Chris Satterley raised a number of interesting points, and I shall comment on one of those here. He stated that global warming modelers are frequently asked, "What are your assumptions?" My issue is, if there is doubt about an assumption, will anybody do anything about it? I would like to explain, using an example from my own past, why I am a little skeptical. (This is most certainly not a whinge. The only reason I raise it is that I know the details.)
 
Early in my career, I was interested in strained molecules, and consequently I became interested in bond bending. Some time ago I went to a conference and in a session where there was nothing particularly relevant to my direct interests, I sat in on a session on molecular mechanics. In the presentations, bond bending was modeled on the assumption that it was simple harmonic, in which restoring force is proportional to deformation, hence the energy of the deformation is proportional to the square of the deformation in radians.
 
So, what else could it be? Consider a C-H bond in methane, in its equilibrium position. If a plane is drawn through the carbon atom normal to that bond, all the other bonds are on the distal side of the plane, and the charge distribution is more or less symmetrical. To my mind, that indicated that the repulsive force should be along the line of the bond, and to the extent that the deformations did not remove the symmetry, the dynamics would be similar to those of a pendulum, in which case the deformation energy would depend on the sine of the angle of deformation. If anyone is interested, the calculation of overtones was quite respectable (at least in my opinion) for a very few limited molecules (Aust J. Chem 22: 2575-2580).
 
The point of all this is, when I raised this to one of the main speakers, (a) he hadn't heard of this alternative, but more importantly, (b) he was not going to do anything about it. Why not? My view is that it depends on funding. The main function of a project leader is to get the project funded. While the nature of this problem varies from country to country, usually some form of performance review is required. I doubt anybody has the nerve to write in a fund application that they intend to turn over the last ten years' work to determine whether a primary assumption was wrong when during that ten years they have been funded based on their "remarkably good" results. (The fact that there are so many validation constants in the programs is beside the point.) Of course, there should be some form of evaluation of scientists' performances, but I am far from convinced that the current methodology is good for science. The problem is, this procedure is almost designed to lock in any previous incorrect assumptions, and while I am sure that was never the intention, it is one of the unfortunate unwanted consequences. Yes, pointing out a problem is easy; solving it is not, nevertheless, pointing it out may be a start.
 
This may seem harmless, and some may think, even if the underpinning relationships are wrong, if your models reproduce observation, does it matter? To me, the answer is, yes. The problem arises when the model is taken into new territory. The most successful theory, at least in terms of time over which good results were always predicted, was the model of Claudius Ptolemy. It always predicted where the planets would be, when the eclipses would be, etc. However, because it is wrong, if NASA used it for manned flights there would be a lot of dead astronauts. In this bond bending example, maybe it doesn't matter if we don't know exactly why certain polymer solutions have certain properties, but suppose we want to devise new biocatalysts – effectively, synthetic enzymes. Would it not be desirable that we know what we are doing?