At Professor Solum's blawg on Legal Theory, he has a long posting on the theory of conflicts in class litigation. N.B. Not so long ago, I was a plaintiff's-side class action attorney; I think I still remember quite a bit about it. In any event, Professor Solum's basic conclusion is this:
It is now obvious that something has gone dreadfully wrong with our application of the veil of ignorance to the problem of hypothetical consent to intraclass conflicts…. Putting the class members behind a veil of ignorance that excludes knowledge of position in the class is an inappropriate way of representing deliberation by a reasonable party deciding whether to consent to a conflict, because “position in the class” is what is at stake…. Our problem was conflicts of interest. You can’t reason about conflicts of interest by imagining that they don’t exist and then asking, “If there were no conflict, would you consent to it.” In other words, Miller’s idea is incoherent at a deep level. (emphasis in original)
An even simpler statement of why the "veil of ignorance" is not a useful tool in evaluating potential intraclass conflicts comes from recognizing that a conflict has both a quantitative and a qualitative dimension. That is, evaluating a conflict requires understanding both its prevalence and its seriousness. The burden of proof in a criminal prosecution is a good example of a balancing to limit the "value" of intraclass conflicts. Behind the veil of ignorance, one must assume that one might be a member of the class of wrongly accused individuals at some time. The "qualitative" dimension of a wrongful conviction is considered so great that the burden to convict is set high enough to limit the "quantitative" dimension of that wrongful conviction—that is, limit its prevalence to an "acceptable" level.
The "veil of ignorance" conflates these two dimensions. It asks a single, binary question: is the intraclass conflict "serious enough" that a representative action is inappropriate? However, since the very nature of a conflict is not binary, but multivalued, it is asking an unanswerable question—because the "high quality and high quantity" quadrant of the truth table (in our example above, letting the guilty go free) can easily be construed to overwhelm the actual conflict. Without the direct data on both the qualitative and quantitative dimensions, no answer to the question is sound, because it relies upon incomplete data.
Now look at Amazon's "search inside the book" feature. For those who do not object, there is no possible conflict. For authors of book-length works who do object, the initial resolution must be with the publisher, because typical publishing contracts give the publisher the right to put out excerpts as publicity. Those who do object, and write shorter works that are collected in longer works, must resolve their problems with multiple actors: Amazon and the publisher (see Tasini). This creates a clear conflict; but using a "veil of ignorance" analysis would hide it, because it is not possible to balance the interests without knowledge of the initial position. The question then becomes this: does that last quadrant indicate that Amazon's program should/must be opt-out (that is, include everything unless objected to) or opt-in (include nothing without adequate permission)? The follow-up question is who may make that decision?