• maybe if we really focus in on what people mean by "nice", and do lots of studies into what makes them think that people are nice, and think really hard, then we can come up with a precise concept of niceness that we can stick a unit on.

Jasay addresses this very counterargument a few paragraphs later:

On the other hand, if they are to be understood as verifiable, refutable matters of fact, interpersonal comparability must mean that any difficulties we may have with adding up are technical and not conceptual; they are due to the inaccessibility, paucity or vagueness of the required information. The problem is how to get at and measure what goes on inside people's heads and not that the heads belong to different persons. Minimal, widely accessible information about Nero, Rome and fiddling, for example, is sufficient for concluding that, for a fact, there was no net gain of utility from the burning of Rome while Nero played the fiddle. Progressively richer, more precise information allows progressively more refined interpersonal findings. Thus we move forward from the non-addibility resulting from sheer lack of specific data to an at least quasi-cardinal utility and its at least partial interpersonal comparison. At least ostensibly, the contrast with proposals to ignore specificity and strip individuals of their differences, could not be more complete. The proposal here seems to be to start from admitted heterogeneity and approach homogeneity of individuals by capturing as many of their differences as possible in pairwise comparisons, as if we were comparing an apple and a pear first in terms of size, sugar content, acidity, colour, specific weight and so on through n separate comparisons of homogeneous attributes, leaving uncompared only residual ones which defy all common measure. Once we have found the n common attributes and performed the comparisons, we have n separate results. These must then be consolidated into a single result, the Comparison, by deciding their relative weights.

Would, however, the admission that this procedure for adding up utilities is intellectually coherent, suffice to make it acceptable for choosing policies? If the procedure were to be operated, a host of debatable issues would first have to be somehow (unanimously?) agreed by everybody whose utility gain or loss was liable to be compared in the operation. What distinguishing traits of each individual (income, education, health, job satisfaction, character, spouse's good or bad disposition, etc.) shall be pairwise compared to infer utility levels or utility differences? If some traits can only be subjectively assessed, rather than read off from Census Bureau statistics, who shall assess them? What weight shall be given to each characteristic in inferring utility, and will the same weight do for people of possibly quite different sensibilities? Whose values shall condition these judgements? If some "equitable" way were unanimously agreed for delegating powers for taking comparative readings and setting the weights, the delegate would either go insane, or would just produce whatever result looked right to his intuition.

The long and short of it is that objective and procedurally defined interpersonal comparisons of utility, even if they are modestly partial, are merely a roundabout route all the way back to irreducible arbitrariness, to be exercised by authority. At the end of the day, it is the intuition of the person making the comparison which decides, or there is no comparison.

I apologize for the excessive quotation length, but I couldn't think of a good chunk to cut.