Sydney Law Review
|
|
Home
| Databases
| WorldLII
| Search
| Feedback
Sydney Law Review |
|
ANDREW CHRISTIE AND SERENA SYME[*]
The enormous growth that has occurred in recent times in patenting software related inventions is most evident in the US. Between 1990 and 1997, the number of software patents issued annually in the US increased nearly ten-fold, from approximately 1300 to about 11000.[1] Whilst the absolute number of software related patent grants in Australia is relatively small in comparison to the US, a similar rate of growth seems to have pertained in this country.[2]
This phenomenal growth has not been greeted with enthusiasm by all in the software industry. Even as early as 1990, some software producers were expressing the view that many “bogus” patents were being granted.[3] In response to this industry disquiet, the US Patent and Trademark Office (USPTO) called a Public Hearing in 1994 to gauge the reactions of software companies and their legal advisers.[4] While the debate was wide-ranging, it tended to separate along vocational lines, with the majority of lawyers coming down in favour of patent protection, and a high proportion of the software companies expressing opposition to it.[5]
Despite the divided view of the stakeholders, the trend in the US and Australia is towards a more liberal application of the requirements for patentability to software-related inventions. This is shown by the approach of the courts,[6] and the subsequent adoption of new examination guidelines by the patent offices,[7] of those two countries. It would seem, however, that the position in the UK is somewhat different. A UK Patent Office report issued in late 1994[8] disclosed that there was “remarkable unanimity” within industry in favour of abolishing the legislative prohibition on software patents.[9] Nevertheless, the clear trend of the UK cases has been to deny the patentability of software-related inventions.[10]
To properly evaluate whether inventions in the software field can and should be patented, it is necessary to address the specific issue of patenting algorithms. This is because algorithms are the building-blocks or “back-bones” of computer programs, and as such are the essence of the subject matter for which the contentious patents are being sought.[11] This paper considers the availability and scope of patents for algorithms in Australia. It discusses whether algorithms can properly be regarded as inherently patentable subject matter.[12] It also explores the application of the patent law requirements of novelty, inventive step, utility and sufficient disclosure to algorithmic inventions.[13] In addition, the application of principles of patent infringement to algorithm patents is canvassed.[14] Finally, consideration is given to the practical implications of patenting algorithms, and to the extent it is desirable and feasible, in law and practice, to reform Australian patent law as it applies to algorithms.[15]
In many discussions about the patentability of software, little attention is paid to the essence of the subject matter – namely, the algorithm. Without an understanding of the nature of an algorithm, and its role in software, it is not possible to grasp and to address properly the many issues raised in regard to protection of this technology. It is for this reason that the article begins with an examination of the characteristics and function of algorithms. The word “algorithm” has mathematical origins,[16] but is used in a variety of contexts. At its most general, “algorithm” is used to describe a process or set of rules (even a type of “recipe”) for carrying out a procedure, function or task. Algorithms are probably most noted for their association with computer programming, but are also important in mathematics and in engineering, especially manufacturing processes.[17] In fact, all step-by-step processes involve algorithms in the broad sense.[18]
An algorithm must possess five essential features – definition, finiteness, input/ output, effectiveness and order.[19] These features are common to all varieties of algorithm; from a cooking recipe to a computer programmer’s algorithm.[20] The feature of definition relates to the certainty of description of an algorithm’s steps. The degree of certainty of description required of an algorithm depends to a large extent on the user of the algorithm. Algorithms to be used by mathematicians may not need to be very rigorous, because a large amount of background knowledge and experience can be assumed. Most computers and robotic devices, however, lack this knowledge, so computer programmers require their algorithms to be highly definite, in the sense that they must be specified rigorously and without ambiguity.[21]
The feature of finiteness relates to the need for an algorithm to specify a route to a final step in all possible cases to be handled by it. Thus algorithms need to terminate after a finite number of steps, and each step must be able to be accomplished in a finite (but not necessarily short) amount of time. The need for an algorithm to have input and output follows from the fact that it is designed to be useful. Thus, an algorithm must be capable of accepting zero or more starting conditions,[22] and must produce one or more meaningful results.[23] How can it be determined whether what an algorithm produces is the desired output? We can only really be sure of this if each of the steps are simple enough to be performed manually, even if such performance would be tedious and inefficient. An algorithm which can be performed in this way is said to be effective. Effectiveness is more evidently important when dealing with complex numerical algorithms; unless each step of the operation can be manually mimicked, we cannot be certain that the correct procedure is being effected. Where the operator is a human rather than a computer, the requirement of effectiveness is necessarily met.
