Sunday, June 17, 2007

Falsifiability and the Importance of the Theory Creation Process


Karl Popper came up with this great idea: Good scientific theories should be falsifiable. If you are already familiar with this idea and what it means, go ahead and skip the next paragraph.

If a theory is falsifiable, it makes predictions about what cannot be observed. That is, it sticks it's neck out and says that there are some particular things that cannot happen. For example, if I say that all swans are white, this is falsiable because all you would need to do is observe a single non-white swan to falsify my
theory. A theory that says that people like to dance to songs with "energy," on the other hand, could very well be non-falsifiable if your measure of a song's "energy" is how likely people are to dance to it. In this case, you'd need an independent measure of "energy," and then see if that measure correlates with the probability of dancing. When a theory is "falsifiable," what this means is that it's potentially falsifiable, given certain possible observations. When a theory makes strong falsifiable predictions, yet fails to get falsified in spite of the efforts of scientists, it is very impressive. For example, relativity predicted that starlight would bend with the gravitational pull of the sun. We had to wait for an eclipse (after Einstein was dead) to measure the displacement of the stars around the sun (because the sun is too bright otherwise). Turns out yes, the stars looked out of place, just like Einstein predicted! If they had not, his theory would have been falsified," which, strictly speaking, means it had been found to be not ompletely true, and in need of revision or abandonment.

The problem is that some scientists take this idea a little too
seriously and think that unfalsifiable theory has no place at all in
the scientific process. Such people seriously underestimate the
complexity and importance of the process of theory creation.


Creating a theory, is, in essence, a highly creative act, and as such
can be affected by all kinds of things: random events in a person's
life, art, and yes, other people's ideas, be they falsifiable or not.

It's kind of understandable why people downplay scientific
creativity. Lay people think science is a straightforward, logical
progression from data to truth. This is wildly off-base.

But even people who study science don't pay enough attention to theory
creation. The field of "science studies" consists of the following
kinds of people:




  • philosophers of science

  • sociologists of science

  • historians of science

  • psychologists of science



Philosophers of science tend to focus on the normative aspects
of science. That is, how scientists ought to do things. Popper
was one of these. The ones who do descriptive work, that is,
describing how scientists actually do what they do, is mostly
limited to theory evaluation, or how scientists choose between
competing theories, rather than how the theories are created in the
first place.

Sociologists of science view science, for the most part, as a
social activity where power relations dominate. They are not
particularly interested in where theory comes from, or, sometimes,
what the theories even are.

Historians of science often do not speculate on the inner
mental workings of the scientists in question. They stick to their
field of expertise, which is telling a coherent story from historical
facts.

Psychologists of science are few. They tend to study things
like theory preference and how hypotheses are tested. It's harder to generalize about this group, because their work is very diverse.

Of all of the scholars working in science studies, only a handful are
actually approaching theory creation as a creative endeavour. The ones
that do include myself, Kevin Dunbar, Paul Thagard, Ryan Tweney,
Ronald Giere, and my former co-advisor Nancy Nersessian.

What little we know about scientific creativity shows, however, is
that often interesting theories come out of analogies with analogues
from outside the discipline. James Clerk Maxwell, for example, used
an analogy with a gear system to help come up with his electromagnetic
equations (Nersessian 1984, 1990). I modeled a part of this theory creation
as a part of my dissertation work (Davies, Nersessian, & Goel, 2001).

Given that ideas for scientific theories can come from such diverse
sources as moving trains (Einstein), dreams (Kekule), and physical
machinery (Maxwell), it should not be surprising that good ideas can
come out of non-falsifiable theories.

Freud's ideas, the poster children for non-falsifiability, continue to
inspire people to create new, falsifiable theories (e.g. Minsky,
2006).

Another way to look at it is in terms of "research programs." Laudan,
a philosopher of science, coined this term (Laudan, 1978). Behaviorism, for example, is not a falsifiable theory. It's a framework, an approach, that uses a particular way of looking at the
world.

When learning about a theory that is unfalsifiable, try to think of it
as a way to look at something, an approach, a springboard of ideas
rather than something to dismiss out of hand. It just might help you
come up with something wonderful.
The pictured photo is one I took during the Ottawa Tulip Festival.

REFERENCES

Davies, J. R., Nersessian, N. J. & Goel, A. K. (2001). The role of visual analogy in scientific discovery. Model-Based Reasoning: Scientific Discovery, Technological Innovation, Values. Pavia, Italy.

Laudan, L. (1978) Progress and Its Problems. University of California Press.

Minsky, M. (2006). The Emotion Machine. Simon & Schuster.

Nersessian, N. J. (1984, 1990) Faraday to Einstein: Constructing Meaning in Scientific Theories. Kluwer.


Popper, K. (reprint 2002) Conjectures and Refutations. Routledge.

1 comment:

Anthony said...

Good article. BTW, Einstein was alive when the eclipse experiment was performed in 1919 - in fact, it's the experiment that made him famous:

http://www.firstscience.com/site/articles/coles.asp

Everyone, I think, underestimates the importance of research programs. A good research program produces a lot of testable ideas even if it's not testable itself - by which mark behaviorism was a great research program and string theory is unfortunately not (given our current ability to test; of course that may change as technology improves).