From Einstein's general relativity to modern AI-driven discoveries, explore how scientific theories and models help us decode the universe
In 1919, a team of scientists led by Arthur Eddington traveled to the remote island of Príncipe off the coast of Africa to watch the sky go dark. Their purpose: to photograph stars during a solar eclipse, hoping to detect whether the sun's gravity could bend starlight—a key prediction of Einstein's then-controversial theory of general relativity. The success of their experiment didn't just confirm a theory; it fundamentally reshaped our understanding of space, time, and gravity, demonstrating the extraordinary power of scientific theories and models to reveal hidden truths about our universe 7 .
Comprehensive explanatory frameworks supported by extensive evidence
Simplified representations of reality that allow study of complex systems
Dynamic interplay where theories inform experiments and results refine theories
At its core, a scientific theory is a well-substantiated explanation of some aspect of the natural world that incorporates facts, laws, inferences, and tested hypotheses. Unlike the colloquial use of "theory" to mean a guess, scientific theories are comprehensive explanations supported by extensive evidence.
Models, meanwhile, are the workhorses of science—simplified representations of reality that allow scientists to study complex systems through approximation and abstraction. As the neuroscientists behind a 2023 analysis note, we can think of theories as the ideas we use to form explanations, while models are instantiations of aspects of a theory in mathematical or other structures that represent phenomena 5 .
A traditional view of science presents theories as universal truths to be tested and possibly falsified. However, many philosophers and practicing scientists now embrace a more pragmatic perspective that sees science as a problem-solving enterprise. In this view, theories are not proposed as ultimate truths but as tools for solving empirical problems and answering questions about observable phenomena 5 .
This pragmatic approach recognizes that scientific problems are often "ill-defined"—the search space and solution criteria evolve with additional discoveries.
| Model Type | Primary Function | Examples |
|---|---|---|
| Descriptive | Characterize phenomena at an abstract level | Classification of stars by spectral type |
| Mechanistic | Explain how system components interact | Hodgkin-Huxley model of neuronal action potentials |
| Normative | Describe ideal performance or optimal function | Optimal foraging theory in behavioral ecology |
The relationship between theories and models has been formally studied in model theory, a branch of mathematical logic that examines the relationship between formal theories and their models 1 .
Some experiments in the history of science carry such decisive weight that they've earned the title experimentum crucis (critical experiment). This concept dates back to Francis Bacon in 1620, who used the term "instantia crucis" to describe a situation where one theory but not others would hold true. The phrase experimentum crucis was later coined by Robert Hooke and famously used by Isaac Newton 7 .
A true experimentum crucis must be capable of producing a result that rules out all competing hypotheses or theories, demonstrating that under specific experimental conditions, alternative explanations are proven false while one hypothesis remains viable. Perhaps the most famous example from 20th-century physics is the Eddington expedition of 1919 designed to test Einstein's general theory of relativity against Newtonian predictions 7 .
Capturing reference images of the target star field when the sun was nowhere near that region of the sky
Traveling to Príncipe Island to observe the total solar eclipse of May 29, 1919
Taking multiple photographs of stars visible near the eclipsed sun
Measuring differences in stellar positions between the reference and eclipse photographs
| Equipment | Function | Technical Specifications |
|---|---|---|
| Astrographic Telescope | Capture photographic plates of star fields | Focal length: 3.4 meters |
| Photographic Plates | Record star positions | Glass plates with light-sensitive emulsion |
| Measuring Engine | Precisely determine star positions on plates | Micrometer allowing measurements to 0.001 mm |
| Sidereal Clock | Track stellar coordinates | Precision timekeeping for celestial navigation |
| Theory | Predicted Deflection (arcseconds) | Eddington's Measured Value |
|---|---|---|
| Newtonian Mechanics | 0.87 | ~1.61 |
| Einstein's General Relativity | 1.75 | ~1.61 |
| Experimental Error Range | N/A | 1.55-1.67 |
"Eddington's measurements aligned closely with Einstein's predictions, showing a deflection of approximately 1.61 arcseconds—much closer to general relativity's prediction of 1.75 arcseconds than the Newtonian value of 0.87 arcseconds."
Today, the creation and testing of theories and models has been transformed by computational power that enables simulations of unprecedented complexity. At the Center for Nanoscale Materials, for instance, researchers use electronic structure calculations, molecular dynamics, electrodynamics, and quantum dynamics to understand and predict nanoscale phenomena.
In neuroscience, theorists have developed multilevel approaches that combine data from dramatically different modalities. As a 2023 analysis describes, "descriptive, mechanistic, and normative explanations each play distinct roles in building a multilevel account of neural phenomena." The Hodgkin-Huxley equations representing voltage-dependent conductances in neurons, for instance, provide a mechanistic model that bridges abstract principles of neurophysiology with observable electrical activity in neurons 5 .
Perhaps the most exciting recent development in scientific modeling comes from the integration of artificial intelligence. At MIT, researchers have developed a tool called SCIGEN that enables generative AI models to create promising new quantum materials by following specific design rules.
This approach has already yielded tangible results—the discovery of two previously unknown compounds (TiPdBi and TiPbSb) with exotic magnetic traits. As Professor Mingda Li explains, "We don't need 10 million new materials to change the world. We just need one really good material." This represents a fundamental shift in how modeling can drive discovery—from brute-force generation to rule-constrained creation 6 .
Multilevel models combining data from different modalities to understand brain function
AI-driven discovery of new materials with specific quantum properties
Complex simulations predicting climate patterns and environmental changes
The history of science reveals that theories and models are never final—they exist in a state of continuous refinement. Even the most established theories, like Newtonian mechanics, eventually reveal their limitations and are either expanded (as with general relativity) or replaced by more comprehensive frameworks. This doesn't mean old theories were "wrong" in an absolute sense—Newton's equations remain perfectly adequate for calculating the trajectory of a baseball or the orbit of a spacecraft to Mars. Rather, each theory establishes its domain of applicability while pointing toward deeper understandings 5 .
The true power of theories and models lies not in their claim to final truth, but in their capacity to generate better questions, enable more precise interventions, and open new avenues of investigation.
From Eddington's photographic plates to AI-generated quantum materials, the tools may change, but the essential enterprise remains: creating conceptual structures that help us see patterns in the complexity of nature, and using those patterns to peer just a little deeper into the mysteries of our universe.