Wednesday, January 18, 2017

Scientific Certainty

I recently had a conversation with a fellow skeptic regarding scientific certainty, specifically that we cannot be 100% sure of the correctness of our theories. He was adamant that nothing in science is certain, and that some level of skepticism regarding scientific knowledge was always appropriate. I, however, argued that that was an overstatement, possibly dogmatic. I felt that some things are certain (such as the mathematics used by scientists, for reasons stated in the last paragraph of this section), and many things are almost (though not completely) certain, such as what are sometimes called "our best theories". Admittedly, we lack full certainty for things "of the world" - the world of experience - for well known philosophic reasons. Anything based on empirical evidence, collected through repeated observations, and accounted for by theory, is subject to some level of uncertainty. Conclusions reached through inductive reasoning and evidence, rather than deduction, will always be subject to potential future updating, modification, or even disproof through new observations and tests, or revamped theories. We can't say, for example, that we have "proved" that all copper transmits electricity, because we haven't tested all instances of copper. We can only infer that all copper probably conducts electricity because we have never seen a case where it doesn't, and because our theories about the properties of copper and of how electricity works supports this idea.

A Bayesian approach can shed light on this. We assign very high prior probabilities and confidence in the correctness of our best and most rigorously tested theories and hypotheses. The likelihood of encountering counter-evidence given that these hypotheses are true is very low (will we find a crystal sphere of stars in the heavens? will we find rabbit fossils next to trilobite fossils?). Given how Bayes Theorem works, the probability of discarding our existing theories is extremely low - low enough not to waste much time thinking about.

So, some theories are so well supported that they are "practically" certain. The confidence we can have in some well supported theories is very high, asymptotically approaching 100%, and is established to such a degree that no one spends any research time trying to find alternate explanations (e.g., there is no serious research into Flat Earth, Phlogiston, Geo-centrism, Astrology, or Young Earth Creationism). But, philosophically, the level of confidence cannot be 100%. To put any weight on the lack of certainty for quantum theory, evolutionary theory, atomic theory, or the helio-centric model of the solar system would be perverse, arbitrary, contrary, and unproductive. These are among our best theories, and until something better comes along, they are best treated as being essentially correct (though, of course, subject to continued refinement and elaboration).

I am confident (though not certain) that 100 or 1000 years from now there will be some logical entity that occupies the concept that we currently call an "atom", and that two oxygen atoms and one hydrogen atom will continue to be the basis of a wet substance called water. The basis of the feeling of certainty is the knowledge that, although the details of what makes up the entity we call an "atom" have changed, the "placeholder" for an atom-like thing persists. The model of the atom has evolved, and will probably continue to change in the future (i.e., the models proposed by Democritus, Dalton, Thompson, Rutherford, Heisenberg/Chadwick/Bohr, up to the current quantum mechanical "cloud" model). Atomic theory may be subsumed as a special case of some more comprehensive theories of matter and energy. But it will still have applicability in chemistry and normal daily living, just as Newtonian physics, though extended by Relativistic physics, is still appropriate in a subset of cases. "F=MA" is probably not going away, though exceptions outside the bounds of normal human experience can be found. Newton will almost certainly continue to be applied in low-mass, low-velocity scenarios for which it is appropriate, while more sophisticated models will be used for cases outside those environments. As Stephen Jay Gould wrote,

In science, "fact" can only mean "confirmed to such a degree that it would be perverse to withhold provisional assent." I suppose that apples might start to rise tomorrow, but the possibility does not merit equal time in physics classrooms.

As mentioned above, there are some things in which we do have 100% confidence. These involve "analytic statements" (Kant), also called "relations of ideas" (Hume), where a statement is true or false based on its meaning or definition. For example - no bachelors are married, 2+2=4, and triangles have 3 sides. These statements are "analytically true", by virture of their meanings. When the premises are true, and valid logic is used, then the conclusion is fully contained in the premises and 100% certainty is established.