New Initiatives Aim To Emancipate 'Scientist-Programmers'

Cornell University’s Eldredge Bermingham spends hours each day, against his will, in front of a computer. Bermingham isn’t a reluctant computer scientist; he’s just a biologist who needs some good software. He uses the latest techniques in molecular genetics to understand how new species are formed. His work requires computer programs so specialized that most of them aren’t available commercially. So Bermingham has to write his own, or hire programmers to help him. Overseeing the process often takes a quarter of his working day and keeps him from his laboratory. “It’s just a gross expense of time,” he complains. Berrningham is one of a swelling number of researchers who have experienced the dark side of the computer revolution. As workstations and PCs infiltrate nearly every corner of scientific endeavor and grow ever more powerful and capable, scientists are spending more time trying to satisfy their machine’s insatiable appetite for better software. For these scientists-programmers, help is on the way. This summer, officials from the National Science Foundation and the Defense Advanced Research Projects Agency (DARPA) told Congress about a series of new initiatives aimed at easing the programming burden of computer-based science. The efforts range from DARPA’s multimillion-dollar Common Prototyping System campaign for a “programming environment” that can automate most of the steps in software development, to NSF’s experimental Software Capitalization Program, which simply encourages scientists to buy or develop reusable software tools. The problem lies in the enormous power of the computer to perform tasks previously too complex or time-consuming for the average scientist. But those simulations, data acquisition systems, theoretical models, and graphics often require writing new software, sometimes from the ground up. In the midst of their labors, those scientists-turnedresentful-programmers can’t help wondering whether somewhere, in some other lab, someone is doing exactly the same thing. The answer is yes, says William Wulf, NSF assistant director for computer and information science and engineering. And that duplication of effort is an inefficient use of valuable resources. Says Wulf: “Too often [researchers] with sketchy Backgrounds in software development are being averted from their primary scientific investigation to develop the software they need. Productivity suffers, he adds, when scientists are forced to write then own software from scratch because they don’t trust—or even know about—existing programs. So concerned is WuIf about the problem of science’s “software gap” that he has diverted $1 million of his own office’s discretionary funds to create the trial Software Capitalization Program. The funds are matched by other NSF programs that pay for scientific software as part of their research grants. Wulf, who had previously headed a private software company and was a member of the University of Virginia faculty before coming to NSF last summer, says he was “really struck by the poor quality of the software tools academics have, compared to their private counterparts. The reason is that NSF has not usually wanted to pay for software development. I’m willing to pay for that.” The program is meant to help researchers develop software that other scientists can use, explains NSF Software Engineering Program manager John Gannon. One group, for example, is designing a simulation of a working railroad. Once it:s completed, researchers who are investigating “real-time systems” will be able to use the simulation to test their work on a typical real-world situation. Neural Simulation Package Neurobiologist James Bower is already proving how useful such efforts can be. Over the last two years, the Caltech researcher and his collaborators have created a neural network “simulation tool” that promises to shorten future neural simulation efforts by months and even years. The package, known as GENESIS (for GEneral NEtwork SImulation System), lets neural researchers build new simulations with simple, yet powerful, commands and “point-and-click” graphics, without the need to ever see a line of standard programming code. Bower has begun to teach a new generation of neurobiologists at the Woods Hole Marine Biology Laboratory (MBL) how to build networks with the new system. When his MBL students return to their home institution this month, each one will take a copy of GENESIS with them. The program will find even wider distribution as Bower releases portions into the public domain. Eventually, Bower hopes, the common practice of building neural network simulations from scratch will become a thing of the past. Bower says that scientists immediately see the advantages of his approach. “It’s the best neural network simulation system there is,” he says. “It will sell itself.” But researchers in other fields are less confident that the innovation will be useful for them. Chief among the roadblocks to software reform is the ubiquitous FORTRAN programming language. Although FORTRAN has evolved considerably since its introduction in the 1960s, the world out- side science has passed it by. The best programming tools are now being written for languages such as C and the Defense Department-standard Ada. Yet vast libraries of code written exclusively in FORTRAN serve as the foundation for countless scientific programs. It’s a dinosaur, scientists agree, but one that is unlikely to become extinct. “What will be the niost popular scientific language in the year 3000?” NSF’s Wulf asks facetiously. “I don’t know, but it will be called FORTRAN.” Experts say that scientists are paying a high price for their reluctance to wean themselves from FORTRAN. Many sophisticated programming tools already available for other languages remain tantalizingly out of reach. “Interface builders”— which allow programmers to design how their software should look on the screen, then automatically generate the appropriate code—are becoming common in the business community. And even the common Apple Macintosh computer now comes equipped with HyperCard, a program that allows even non-programmers to design sophisticated software that might take a scientist months in FORTRAN. — At DARPA, the Information Science and Technology Office (ISTO) will spend some $20 million this year on projects that fall within the generic category of Computer Aided

New Initiatives Aim To Emancipate 'Scientist-Programmers'

Interested in reading more?

Become a Member of