The processing capabilities of the prototype described in this chapter are, of course, extremely modest, indeed even minimal, in comparison to the architectural projections of the previous section. It is to be regarded only as an initial step designed to concretize the conformation-driven computing concept and to demonstrate its technological feasibility at the level of what might be called macroscopic fluidics. The step to lab-on-a-chip integration can readily be seen.
The important question concerns the basic claim—namely, that the conformationdriven approach should provide access to computational processes that cannot practically fit into a conventional architecture. The term transformal computing is apt. How would we even recognize whether a computational system performs an operation that is refractory to digital (i.e., formal) machines?
The famous thesis of Church and Turing asserts, in its strong form, that all processes in nature can be brought into the circle of formal computation (Hofstadter 1980). This is an open question. Whether the answer is affirmative or negative is not the issue with which we are concerned here. It is the practical question that is relevant. Many examples could be cited: human aesthetic judgments, legal judgments, ethical rules (like the Golden Rule), or any decision that involves an indefinitely large number of situations. Arguably, an unambiguous description of such general decision rules by formal rules (i.e., by a program in the Turing sense) is infeasible. We here enter the realm of what was referred to above as transformal computations.
Of course, we do not expect the conformation-driven technology proposed here to perform such complex human operations either. Constructing an artificial brain that comes close to the human brain, even under the reasonable assumption that conformational processing plays a key role in the human mental process, exceeds by far any expectations that we would care to project. The proper question is: Can conformational processors perform transformations that exceed the practical capabilities of formal machines; and how could such transformations be identified?
Take as a concrete example the functioning of an assembly line. Automation is limited by the speed of visual processing and by the fact that quality control problems are often ambiguous. If conformational processors were evolved and harvested that could preprocess ambiguous patterns in a manner that made them suitable for processing by vision algorithms, it would constitute what in practice might be called a transformal computation.
By choosing to look at the benchmark XOR operation, we have a fortiori precluded the possibility of finding a transformal transformation. Our objective was to demonstrate that even a single enzyme species could do more processing than is standardly attributed to the threshold elements utilized in many current neural net models. Our working hypothesis—that we can use the conformation-driven approach to escape the practical limitations of programmable machines—is based on three considerations: the complexity arguments indicating that systems with selforganizing dynamics can perform more-complex operations than systems with programmable architectures, the technological feasibility of fabricating conformationdriven modules that utilize self-organizing dynamics, and the feasibility of using an evolutionary response surface methodology for developing a repertoire of highcomplexity basis transforms that can be embedded in or conjoined with higher level architectures. This is a three-point landing on theory, technology, and architecture. The pieces are present; bringing them together should yield computational capabilities complementary to and synergistic with digital capabilities.