innovators but promote a rapid evolution of their product lines.

Is there a way to protect innovators without strangling further innovation? Congress could create a separate protection system for software, taking elements from both patent and copy

Commentary by Esther Dyson

ware, that would still leave the ong nator of a language with a valuable lead of several years over its rivals. It would, to fall back upon the automotive analogy, let inventors patent gear arrangements but not the idea of hav ing a car with a stick shift. ■

For New Year's, you resolved to exercise regularly, count to ten before losing your temper, and take a serious look at computer-aided software engineering (CASE). Have you been keeping all your resolutions? If not, you can join the crowd. People shy away from CASE even though they know it's good for them.

Like other disciplines, CASE demands a sacrifice up front-a learning and adjusting process-before the benefits arrive. Those benefits are clear and simple to understand: CASE makes it easier and faster to build software, and makes the programs that result higher in quality, more consistent, and easier to change later on (from paying people every two weeks to twice monthly, for example).

So why don't people do it? Because it's difficult to change habits. Programmers get fixed in their ways. If they write a program with the help of another program, the work may go faster but the resulting code doesn't come out looking just the way they'd do it on their own. And they have an excuse: There are too many products on the market, and most customers want to wait until things shake out and the right choice is clear.

Howard Arnold is president of Sage Federal Systems, a Rockville, Md. company that has made a business out of market confusion by understanding all these tools and using them to develop software. The central fact about computer-aided software engineering, he says, is that it's not a quick fix. The tools have to be used regularly and consistently to pay off, because the benefits in large part lie in the consistency of the results. Like weight or temper control, CASE is really a way of life.

Esther Dyson is editor and publisher of the newsletter Release 1.0.

But the resistance is more than just personal habit. There's politics, too: Middle managers don't want to adopt CASE on the broad scale that's necessary for benefits because they would lose control to a central corporate authority. Top management doesn't want to adopt CASE wholesale because it's risky, and who wants risk? Besides, top managers may not trust their own data processing people to use it right.

What will shake companies out of this lethargy? Seeing their competitors get ahead by using the new technology effectively. Helping a few early adopters (or customers with problems so great they're will ing to take the risk) is the job of Howard Arnold's company. A professional services firm carved out of CASE vendor Sage Software just be fore the latter went public, Sage Federal isn't wedded to tools sold by its former parent. It uses not only products from Sage Software but also Excelerator from Index Technology and Recoder from Language Technology.

Sage Federal, privately held with revenues last year over $20 million, has overcome the natural resistance to CASE technology that infects its customers. Its tools (and its 200 people trained to use them) allow it to build prototype systems quickly, so that clients can get early feedback that something is happening and make sure it's the right thing: Customers frequently don't correctly specify what they want at first.

Keeping the customer's mind focused is a big part of a programmer's

job. He may hear, halfway through an inventory accounting project: "Come to think of it, we have to have the unit prices for the videos in both dollars and yen." It is a universal gripe of programmers that they are aiming at a moving target.

From the prototype, Sage develops a finished system using a standard methodology and assemblyline quality controls that guarantee the finished system will match the prototype.

It all sounds so easy. Why can't everyone do it?

In fact, a lot of people are starting to do it, and it will just take a number of success stories to speed up the process. Traditional software development houses have quietly been using CASE all along (without necessarily touting its benefits to competitors or customers paying full price), and the practice is starting to spread. In particular, many management consulting firms-notably Index Group, Booz, Allen and Arthur D. Little, and accounting firms such as Arthur Andersen, Peat Marwick, Coopers & Lybrand and Price Waterhouse-are doing systems development for customers and starting to use CASE tools to do so. Arthur Andersen, for example, used its own CASE tool, Foundation, in building a multimillion-dollar materials management system for Nynex.

Meanwhile the widespread adoption of CASE could have an interesting effect on the balance of power between software specialists and management consultants. The more automatic the development process, the more the value-added shifts toward consulting rather than development skills-if the consultants can use the tools properly. The tools make the coding automatic, but there are no tools yet to automate the understanding of business problems that the consulting firms bring to the party. ■

Ronald Bracewell, a professor of electrical engineering at Stanford who developed a new way to calculate what is known among mathematicians as a Fast Fourier Transform, said that without patent protection future research would be inhibited.

"By the time you've spent six months writing 10,000 lines of computer code, you know darn well it isn't something nature put there," he said. "The world of industry is demanding protection for what they very well know is proprietary information. Without the protection of patents, the incentive to get into the hard work of development would be lost."