The final feature of an algorithm is sometimes taken for granted but is nevertheless crucial; the algorithm must be performed according to a specified order. Order is essential to all types of algorithm; an algorithm would be less than serviceable if its steps were described in no particular order, even if an experienced operator could work out the correct sequence for him or herself.
As will seen below,[24] courts have consistently referred to algorithms as “mathematical algorithms”, without troubling to define what, if any, is the meaning or limitation of the word “mathematical” in that phrase. Indeed, in a number of cases the courts have equated all algorithms with mathematical algorithms, and hence have incorrectly used the two terms interchangeably. What then is the correct meaning of a mathematical algorithm?
Once it is established that an “algorithm” is an ordered and highly definite method for solving a problem, it would seem to follow that a “mathematical algorithm” is an algorithm which solves a mathematical problem. The phrase “mathematical problem” is used here in a narrow sense, to mean a problem solely concerned with mathematical concepts. It does not include a general (nonmathematical) problem the solution of which may involve mathematical concepts, but it does include a specific problem concerned solely with mathematical concepts even though that problem may form part of a larger, more general (nonmathematical) problem. As is demonstrated in Section 3.4.2 below, this approach is not inconsistent with that taken by Australian courts.
When courts come to consider the types of algorithms used in the computer context, they often incorrectly equate them with computer programs or software.[25] This confusion makes it more difficult to understand clearly the issues surrounding the patenting of algorithms. It is therefore important to make clear the conceptual relationship between these two subject matters.
As a method for solving a problem, an algorithm, standing alone, is nothing more than a concept or idea. In order to be commercially useful, an algorithm must be expressed in some form of language or code.[26] In the case of an algorithm designed for use solely by a human operator, the tangible embodiment of the algorithm is a written description in a human language of the set of instructions of which it is comprised.[27] As we are used to expressing ideas in a human language, the distinction between the concept of the algorithm and the expression of the steps of the algorithm is a fine one.
The distinction is clearer in the example of an algorithm designed for use by a mechanical device, rather than a human operator. In that situation, the algorithm must have a tangible embodiment beyond simply a written description in a human language of the set of instructions of which it is comprised. A plant or machinery must be “built” by reference to each step of the algorithm, so as to allow every step to be performed in the appropriate order by the non-human operator. Here the algorithm is being used as the “back-bone” of the mechanical device, and is identifiable separately from the device itself.
In the current day, the most common user of an algorithm is a computer, and so the typical tangible embodiment of any algorithm is a computer program. A computer program is a complete series of definitions and instructions, conforming to the syntax of a given computer language, that when executed on a computer will perform a required task.[28] By contrast, an algorithm is just the method for solving a problem; it may be suitable for implementation in a computer program, but is not itself a computer program.[29]
Euclid’s famous algorithm can be used to illustrate this distinction. This algorithm is designed to find the greatest common divisor of two positive integers, and is set out in various forms in the Appendix to this paper. In the first form, it could be used by a human with an understanding of English and of basic mathematical concepts.[30] But if the algorithm was to be used by a computer, it would need to be expressed in a different way; that is, it would need to be embodied in a computer program. Two computer programs embodying this algorithm (in the languages Pascal and C, respectively) are set out in the Appendix.[31] The algorithm is clearly distinct from each of the programs embodying it. The algorithm is expressed in far more detail in the programs than in its English language description, and the programs contain elements which would not be necessary if the algorithm was being performed by a human operator, such as defining the variables to be used as integers, or displaying the final result on the screen.
