Theory and Data, Part 2
Episode #3 of the course How science works by Benjamin Keep
Welcome back.
Last time, we talked about one proposed idea that purported to describe how science works and explain why we should “trust” scientific knowledge: Popper’s falsification approach. Ideas in science do seem to go through a crucible of sorts. At least on its face, it seems like good theories rise to the top because of rigorous tests. Prescriptively, the idea that our most cherished scientific theories and statements have withstood the harshest scrutiny is nice. But does it really explain how science works? Three ideas challenge this view.
The Web of Belief
Popper’s view implies that when data conflicts with theory, we throw the theory out. The theory has failed the crucible. And we should trust only theories that survive these crucibles. He introduces the idea of a “crucial” experiment that will vindicate (or trash) an existing theory. The experiments in the early 1900s that examined the speed of light are a good example. Einstein’s theory of relativity says that the speed of light is constant regardless of whether we’re stationary or moving with respect to the light, and ever-more precise experiments measuring the speed of light continued to illustrate that prediction. It seems like a vindication of Einstein’s theory and a vindication of Popper’s philosophical view.
But what happens when theories are falsified? Much is going on with those light experiments. What if the experimentalists had gotten evidence that the speed of light varied? It could be that the theory of relativity was just wholesale incorrect. Or the theory may have been misinterpreted. Or the experimental instruments could have been miscalibrated. Or the measurements could have been contaminated by a previously unknown phenomenon.
The web of belief is the idea that any sufficiently complex experiment involves multiple mutually supportive theories that work toward making the key prediction that the experiment tests. When the experiment goes against the theory, we might question any number of those beliefs.
A great deal of evidence suggests that scientists, in fact, don’t throw out theories when evidence contradicts them. Scientists ignore, reject, and reinterpret data all the time. Often, theories don’t need to be wholesale rejected but slightly modified to accommodate new information. Did we throw out Newtonian mechanics when it failed to precisely predict the tides? No. It’s not that Newtonian mechanics was “wrong,” it’s just that tidal forces are complex.
If falsification says, “When the data conflict with theory, the theory has been falsified,” the web of belief says, “Which theory?”
The Underdetermination of Theory by Data
Suppose you tell me that if I fall off my bike, I’ll scrape my knee. Later on, you see my knee is scraped. Did I fall off my bike? No, not necessarily. Many things might have led to me having a scraped knee. More than one “theory” explains the evidence.
Consider the discovery of Neptune. Newtonian mechanics predicted that Uranus should be in one place in the sky. But it was somewhere else (note that astronomers didn’t throw out Newton’s theory at this point). Astronomers, using Newtonian mechanics, predicted the presence of another planet that was disturbing Uranus’s orbit. They looked at the sky at the predicted place, and there it was: Neptune. A startling prediction and such vindication of Newtonian mechanics! But it only feels startling because Newton’s theory was the only one on the table. Fifty years later, another theory of gravitation—Einstein’s theory of general relativity—could have predicted Neptune’s presence as well. In principle, at least, there is any number of theories that explain existing data equally well.
If more than one theory can always explain the data, then which is supported by “failing to be falsified”?
The Theory-Ladenness of Observations
Falsification and several other accounts of how science works rely heavily on the distinction between what we might call “observations” and “theories.” Theories are ideas or claims about how the world works, from which we can derive predictions. Observations are what we “see,” either through experimenting or merely perceiving some phenomenon.
How firm is this distinction? Hanson, a philosopher of science from the mid-20th century, argued that through scientific training and experience, scientists acquire “disciplined perception.” Compare what I, a complete chess novice, see when I look at a board position to what Garry Kasparov or Magnus Carlsen, chess grandmasters, see when they look at a board position. The experienced chess player has been trained to perceive things that I simply haven’t been. Hanson argued that something similar goes on for scientists. Aristotle and Galileo could watch the same event—an arrow sailing through the air, for example—and attend to very different aspects of the experience. Their prior understandings influence both what they pay attention to and how they articulate what happened.
A few years after astronomers discovered Neptune, they used the same exact line of reasoning to propose the existence of a new planet, this one closer to the sun: Vulcan. Vulcan’s orbit would explain oddities in Mercury’s orbit. Why hadn’t it been discovered before? It was too close to the sun for astronomers to see it. So, just like with Neptune, astronomers pointed their telescopes to where Vulcan should be. Did they see it? Well, yes, actually. Over several decades, many astronomers, both amateur and professional, saw “Vulcan.” Of course, we’re pretty sure now that the planet doesn’t exist. Einstein’s theory of relativity solved the discrepancy in Mercury’s orbit.
This observation—that observations depend on theoretical background—does not mean that “we only see what we want to see.” An arrow flying backward would shock both Aristotle and Galileo. But it does complicate the falsificationist argument because it suggests that the “tests” that theories undergo may very well be influenced by related theories.
Next time: finding the best explanation.
Questions to ponder
Are there some situations more vulnerable to the critiques of theory testing in this lesson than others? Is there anything scientists can do to address these concerns?
Recommended book
Share with friends