in any one formal framework.

Chaos adds a different kind of boundary. Even where the laws are deterministic, long-range prediction can fail because errors in initial conditions amplify exponentially. The weather’s roughly 10-day ceiling is a practical expression of this deeper fact. Planetary motion can be stable over vast periods, yet still exhibit chaotic zones. One can know the equations and still lose the future.

Quantum mechanics introduces another in-principle difficulty. The theory is the most accurate framework in physics, but what exactly it says about reality remains disputed. The measurement problem is the live nerve. The Schrödinger equation describes smooth, deterministic evolution of the wavefunction, yet measurements yield definite outcomes. How, exactly, does one get from superposition to a single result? Copenhagen-style views, many-worlds interpretations, Bohmian mechanics and objective-collapse theories offer different answers, and no consensus interpretation has been forced by experiment. Science has here produced astonishing predictive success without a settled explanatory picture of what the formalism means.

Then there are cosmic horizons. Because the universe has a finite age and space is expanding, there are regions from which light has not had time to reach us, and perhaps never will. The observable universe is not the universe full stop. Cosmologists can infer earlier conditions from the cosmic microwave background and other relics, but there are hard limits to observation. Inflation, if it occurred, may have erased information about still earlier states. Questions about what lay beyond the observable patch, or before a putative inflationary phase, may be constrained not only by ingenuity but by the causal structure of spacetime.

Some questions are not scientific in form

Another source of confusion is category error. Science is superb at descriptive and causal questions. It can tell you what stars are made of, when species diverged, how synapses change with learning, what atmospheric carbon dioxide does to radiative balance. It is not designed to settle every normative or aesthetic claim, because those claims are not simply about facts in the same way.

Take ethics. Science can illuminate the consequences of actions, the evolutionary roots of cooperation, the psychology of moral judgement and the effects of policies on health or welfare. It cannot, by experiment alone, derive that one ought to value equality over liberty, or future generations over present consumers, or Beethoven over drill. David Hume’s old point about the gap between ‘is’ and ‘ought’ still bites. Facts constrain values; they do not by themselves generate them.

Aesthetics is similar. Neuroscience can study musical expectation, visual processing and reward circuits. Sociology can trace how taste tracks class, fashion and institutions. None of that settles whether Turner is better than a tourist sunset or whether a late quartet matters more than a catchy chorus. Those are argumentative judgements embedded in traditions, criteria and human ends. One may try to naturalise them; one has not thereby turned them into laboratory findings.

This is not a defeat for science. It is a reminder that explanation is plural. Courts, parliaments, critics and historians answer questions that are disciplined, rational and evidence-sensitive without thereby becoming branches of physics. If someone asks whether a punishment is just or whether a poem is beautiful, the request is not for another decimal place in a spectrometer readout.

Science limits itself to gain reliability

The most revealing limits are self-imposed. Science works by public methods: measurement, replication, quantification where possible, explicit models, and claims exposed to refutation. Those rules exclude some sources of conviction not because they are always false but because they are not reliably shareable. A private revelation, a mystical certainty or a single unrepeatable experience may be existentially decisive for the person who has it. It is not, on its own, scientific evidence.

That self-limitation is a strength. It explains why science can converge across cultures and centuries. The kilogram is no longer tied to a metal cylinder in Sèvres but defined via physical constants because metrology prefers stable, public standards. Clinical trials use controls and statistics because intuition is a poor guide to efficacy. Astronomers build instruments on Mauna Kea and in Chile because careful observation beats armchair cosmology.

But method also narrows the target. Science generally brackets first-person meaning in order to secure third-person agreement. It can correlate reports of pain with C-fibre firing, map the neural signatures of decision-making, and model perceptual binding. Whether such work exhausts consciousness as lived remains contested, which is why debates in philosophy of mind persist alongside neuroscience. The point is not that consciousness is supernatural. It is that the method prefers what can be standardised, and some features of experience resist easy standardisation.

So, can science explain everything? If ‘everything’ means every physical process accessible to observation and modelling, science is plainly our best explanatory engine and still expanding. If it means every truth, every value, every first-person significance, every mathematically expressible fact, and every fact about regions forever outside our causal horizon, the answer is no. The interesting part is not the refusal. It is the map of the boundaries. Some will retreat as tools improve. Some are fixed by logic or by the universe’s design. Some mark the difference between facts and judgements. And some are the price science pays for being the most reliable way yet devised of finding out how the world works.