As will be appreciated from this example, a computer program is built by expressing each step of the algorithm in a computer language. Programs in different programming languages may be created by reference to the same algorithm; while the programs produced may appear very different, they will perform broadly the same function.[32] Thus, in relation to software-related inventions, it is the algorithm or algorithms embodied in the software, rather than the software itself, for which a patent is likely to be sought.
Having determined the nature of an algorithm, it is necessary to consider how the various requirements of the patent legislation apply to an invention involving an algorithm, commencing with the threshold requirement that an invention be inherently patentable subject matter.
Section 18(1) of the Patents Act 1990 (Cth) (hereinafter Patents Act) states:
A patentable invention is an invention that, so far as claimed in any claim: (a) is a manner of manufacture within the meaning of section 6 of the Statute of Monopolies ....
A similar provision was contained in the previous legislation, the Patents Act 1952 (Cth),[33] and was construed by the High Court in National Research & Development Corp v Commissioner of Patents (hereinafter NRDC).[34] In that case, the court warned against the use of “manufacture” to reduce the question of patentability to one of verbal interpretation.[35] Their Honours noted that the purpose of section 6 of the Statute of Monopolies was to encourage national economic and social development by granting letters patent to the appropriate subject matter. In this context, “manufacture” was used as a label for the general class of patentable inventions; consequently, any attempt to precisely define “manufacture” was bound to fail.[36] The court stated that the scope of patentability should not be restricted by reference to the strict nature of a “manufacture”, but held that the legislative provision involved a question of public policy – being whether the invention is “a proper subject of letters patent according to the principles which have been developed for the application of section 6 of the Statute of Monopolies?”.[37] That provision expressly provided that a patent should only be granted where to do so was “not contrary to the law nor mischievous to the State, by raising prices of commodities at home, or hurt of trade, or generally inconvenient”.[38] Their Honours concluded that, to be patentable, a process “must be one that offers some advantage which is material, in the sense that the process belongs to a useful art as distinct from a fine art”.[39]
In NV Philips Gloeilampenfabrieken v Mirabella International Pty Ltd (hereinafter Philips),[40] the High Court considered the construction to be given to section 18(1) of the Patents Act. It held, by a majority, that the opening words of section 18(1) (“a patentable invention is an invention that ...”) imported into that section an additional requirement of inherent patentability – namely, that the subject matter of the claim must not, “on the face of the patent specification, [lack] the quality of inventiveness necessary for there to be a proper subject of letters patent under the Statute of Monopolies”.[41] A claim will lack this quality where it is for “nothing more than a new use of an old product”[42] – that is, where it is for a new use of a previously known subject matter in circumstances where the previous use of the subject matter is merely “analogous to” the claimed new use.[43]
The majority of the High Court in Philips stated expressly that this “threshold requirement of inventiveness”[44] is a requirement in addition to the other patentability requirements of novelty and inventive step,[45] each of which are discussed below in sections 4.1 and 4.2, respectively. The High Court majority found such a construction of section 18(1) to be consistent with the construction given by the High Court in NRDC to the equivalent provision on inherent patentability in the Patents Act 1952 (Cth).[46] Whilst some of the reasoning of the majority in Philips may be criticised,[47] the general principle enunciated by it was subsequently confirmed by a majority of the High Court in Advanced Building Systems Pty Ltd v Ramset Fasteners (Aust) Pty Ltd.[48]
The question that arises is how the threshold requirement of inventiveness will operate in relation to a patent claim for an algorithm. It is noted later that one class of patent claim for an algorithm is to a computer-automated form of a previously used algorithm, where the previous use of the algorithm was in a field other than the one to which the claim relates.[49] In this situation, the principle enunciated by the majority in Philips would seem to have particular relevance, and could operate to preclude the inherent patentability of the algorithm if the fact of the previous use of the algorithm was disclosed on the face of the specification.
There has not been a great deal of Australian authority on the question of the inherent patentability of algorithms; the Federal Court’s 1992 decision in International Business Machines Corporation v Commissioner of Patents (hereinafter IBM)[50] was the first major consideration of this issue. The claim in IBM concerned a method for producing a curve image on a computer screen from a set of control points. The method was to be embodied in a computer program, and was ordered, highly definite, and could be manually calculated. It accepted input control points and produced the completed curve. Hence the claim clearly recited an algorithm.
