Provisional Models

For over a thousand years, Western astronomy revolved around a simple concept: Earth sits still at the center and planets move in circles (epicycles) around it[1]. This system was set down by the astronomer Claudius Ptolemy in the second century, and his model became the definitive account of the heavens for the next 1,400 years.

Of course, the problem with this theory is that planets don't actually move in neat circles around earth so naturally, predictions drifted from observations. However, when reality didn’t follow the models, astronomers didn't rethink the setup. They just added geometric machinery: extra circles, offsets, new reference points. The details don't really matter, what matters is that it worked: the patched model predicted planetary positions accurately enough to run calendars and navigation for over a thousand years. Every time the model didn't predict, a theory was bolted on and the numbers fell back into place.

This worked for a long time until Kepler replaced dozens of hand-tuned circles with a single idea: planets move in ellipses around the Sun. Three clean laws did the work that hundreds of epicycles couldn't, complex calculations collapsed into a single theory every planet obeyed.

Provisional Models

Ptolemy's system is what I call a Provisional Model: not strictly false, but "true" in a limited way. They are able to produce some desirable outcome or prediction, but without correctly capturing internal mechanisms accurately. I like to think of it is being 'decent enough' for real world utility, in spite of its inaccurate modelling.

However, I think there is danger in treating these frameworks as definitive. The moment a provisional model stops being held as provisional and starts being treated as the truth, it quietly changes what we do with evidence. Each contradiction to the existing framework is met with another epicycle, a new caveat or a special case until the framework becomes less a description of reality and more a machine for defending itself.

In hindsight, provisional models of reality are so obvious its hard to image how people took them seriously as fundamental truths. Indeed, the idea that the Earth is the centre of the universe is laughable retrospectively. But, astronomers took these archaic models as fundamental laws of the universe for thousands of years because they guided millions of ships across the globe safely for thousands of years.

The question that we should be asking ourselves is not how could they have been so wrong[2], but what models are we building now that predictively correct, but mechanistically wrong?

Every era and field has its Ptolemy Model, frameworks that are so embedded, and skilled at absorbing contradictions, that the possibility of the model being wrong is difficult to register.

Possible Provisional Models

Some concepts I personally believe to be provisional models:

“Structuralism” in medicine: The focus on physical abnormalities in musculoskeletal medicine (slipped disc, “wear and tear”, tendonitis) as root causes of pain.

  1. An uneven pelvis is a poor predictor of back pain
  2. Does the shape of your spine explain neck pain?
  3. Knee “misalignment” cannot reliably explain knee pain
  4. Why posture and “bad alignment” aren’t big causes of pain

Serotonin as the main mechanism for antidepressants: the idea that depression is caused by a “chemical imbalance” (low serotonin) that antidepressants correct.

  1. No convincing evidence that depression is caused by low serotonin
  2. How the “chemical imbalance” idea spread despite thin evidence
  3. SSRIs raise serotonin within hours, but relief takes weeks — the mechanism puzzle

Better Models

It’s important to note that I am not making a claim that antidepressants do not work, nor that physiotherapy is a complete sham. The main idea I am espousing is being able to predict or fix something is not the same as understanding what actually causes it.

We should strive for better models. More mechanistically accurate models that mirror reality allows us to make more accurate decisions.

Beyond predictive accuracy, perhaps the most important part is not even the predictive accuracy but the intuition that it lends us when expanding knowledge. A provisional model only works within the exact conditions it was built for—when it faces an anomaly, it simply adds another epicycle.

A mechanistic model, because it captures why things happen, gives us the intuition to ask better questions and navigate new problems.

[1] Slightly more complicated with deferrants
The deferent is the big circle — a large loop centered (roughly) on Earth. Nothing physical sits on the deferent itself; it's just a track.
The epicycle is a small circle whose center travels along that big track. The planet sits on this small circle.

[2] Other examples I wanted to include are Humoral theory, blood letting