Recently, Facebook CEO Mark Zuckerberg chose Thomas Kuhn’s classic The Structure of Scientific Revolutions for his book discussion group. And although I don’t usually try to update this blog with the most recent controversy of the day, this time I can’t resist jumping on the Internet bandwagon and delving into this difficult, challenging book.

To briefly summarize, Kuhn disputes the traditional notion of science as one of cumulative growth, in which Galileo and Kepler build upon Copernicus, Newton builds upon Galileo and Kepler, and Einstein builds upon Newton. This picture of cumulative growth may be accurate for periods of “normal science,” Kuhn writes, when the community of scientists are working from the same general picture of the universe. But there are periods when the common picture of the universe (which Kuhn refers to as a “paradigm”) undergoes a revolutionary change. A radically new picture of the universe emerges in the community of scientists, old words and concepts obtain new meanings, and scientific consensus is challenged by conflict between traditionalists and adherents of the new paradigm. If the new paradigm is generally successful in solving new puzzles AND solving older puzzles that the previous paradigm solved, the community of scientists gradually moves to accept the new paradigm — though this often requires that stubborn traditionalists eventually die off.

According to Kuhn, science as a whole progressed cumulatively in the sense that science became better and better at solving puzzles and predicting things, such as the motions of the planets and stars. But the notion that scientific progress was bringing us closer and closer to the Truth, was in Kuhn’s view highly problematic. He felt there was no theory-independent way of saying what was really “out there” — conceptions of reality were inextricably linked to the human mind and its methods of perceiving, selecting, and organizing information. Rather than seeing science as evolving closer and closer to an ultimate goal, Kuhn made an analogy to biological evolution, noting that life evolves into higher forms, but there is no evidence of a final goal toward which life is heading. According to Kuhn,

I do not doubt, for example, that Newton’s mechanics improves on Aristotle’s and that Einstein’s improves on Newton’s as instruments for puzzle-solving. But I can see in their succession no coherent direction of ontological development. On the contrary, in some important respects, though by no means all, Einstein’s general theory of relativity is closer to Aristotle’s than either of them is to Newton’s. (Structure of Scientific Revolutions, postscript, pp. 206-7.)

This claim has bothered many. In the view of Kuhn’s critics, if a theory solves more puzzles, predicts more phenomena to a greater degree of accuracy, the theory must be a more accurate picture of reality, bringing us closer and closer to the Truth. This is a “common sense” conclusion that would seem to be irrefutable. One writer in Scientific American comments on Kuhn’s appeal to “relativists,” and argues:

Kuhn’s insight forced him to take the untenable position that because all scientific theories fall short of absolute, mystical truth, they are all equally untrue. Because we cannot discover The Answer, we cannot find any answers. His mysticism led him to a position as absurd as that of the literary sophists who argue that all texts — from The Tempest to an ad for a new brand of vodka — are equally meaningless, or meaningful. (“What Thomas Kuhn Really Thought About Scientific ‘Truth’“)

Many others have also charged Kuhn with relativism, so it is important to take some time to examine this charge.

What people seem to have a hard time grasping is what scientific theories actually accomplish. Scientific theories or models can in fact be very good at solving puzzles or predicting outcomes without being an accurate reflection of reality — in fact, in many cases theories have to be unrealistic in order to be useful! Why? A theory must accomplish several goals, but some of these goals are incompatible, requiring a tradeoff of values. For example, the best theories generalize as much as possible, but since there are exceptions to almost every generalization, there is a tradeoff between generalizability and accuracy. As Nancy Cartwright and Ronald Giere have pointed out, the “laws of physics” have many exceptions when matched to actual phenomena; but we cherish the laws of physics because of their wide scope: they subsume millions of observations under a small number of general principles, even though specific cases usually don’t exactly match the predictions of any one law.

There is also a tradeoff between accuracy and simplicity. Complete accuracy in many cases may require dozens of complex calculations; but most of the time, complete accuracy is not required, so scientists go with the simplest possible principles and calculations. For example, when dealing with gravity, Newton’s theory is much simpler than Einstein’s, so scientists use Newton’s equations until circumstances require them to use Einstein’s equations. (For more on theoretical flexibility, see this post.)