Mr. Bracewell's algorithm embodies a technique that the university said could double the speed of some complex calculations, regardless of the kind of computer used. The new formula has potential applications in almost any instrument that processes signals, from radar to medical imag ing, and in systems that process large volumes of data.

The formula is at the heart of a computer system that A.T.&T. says can be used to route telephone calls more efficiently, help airlines coordinate their planes and supplies, and even help banks invest their reserves.

Mathematical algorithms are stepby-step procedures to solve complicated numerical problems and usually employ novel strategies that can be translated into computer code to solve problems.

Computer chips, for example, use algorithms to sample music signals and code them into digital information stored on compact disks. Algorithms are also central to computers that imitate the sound of a piano, analyze airflow over the wing of a jet and route telecommunications traffic.

Mathematics may

'become poorer as mathematicians become richer.'.

Although computers have long been considered patentable inventions, the mathematical algorithm has not. Like Newton's discovery of the law of gravity, such expressions have traditionally been viewed by courts as descriptions of eternal relationships, which can be discovered but never owned.

A patent, by contrast, is the right awarded to the inventor of a new device or process that allows him or her to prevent anyone from using the invention without permission for 17 years.

The debate turns on whether one views algorithms as statements about universal relationships or as new tools invented to solve concrete problems.

As early as 1939, the Supreme Court held that "scientific truth, or the mathematical expression of it, is not a patentable invention." In 1972, the Supreme Court became more explicit, rejecting the patent for an algo rithm used to convert decimal numbers to the binary code used in a computer. The Court ruled that an "algorithm, or mathematical formula, is like a law of nature, which cannot be the subject of a patent."

But in 1980 the Supreme Court ruled that living bacteria altered by genetic engineering could receive a patent, because the bacteria had not previously existed in nature. The Patent Office has since used the ruling to issue a patent for a genetically altered mouse.

The Supreme Court first eased its stand on mathematical expressions in 1981, by noting that the mere presence of an algorithm should not render an invention unpatentable. "It is now a commonplace that an application of a law of nature or mathemati cal formula to a known structure or process may well be deserving of patent protection," the Court ruled. The case, known as Diamond v. Diehr, involved a computerized system for curing rubber that used an equation in the process.

Subsequently, lower courts and the Patent Office have applied the ruling to accept increasingly abstract applications. In 1982, the Federal Court of Customs and Patent Appeals ruled that an algorithm could still be protected as long as it could be "applied in any manner to physical elements or process steps" of the invention.

Guided by such rulings, the Patent Office is now approving patents for inventions in which an algorithm constitutes most of the claim and applications are described only generally.

AT&T.'s patent last May was titled "Methods and Apparatus for Efficient Resource Allocation." What it describes, however, is a purely mathematical model, based on new insights about the way to juggle thousands of variables in search of opti mal combinations of "users" and "resources." The variables can be anything, and the patent claims broad rights for any use of the algorithm in telecommunications, manufacturing, data processing and information handling

The Bracewell patent actually says it covers a "special purpose comput. er," but the patent includes only minimal descriptions of any equipment and notes that even these are "preferred embodiments," examples of equipment that do not exhaust the scope of the claims. The "comput er's" only stated purpose is to execute the algorithm.

Patent attorneys say that a patent `should not prevent academic use of an algorithm, because the patent discloses the invention in a public document that anyone may study. Indeed, the A.T.&T. patent specifically waives rights to any academic uses of its algorithm.

"When Newton discovered gravity, he discovered a law of nature, something that had always existed and certainly we wouldn't want him to be able to patent that," said Henry T. Brendzel, a senior attorney for Bell Labs. "On the other hand, if someone takes the knowledge of gravity and invents a rocket, why shouldn't we let the guy get a patent?"

But most of the algorithms developed for computer science, including the forerunners to both the Bracewell and the AT&T. formulas, were developed in the last 25 years and were never patented.

L. Thorne McCarty, professor of law and computer science at Rutgers University, said: "What if the court had gone the other way in 1964 or 1965, and we had freely granted patents on algorithms? All of the basic texts" in computer science "would be under patent, and that is perhaps worrisome."

Those opposed to the patents are concerned that academic mathemati cians will not share ideas among themselves if they are awaiting pat ent approval. In theory, every contributor to an invention must be listed on a patent, and that potential for tur. moll and conflict is worrisome.