"Science is observable, testable, repeatable and falsifiable. Evolution is neither."
This sort of definition of science is often parroted, but rarely understood. Instead of reciting definitions, we need to take a step back and understand the philosophy behind it.
1. The observation part perhaps needs the most explanation:
Science is based on the principle of empiricism which basically states there is a reality external to ourselves and that our senses inform us about it. In other words, a theory is scientific if its truth has an influence on our sensory experience. What needs to be understood is that everything we observe is essentially just a model of external reality. The so-called qualia (the "redness" of red or heat-as-we-feel it) do not exist on a fundamental level of physics, they only exist inside our brains which means that any observation we make passes through a filter of interpretation. As such, witnessing a phenomenon is not more empirical than just seeing the fingerprints it has left. You can be more confident about it, since it is harder to make mistakes, so it is better evidence, but they are not fundamentally different methods of inquiry and you cannot just ignore fingerprints (this is a metaphor for inference in general) outright. This is the reason why it is not necessary to actually witness a phenomenon to vindicate it scientifically. You do diffract as you pass through a door to know that massive objects (like you) have wave properties (albeit negligible ones at your size). The double slits experiment is sufficient.
Likewise, noone would call Einstein a pseudoscientist because he has never observed the literal curvature of spacetime in the way creationists would like it. It was sufficient for him to know that his theory of gravity had different impacts on the orbits of planets and the trajectory of light than classical mechanics and this was enough for him.
In order for a theory to be even in principle unobservable, it would either need to be causally disconnected from our lightcone, have left its fingerprints in inaccessible locations (like black holes or very specific planets in the Triangulum galaxy), have got its traces removed from human inquiry (this would include very specific historical claims, like the first words Caesar said after he woke up on his 37th birthday; nothing recorded that) or be disconnected to our sensory experience (like an invisible dragon in my garage which is intangible, inaudible and breathes cold fire).
Since evolution claims that every living being arose from a common ancestor in a branching process of descent with modification (plus some horizontal forms of inheritance which I will ignore for the sake of simplicity), we can expect to have access to evolution's fingerprints when we investigate living beings.
I think I already covered the power of inference when responding to other claims.
Let me just say this: Inference IS observable evidence.
2. Testability means that a hypothesis or theory is capable of making predictions which rivaling hypotheses do not make or does not make predictions which other hypotheses do make. In more simple words, something is testable if it can be the
best explanation for something.
Critics of string theory or multiverse theories often point out that postulating fundamental strings or several universes do not make any new predictions that a single universe or the standard model of particle physics do not make which is why these theories are not testable (this is actually a controversial point disputed by advocates of the theories, but it still serves as an illustrative example).
Many "woo" based explanations for consciousness suffer from a similar problem, since they do not make any predictions that classical neuroscience does not make.
As for positive examples, the hypothesis that the Earth rotates is testable because it makes predictions that a fixed Earth does not make. The most obvious one is the Coriolis effect which manifests itself in the apparent deflection of the paths of the winds.
Evolutionary theory does make such predictions. No rivaling hypothesis predicts a nested hierarchy classification scheme (i.e. when you classify organisms on the basis of their morphology and genetics, they fall in groups inside bigger groups inside bigger groups etc.). The only rival I can think of would be some sort of fixity of kinds which would predict consistent nested hierarchies inside these "kinds" and none of these between the "kinds".
3. "Repeatability" should better be rephrased as "reproducibility", as this captures closer what is meant.
Reproducibility is a criterion added to avoid subjectivity. Eyewitness testimony is an example of non-repeatable and hence non-scientific evidence (it is legal evidence though, since it is accepted in a court). If I see someone in my garden, this is not reproducible, since only I can verify it. If this person leaves footprints though, others can verify it as well (unless something washes the footprints away).
An example of reproducibility comes from the Hockey stick graph of Michael Mann (which illustrates climate change over the last two thousand years). Climate change deniers have disputed it, but other scientists were able to reproduce his graph.
Evolutionary theory passes this one. If you make a phylogenetic analysis, other scientists should be able to reproduce your tree.
Reproducibility separates science from history.
4. In order to actually understand falsifiability, we must understand the concept of fallibilism. Empiricism can never prove anything with 100% certainty. Several thought experiments have shown that. A Cartesian demon could delude you by sending you false sensory experiences, a cosmic jester could have created the whole world 3 seconds ago and implanted false memories, records and diary entries of events that never happened and holes in our socks to fool us. You get the point. What is worse, we do not have access to 100% of the knowledge in the universe. No matter how many correct predictions our theory has made, we can never be absolutely sure that this will continue. We can observe the Sun rising a 10,000 times, but this does not guarantee it will do so tomorrow. If a chicken gets fed by a farmer everyday, it could formulate a theory which says "If the farmer comes, I get fed". Then he comes not with food, but with a knife.
Hence, any scientific theory must be falsifiable.
What does that mean?
If you cannot know something with a certainty of 100% or 0%, you need
probabilities. Let's look at Bayes theorem:
I'm gonna steal the description from Yudkowsky:[1]P(A
j) and P(A
i) are the priors of the respective hypotheses. A prior is the probability of a hypothesis before evaluating the evidence. Something like evaluating the probability of a dice roll before rolling it (these priors are irrelevant, I was just explaining in case you are confused, the relevant part is in the Yudkowsky quote).
A hypothesis is falsifiable if some outcomes are more probable than others. It must rule out possibilities.
Freudian psychoanalysis was deemed unfalsifiable because it could come up with equally good explanations for every scenario. Likewise, creationism suffers from the problem that without knowing anything about the abilities or the psychology of the creator, we cannot assign any higher probability to some scenario over another. There is no reason why God should use a common design rather than not using it. Uncommon design in fact covers a broader scope of possible designs.
An example of a falsifiable idea is the existence of protons. When firing particles alpha particles at an atom, the hypothesis that there is a positive particle in the middle of the atom predicts that some alpha particles should be deflected and there are statistical trends in the deflection patterns favored by the proton hypothesis.
Evolution passes this one as well. The likelihood of finding fossils of rabbits in the Neogene is quite high under evolution. So is the likelihood of finding fossils of them in the Paleogene. The probability of finding them in the Cretaceous is lower. The probability of a rabbit in the Jurassic is low. The probability of a rabbit in the Triassic is even lower. The probability of a rabbit in the Triassic is fantastically improbable under evolution.
Other possible falsifications:
-Things like crocoducks or pegasi which are very difficult to classify and violate the principle of homology.
-Huge morphological leaps in a generation, like a monkey giving birth to a human.
-Nonsensical adaptations, like blind creatures being brightly colored and clearly visible to everyone for no purpose.
-Inconsistent phylogeny (a phylogenetic tree based on pseudogenes having no resemblance to any tree based on ERVs, for instance).
None of this would drop the likelihood of evolution down to 0%, but depending on the statistical significance of findings, they could damage it so much that it could fall out of favor.
[1]
lesswrong.com/lw/q4/decoherence_is_falsifiable_and_testable/This comment = The scientific method in a nutshell.
