DNA computing still labors under a preponderance of theory. Laboratory experiments have only explored the nearest shoals. All the techniques from the section above, entitled Tools of the Trade, need to be examined with the same kind of rigor that Khodor and Gifford (1999) used to examine the sequence separation procedure. Potential improvements to the techniques should be explored. An efficient computational model should be derived from the least (or tolerably) error-prone manipulations along with algorithms for handling those errors that are unavoidable. This is the rather mundane work that would bring civilization to the hinterlands of current DNA computing. Nevertheless, glory will surely be heaped on she who demonstrates a DNA solution to a problem beyond the ken of silicon computation. This is the fabled fountain of youth for the field: The so-called killer app that will firmly establish DNA computing as more than just a curiosity. Yet no one truly knows if such a killer app exists. The catastrophes of errors may still sink all our hopes.
Meanwhile, one cause for real optimism in the future of DNA computing is its synergy with molecular biology. Almost any advances in techniques in one field will be useful to the other. The work of Seeman and others (Seeman et al. 1998; Gardner, Cantor, and Collins 2000; Elowitz and Leibler 2000) suggest that the major contribution of DNA computation may not even be to the field of computation but rather to the development of nanotechnology. Even if DNA computers never manage to leapfrog traditional silicon computers, they are still bound to be of value. Even if it proves impossible to implement a universal DNA computer—able to run all our software at blazingly fast speeds—the massively parallel nature of DNA computing is sure to be useful for certain practical problems. I take the position that DNA computation is likely to turn out to have a role similar to neural networks. They will be better than traditional methods for some problems, and worse for others. In this way, I expect DNA computation to complement rather than replace our current computational techniques.
Significant portions of this chapter have been reproduced from Maley (1998). I would like to thank MIT Press for their generosity in allowing me this freedom. I would also like to thank Erik Winfree and Laura Landweber for helpful comments and discussion. I would like to thank Stephanie Forrest for her patience and support. This work was carried out under her aegis and ONR Grant N00014-99-1-0417. Finally, I would like to thank Marvin Minsky for ushering me down this path.