Hull, Rainbows in Retrospect 
537 
Generations of systematists have also thought that a fairly direct and unproblematic 
relation exists between phylogenetic trees and hierarchical classifications. The problem 
was only in reconstructing the tree. Given the tree, dividing it up into indented taxa 
is just busywork. I think all of the back and forth over cladism has finally disabused 
systematists of this appealing, though mistaken view. Couldn't this literature have 
been made more efficient and succinct? To be sure, early confusions of trees with 
cladograms might well have been avoided, but science in general is not an extremely 
efficient process. All the false starts and dead ends are intrinsic to science. We all 
tend to think that at least we know what ive mean, but time and again we don't 
really understand what we said until we discover what others thought we said. 
Science does not consist in the definitive pronouncements of infallible authorities 
but numerous conversations among extremely fallible human beings. 
Time and again systematists have found the notion of a theory-free classification 
appealing. The claim that systematists can start only where they happen to be sounds 
too haphazard and perilous. There must be some one place where all systematists 
must begin. What better place to begin than observations? Observations as the one 
and only starting place for systematics looks attractive because observations are 
closest to the world that we are trying to classify, but as close as observation statements 
are to the real world, they can be mistaken. Mistaken observations can introduce 
error into science just as readily as mistaken theories. Although the preceding 
conclusion does not come through loud and clear in the taxonomic literature, it does 
seem to be one of the messages to be learned from the disputes that have taken place 
in systematics over the past thirty years. 
In retrospect, on what issues were Johnson and I right — or at least on the winning 
side? Both Johnson and I objected to the notions of overall similarity and general- 
purpose classifications based on overall similarity. Today numerical taxonomists no 
longer advocate such a view, although they play down the importance of any changes 
they may have made on this score. Both Johnson and I interpreted the numerical 
taxonomists as advocating a phenetic philosophy which precluded 'theories' entering 
in the classificatory process, especially in the early stages. We both argued against 
such a notion. In addition, we both recognized a distinct group of cladists termed 
pattern cladists who we thought held views similar to those of the pheneticists with 
respect to the role of theories in classification. 
In the mean time, both numerical taxonomists and cladists have distinguished 
between several different sorts of 'theories.' Some of these theories can enter into 
classification from the start (e.g., the theory of the character); others not (assumed 
phylogenetic relationships). Who was right on this score? The issues are so tangled 
that I cannot say. Perhaps the apparent misinterpretations of phenetics and pattern 
cladists helped to point out places where greater clarity was needed. Perhaps such 
criticisms introduced unnecessary confusions. Who knows? 
With respect to operationism I think that I was not sufficiently sensitive to the issues 
involved. As a philosophical thesis, 1 think that operationism won't do, but as a call to 
connect our theoretical claims with some sort of observational base, I think that it is 
more than appropriate; it is necessary. Perhaps the meaning of theoretical terms cannot 
be totally cashed out in terms of observational consequences, but the more the better. 
Johnson and I had similar reactions to cladism in its early days. It looked too 'rigid.' 
We both preferred more 'reasonable' classifications that result from a mixture of 
various relations. As time has marched on, we are increasingly in the minority. Both 
numerical taxonomists and cladists urged that systematists pick one relation to express 
in their classifications and apply it consistently. Better one relation expressed clearly 
and unambiguously than several relations expressed only impressionistically. 
