Mr. Burkhead maps back and forth between cost, intelligence and complexity, between biological systems and nanites, and (especially!) between robots and software agents so often, and with so little justification, it's hard to see just WHAT he ends up establishing.
For the record:
Intelligence does not necessarily map to complexity, and NEITHER of them necessarily map to cost. This is particularly relevant to the argument in section 21 referring to bootstrapping intelligences.
For reasons given already, both on this list and in Mr. Yudkowsky's "Coding a Transhuman AI", we have NO REASON WHATSOEVER to believe that our brains are the minimum complexity required for our level of intelligence.
Additionally, many engineering projects ARE mostly pre-designed. I can assure Mr. Burkhead that almost all of the discrete components of my computer were not specially designed for my particular configuration. The re-use of component designs of established quality is one of the cornerstones of computer engineering. This practice s in large part responsible for the rapid pace of technological progress in computers and personal electronics. One of the promises of nanotechnology is to allow electronics-like design principles to be practiced on a wider scale, with the corresponding benefits.
Furthermore, complexity does not map to cost. I would be most interested to find out that my Pentium II 450Mhz, for which I paid approxiamately $900 US, is approxiamately one half the complexity of my father's 80386 20Mhz, for which he paid ~$2000 US in the late 1980's.
The most flagrant abuse of this mapping, or "calibration", occurred in section 19. While the macro-scale "Chinabots" of section 17 would admittedly have the limitations discussed, it is not immediatly clear that any of those limitations map to nanobots or nanites. They most especially do not map to software agents performing complex actions (programming) within a computer's memory. "Microserfs", once created, are neither rare nor particularly expensive. Irrelevant licensing agreements not withstanding, there is no reason I cannot have as many as I want running concurrently. They can be duplicated as easily as any other piece of software. This ease of duplication and tight coupling of description to implementation (for computer programs, they are identical) is another of the promises of this technology.
While the latter section of Geniebusters does not have as many flagrant errors as the earlier sections, its theorems and "proofs" are built upon the badly fractured foundation of what has come before.
To conclude, what Mr. Burkhead refers to as "the calibration meme" seems nothing more then the meme of constructing poor analogies, and hiding them behind the prestige of "mathematical mappings."