Current computers are precise arrangements of reliable parts. Current computations depend upon this precision, and they control logical abstractions that are only loosely coupled to their physical environment.

We look forward to computations performed by myriads of unreliable parts -- microelectronic computing elements, sensors, and actuators -- so inexpensive that they are mixed into bulk materials such as paints, gels, and concrete.

We can imagine constructing such elements, but we have very little idea how to program them. The limitations here are not the limitations of our technology, but the limitations of our minds.

How do we obtain coherent behavior from the cooperation of large numbers of unreliable parts that are interconnected in unknown, irregular, and time-varying ways?

These questions have of course been anticipated, for example, in studies of self-organizing systems. But the phenomena there are largely observed rather than controlled. As engineers, we must assert control. We must devise the programming principles, engineering languages for design, and tools for the analysis of amorphous machines and processes.

One area to look to for guidance is developmental biology, where the processes of differentiation and morphogenesis appear able to carry out such "computations". Many details of the low-level mechanisms are known here, mechanisms such as diffusion, gradient following, growing points, regularization, dominance, activation and inhibition, although the overall organizing principles in developmental biology remain elusive.

Nevertheless, we can exploit these mechanisms to program amorphous machines. We will demonstrate programs based on purely local parallel processes, having no a priori knowledge of the detailed interconnect, that are able to construct patterns with detailed prespecified useful topologies. We will describe some initial attempts at language appropriate for controlling such processes.

By learning how to organize processes for amorphous machines we will begin the development of the information technology necessary to coopt biological processes to allow us to manufacture novel materials and structures at a molecular scale.

Hal Abelson is Class of 1922 Professor Of Electrical Engineering and Computer Science at MIT. Abelson is a Fellow of the IEEE and winner of the 1995 Taylor L. Booth Education Award given by IEEE Computer Society, cited for his continued contributions to the pedagogy and teaching of introductory computer science.

Hal, together with Gerald Sussman, developed MIT's introductory computer science subject, "Structure and Interpretation of Computer Programs," a subject organized around the notion that a computer language is primarily a formal medium for expressing ideas about methodology, rather than just a way to get a computer to perform operations. This work, through their popular computer science textbook, videotapes of their lectures, and the availability on personal computers of the Scheme dialect of Lisp (used in teaching the course), has had a world-wide impact on university computer-science education.

Abelson and Sussman also cooperate in codirecting the MIT Project on Mathematics and Computation MIT Artificial Intelligence Laboratory and the MIT Laboratory for Computer Science, whose goal is to create better computational tools for science and engineering. The Project is currently focusing on amorphous computing, in work joint with Tom Knight at MIT.