Finally, there is a tradeoff between explanation and prediction. Many people assume that explanation and prediction are two sides of the same coin, but in fact it is not only possible to predict outcomes without having a good causal model, sometimes focusing on causation gets in the way of developing a good predictive model. Why? Sometimes it’s difficult to observe or measure causal variables, so you build your model using variables that are observable and measurable even if those variables are merely associated with certain outcomes and may not cause those outcomes. To choose a very simple example, a model that posits that a rooster crowing leads to the rising of the sun can be a very good predictive model while saying nothing about causation. And there are actually many examples of this in contemporary scientific practice. Scientists working for the Netflix corporation on improving the prediction of customers’ movie preferences have built a highly valuable predictive model using associations between certain data points, even though they don’t have a true causal model. (See Galit Shmueli, “To Explain or Predict” in Statistical Science, 2010, vol. 25, no. 3)

Not only is there no single, correct way to make these value tradeoffs, it is often the case that one can end up with multiple, incompatible theories that deal with the same phenomena, and there is no obvious choice as to which theory is best. As Kuhn has pointed out, new theories become widely accepted among the community of scientists only when the new theory can account for anomalies in the old theory AND yet also conserve at least most of the predictions of the old theory. Even so, it is not long before even newer theories come along that also seem to account for the same phenomena equally well. Is it relativism to recognize this fact? Not really. Does the reality of multiple, incompatible theories mean that every person’s opinion is equally valid? No. There are still firm standards in science. But there can be more than one answer to a problem. The square root of 1,000,000 can be 1000 or -1000. That doesn’t mean that any answer to the square root of 1,000,000 is valid!

Physicist Stephen Hawking and philosopher Ronald Giere have made the analogy between scientific theories and maps. A map is an attempt to reduce a very large, approximately spherical, three dimensional object — the earth — to a flat surface. There is no single correct way to make a map, and all maps involve some level of inaccuracy and distortion. If you want accurate distances, the areas of the land masses will be inaccurate, and vice versa. With a small scale, you can depict large areas but lose detail. If you want to depict great detail, you will have to make a map with a larger scale. If you want to depict all geographic features, your map may become so cluttered with detail it is not useful, so you have to choose which details are important — roads, rivers, trees, buildings, elevation, agricultural areas, etc. North can be “up” on your map, but it does not have to be. In fact, it’s possible to make an infinite number of valid maps, as long as they are useful for some purpose. That does not mean that anyone can make a good map, that there are no standards. Making good maps requires knowledge and great skill.

As I noted above, physicists tend to prefer Newton’s theory of gravity rather than Einstein’s to predict the motion of celestial objects because it is simpler. There’s nothing wrong with this, but it is worth pointing out that Einstein’s picture of gravity is completely different from Newton’s. In Newton’s view, space and time are separate, absolute entities, space is flat, and gravity is a force that pulls objects away from the straight lines that the law of inertia would normally make them follow. In Einstein’s view, space and time are combined into one entity, spacetime, space and time are relative, not absolute, spacetime is curved in the presence of mass, and when objects orbit a planet it is not because the force of gravity is overcoming inertia (gravity is in fact a “fictitious force“), but because objects are obeying the law of inertia by following the curved paths of spacetime! In terms of prediction, Einstein’s view of gravity offers an incremental improvement to Newton’s, but Einstein’s picture of gravity is so radically different, Kuhn was right in seeing Einstein’s theory as a revolution. But scientists continue to use Newton’s theory, because it mostly retains the value of prediction while excelling in the value of simplicity.

Stephen Hawking explains why science is not likely to progress to a single, “correct” picture of the universe:

[O]our brains interpret the input from our sensory organs by making a model of the world. When such a model is successful at explaining events, we tend to attribute to it, and the elements and concepts that constitute it, the quality of reality or absolute truth. But there may be different ways in which one could model the same physical situation, with each employing different fundamental elements and concepts. If two such physical theories or models accurately predict the same events, one cannot be said to be more real than the other; rather we are free to use whichever model is more convenient.  (The Grand Design, p. 7)

I don’t think this is “relativism,” but if people insist that it is relativism, it’s not Kuhn who is the guilty party. Kuhn is simply exposing what scientists do.