Coming: Failsafe software

Computerworld - Richard P. Gabriel is Sun Microsystems Inc.'s open-source expert. He has written four books and more than 100 articles, mostly about programming languages and practices. He's leader of the Feyerabend Project, whose goal is "to repair the arena of software development and practice." It was inspired by the philosopher Paul Feyerabend, who wrote in 1975, "Given any rule, however 'fundamental' or 'necessary' for science, there are always circumstances when it is advisable not only to ignore the rule, but to adopt its opposite." Gabriel recently told Computerworld's Gary H. Anthes what's wrong with software today and how it might be improved.

What's wrong with today's programming languages? They are brittle. They were based on the assumption that computers were for scientific and engineering applications. They were designed when computers did thousands of instructions per second, not billions. And when computers cost millions, programmers cost $12,000 a year. Now developers cost $50,000 to $200,000 a year, and they can each have the equivalent of five Cray 2's [supercomputers] on their desk. So it's time to use some of that CPU power to help the programmer.

How could languages be less brittle? The software we build doesn't exhibit homeostasis - that's when an organism is perturbed, it tries to return to its normal state. When someone hacks into a Unix system and does something bad, it doesn't get better until someone repairs it and it is rebooted. But you could make every component in the system tend toward its default state over time. You could have the programming language set up so that your system kind of gravitates back toward its initial state. If you had proposed that in 1959, when programming language ideas were being thought up, they'd have said, "That's crazy. Why would we waste our computers' time doing that?"

What else could be done? From the late 1970s to the late 1980s, folks in the artificial intelligence world tried to program computers differently, with languages like LISP. They had expert systems, which is an easier way to program a computer and not as susceptible to errors. On the other hand, the results that you got might not be precisely what you expected. A rule-based system will give you robust, flexible behavior which is at the 95% level, but certainly sufficient. We have abandoned all of these ways of programming. Java goes a fair way to some of these goals, but still there is a relentless insistence on correctness and precision instead of flexibility and adaptability.

Are you suggesting we write applications today as expert, or rule-based, systems? Is that possible?