Saturday & Sunday, 17-18 March 2007
Two Scandalous Secrets of Philosophy of Science
It was an experiment: a type of workshop we haven't run before. Workshops usually select one theme and marshall speakers and discussants to probe and dissect it. This time, our focus was not a theme, but a text. It was Elliott Sober's draft manuscript, Evidence and Evolution. We had printed off multiple copies, bound them with those awkward plastic combs and mailed them off to the workshop registrants a month ahead of the event. They were the select few. To keep the discussion informed and manageable, we had restricted registration. These few were, supposedly, studying the text minutely in preparation for the event.
On Saturday morning, we squeezed around the table in our small lounge, picking at fruit and breakfast pastries and sipping coffee. Their faces were weary. Perhaps it came from late night study of the manuscript. More likely it was the ordeal of travel that brought participants from far afield. An unusual March storm had struck the east coast in advance of the event. We had spent Thursday tracking flights of participants and finding one after another arriving well enough.
There was one exception. Samir Okasha's flight from Bristol, England, made it to Dulles airport near Washington DC. There everything stalled, a mere 250 miles and four hours drive away. He faced an overnight stay in Washington and at best a 50% chance of a morning flight the next day. In an inspired moment Samir did what would let him "one up" any of the tales of travel woe that circulate over stale beer in conference hotels. Samir caught a cab to Pittsburgh.
There are four chapter in Elliot's draft and we had selected one commentator for each. The first chapter was the "Concept of Evidence" and our own John Earman had agreed to comment. Elliott would then respond.
After the opening formalities had concluded and John stepped towards the overhead projector, I realized what a truly extraordinary event we had before us. In the world of philosophy of science, there are only a few who are acknowledged as unchallenged leaders of their fields. In philosophy of physics, it is John Earman. In philosophy of biology, it is Elliott Sober. One might not imagine that their worlds would intersect. Yet it happens that both have a strong interest in Bayesian confirmation theory and that topic permeates the chapter.
So what would John say? John has a love for precise thought and technical analysis and there were many invitations to it in the first chapter. I fully expected him to take them. He did not. Instead he promised to reveal to us a scandal and open dirty secret in philosophy of science. And he did; twice.
It began with a recalling of the early court decision against creation science in Arkansas. That ruling had drawn heavily on expert testimony to make up a list of five essential characteristics of science. On the authority of that list the judge decreed that scientific creationism is unscientific. That list, John asserted, was so defective that any graduate student could tear it to pieces.
For many, John is a distant authority, known only through his copious writing with their trademark humorous quirks. Who else, as a junior professor, would write a paper, "Who's Afraid of Absolute Space?" When he speaks, that same impishness lurks just millmeters from the surface. He is as likely to recall some obscure theorem as he is politely to urge with delicately controlled passion that an errant philosopher of science should burn in hell.
John began to review the problems of the Arkansas list well-known to us all in philosophy of science. Indeed I had taught some of this in my introduction to philosophy of science the previous term. Falsifiability? Is science demarcated from non-science in that it is falsifiable? We can all think up unfalsifiable, but perfectly respectable, scientific propositions. I think of the proposition that somewhere--who knows where--there is a magnetic monopole. Or what of the idea that the conclusions drawn by science are always tentative? Well that flies in the face of what scientists do and have a right to do. Yes, I thought, chemists would not be too pleased to hear that the oxygen theory of combustion is only tentative. And what of testability? Creation science is testable, John urged. It had been tested and has failed the test.
So what should we do? John allowed that Larry Laudan had posed the challenge correctly. We should not seek to do away with creation science in a single, easily deflected blow. That was the clumsy attempt to denounce creation science as unscientific. It should be discarded because we can show that it fails to do what all good science must: be well supported by arguments and evidence. How to do that is Laudan's challenge.
One would hope that this dismissal of creation science could be effected mechanically. We assemble the evidence at hand; we process it through our best logic of inductive inference; and we find evolutionary theory well-supported and creation science not.
Now came John's second scandalous secret. It tied directly into the content of Elliott's first chapter. In spite of our best efforts, we have been unable to find an unobjectionable account of evidence that would do the job. John now moved into the technical issues. He began to review the various accounts of inductive inference: Bayesianism, subjective and objective; Neyman-Pearson hypothesis testing; likelihoodism; Akaike's criterion for model evaluation; and more. Each succumbed to one or other objection.
For a moment, a familiar John appeared. He rolled up the overhead projector screen and one of his trademark spacetime diagrams grew on the chalkboard. It illustrated what John called the "smart Alec" version of creationism. It showed a world created just recently with the fossil record and all other traces of great antiquity miraculously in place at its first moments.
All too soon it was over. Perhaps, John concluded wearily, we must just pragmatically decide that we belong to the community of Bayesian inducers and insist that ours is the best inferential community. Or fail that, we must judge the import of evidence locally and allow that we cannot call up a global theory of inductive inference to guide us. He sat down.
It was now Elliott's turn to respond. He had been patiently jotting notes as John spoke. He had spent more time facing his writing pad than looking at John. However he had little to dispute in John's saga of two scandalous secrets.
There was some fine tuning to be done. Elliott did correct the account of creation science. It has now been replaced by a much more austere doctrine, intelligent design. ID jettisons virtually all the commitments of the old creation science, retaining only the idea that the irreducible complexity of the biological world is only explicable by an intelligent designer. That, Elliot thought, may well have rendered it untestable. And perhaps we should not be so severe on the ruminations of jurists, for their context and responsibilities are legal. They are not epistemologists.
Elliott clearly shared John's yearning to for the success of Bayesian inductive inference. But he conceded wearily that there are too many reasons against it. A Bayesian must assign prior probabilities to hypotheses. How could one do that for big theories like evolution, general relativity and their unknown alternatives, without admitting that the assignments are subjective, thereby sacrificing intersubjective agreement. He finished, glumly, as did John, by conceding that no account of induction works globally. The best we can do is to work locally, a piecemeal: "cut and paste" epistemology. He sat with John at the table at the front of the room.
Sandy called us to order for questions. The room had a rather different feel from most workshops. We had invited many graduate students to participate. Almost a half of the visitors were graduate students from other universities. So the room was young and eager. I liked it.
There was the obligatory pause. No one really wants to ask the first question. That would suggest a degree of self-confidence that would be foolhardy in room filled with so many astute minds. But soon inhibitions faded and the questions began. I managed to insert my question, picking up on John and Elliott's gloomy ending. Why so glum?! Isn't the recognition that induction is local the first step in finally getting clear on a problem that has vexed us since Aristotle? I tried to hide my own glee since that is a theme I have argued energetically in print.
The hour passed and we broke for lunch. Scanning down the program, it was clear what themes would be dominating this early part of the meeting. After lunch, we would meet to hear Sandy Mitchell's commentary on Chapter 2, "Intelligent Design." Elliot had said that there could only be finitely many errors in his draft, so perhaps we might be able to bring them down to zero. That process had really only just begun; but here my narrative will end.
John D. Norton