Arlin's title is not at all self-explanatory; here's what I now think the words are intended to mean: 'Evolution' means 'a change over time in how a function is accomplished'. 'Constructive' means 'the change is that the function is accomplished in a more complex way'. And as a result of some helpful questions at the end, I now think that 'neutral' means 'the function itself is under stabilizing selection but not under adaptive/directional selection, and how it is accomplished (the change in complexity) is not under selection at all'. Ford didn't define 'complexity' until the question period; he then suggested that one measure of a function's complexity might be the number of components required for it.
Here's the executive summary:
Once organisms have evolved to have many components, some components will inevitably interact with others in 'accidental' ways that have, at least initially, not been shaped by selection. Once these accidental interactions exist, they will modify how selection acts on mutations that affect the function, sometimes making things worse but sometimes mitigating the effects of what would otherwise be deleterious mutations eliminated by selection*. These mitigating effects will weaken stabilizing selection on the function, sometimes allowing the mutations to be preserved (especially if populations are small). Preservation of the mutation effectively creates selection for maintenance of the formerly-unselected interaction. The function has become more 'complex (by Ford's definition), but there hasn't been any selection for the complexity. If mutations with these kinds of effects recur repeatedly, the function will become increasingly complex without having been in any way improved.As evidence that this type of complexity-building is common and important, Ford cited several molecular examples where a process has become ridiculously ('stupidly') complex but doesn't work any better that simpler versions. The RNA editing of trypanosomes is not well known but is a compelling example. So are introns and the spliceosomal machinery that lets eukaryotes cope with them. Simpler examples are the 'maturation proteins' that assist type I and II self-splicing introns. The ribosome itself may be a (not very stupid) example, where proteins have gradually taken over activities originally handled by the catalytic RNAs.
The issue didn't seem very important to the evolutionary biologists in the audience, I think because they don't constantly deal with the just-so-story functions that molecular biologists typically ascribe to any complicating feature of a process. To many molecular biologists, every base pair in the genome, every intermolecular interaction, and every small RNA in the cell is the product of adaptive selection. There are no accidental interactions. Shit never just happens.
*On the other hand (not considered by Ford at all), mutations whose effects are made worse by the accidental interaction will be more efficiently eliminated by the stabilizing selection on the function. I don't think this can be said to reduce the complexity of the function, because the interaction was accidental and thus not included in the complexity count. I don't know if it would it create selection against the interaction.
**Psci Wavefunction has blogged about this concept in some detail, here and here. I confess that I haven't read these very long posts through, but perhaps now I will. (She asked an excellent question after the talk.)
*** Somewhere in his talk Ford was describing clade selection; using the example of how a propensity to speciate can cause a lineage to have many more species than other lineages. He said that more species means more individuals, but that's certainly not true.