In response, Congress was lobbied heavily to change the law and once again allow "functional claiming." After all, it was a lucrative and powerful tool for controlling competition in new markets. Congress introduced the Patent Act of 1952, which reinstated the ability to make functional claims, but imposed conditions on how far they could go. The crucial part of the statute is section 112(f), which says it's OK to write a patent claim that describes a new use for an existing thing in functional terms. The claim has to identify an established "means" -- a thing to be used in a novel way -- and can describe a new function for that item.
The reintroduction of functional claiming on these terms wasn't the disaster one might expect. The new law worked well, greatly reducing the scope of patents claiming ownership of a function; when they were phrased in this way, they could only affect that function in association with the object described in the patent. According to Lemley, patent lawyers in many fields avoid this sort of "means-plus-function" claim language and instead use other linguistic tricks in their attempts to file broad patents.
But in 1952, no one anticipated the problem that would result when this language was applied to software. At some point, a clever patent attorney realized that if "a computer" was used as the "means," it would present no obstacle to the scope of the "functions" that followed it. While "a computer" sounds specific to a judge, the consequences of Turing completeness mean any computer could be used as the "means." As a consequence, section 112(f) of the Patent Act was treated as a gateway to patenting just about any idea imaginable in software. Lemley says:
A patentee who claims "means for calculating an alarm limit" is invoking the limits of section 112(f), and the claim will accordingly be limited to the particular software algorithm or implementation the patentee described in the specification. But if the same patentee claims "a computer programmed to calculate an alarm limit," courts today will read the term "computer" as sufficient structure and will understand the claims to cover any computer that can calculate an alarm limit, however the calculation is programmed.
Taking the statute seriously
The result of this "innovation" was an enormous proliferation of software patents, followed by the eventual emergence of companies whose only role is to extract money from genuine innovators using threats of patent suits. The paper cites Bessen and Meurer's estimate that patent trolls have cost the economy over $500 billion so far. Lemley estimates that tens, perhaps hundreds of thousands, of patents use this conceptual trick in the field of software to gain patents on ideas -- the very act we're led to believe is impossible.
Lemley surveys a range of solutions others have proposed to this problem, from abolishing software patents (he contends it would be impossible because of the lack of a rigorous definition of "software patent") to demanding publication of algorithms. But Lemley asserts there is a far simpler solution. He observes that in other areas of patent law, the courts have required that the "preexisting thing" justifying the use of the "means-plus-function" structure to make a patent claim "must be more than mere window-dressing."