The delegate of the Commissioner of Patents applied the so-called “Freeman- Walter-Abele two-step test”,[51] developed in a trio of US cases decided in the late 1970s and early 1980s.[52] The test is as follows:
First, it must be determined whether the claim directly or indirectly recites an ‘algorithm’ in the Benson sense of that term [ie, a procedure for solving a mathematical problem], for a claim which fails even to recite an algorithm clearly cannot wholly pre-empt an algorithm. Second, the claim must be further analysed to ascertain whether in its entirety it wholly pre-empts that algorithm.[53]
Under this test, the claim in IBM was rejected by the delegate, on the basis that it recited a mathematical algorithm and wholly pre-empted that algorithm because it was not limited to a field of use.[54] On appeal, Burchett J emphasised that the applicant should be given the benefit of any doubt.[55] In referring to the argument that the claim was not limited to a field of use, he stated that the claim and the preceding specification should be read as a whole; if that was done, there could be no doubt that the claim was limited to “the operation of computers”.[56]
Burchett J defined an algorithm as “a procedure for solving a given type of mathematical problem”,[57] and appeared to proceed on the basis that the claim recited an algorithm of this type (that is, a mathematical algorithm).[58] His Honour referred to NRDC, and noted the High Court’s conclusion therein that, to be patentable, the invention must belong to the useful arts as distinct from the fine arts.[59] To belong to the useful arts, an invention must have a commercial effect. It was this requirement that allowed patent law to distinguish between the discovery of a mere abstract principle of science (which is unpatentable), and the making of a patentable invention. Burchett J concluded that the production of an improved curve image is commercially useful in the field of computer graphics, and that therefore the invention was the proper subject matter of a patent.[60] Burchett J did not expressly reject the Freeman-Walter-Abele two-step test; rather, he returned to NRDC, and drew his conclusions from that case.[61] Having drawn his conclusions, he then noted that he did not believe that the US case law precluded this view.[62]
The IBM case was followed soon after by the decision of the Full Court of the Federal Court in Ccom Pty Ltd v Jiejing Pty Ltd (hereinafter Ccom).[63] This case concerned a counter-claim, in infringement proceedings, that a petty patent for a computer processing apparatus which assembled text in Chinese characters was invalid. The relevant claim recited a database of Chinese characters and their graphic representations, and a search and retrieval procedure. The invention required the operator to use a special keyboard to type in character-stroke type categories, each time narrowing down the number of characters which met the criteria, until only a small number remained. These characters were then retrieved from the graphic data storage and displayed; the appropriate one was selected and retained. The essence of the invention was thus an algorithm.[64]
At first instance,[65] Cooper J applied the fine arts versus useful arts distinction drawn in IBM. He concluded that because the algorithmic search and retrieval procedure could be undertaken as a mental process without the aid of a computer, it is the “product of human intellectual activity lying in the fine arts and not the useful arts”.[66] In his view, such a process is not patentable subject matter, impliedly because it is more suited to copyright protection than patent protection.[67]
The case went on appeal. In considering whether the word-processing apparatus was patentable subject matter, the Full Court of the Federal Court followed NRDC, referring to it as the “watershed” case on this issue.[68] Their Honours considered the High Court’s interpretation of “manner of manufacture”[69] and held that, post-NRDC, the question of what is a manner of manufacture requires “a decision as to what properly and currently falls within the scope of the patent system”.[70] In forming such a decision, the court stated that the appropriate subject matters of a patent grant are artificially created states of affairs, that are of practical use and have an economic significance. The computer processing apparatus claimed in the patent was an artificially created and useful state of affairs in a field of economic endeavour, and was therefore a manner of manufacture for the purposes of the Australian legislation.[71]
This appellate decision clearly follows in the conceptual foot-steps of NRDC and IBM, revealing a distinct new approach to the question of patentability of algorithms. As a result of this change in approach, the Australian Patent Office developed new guidelines based on IBM, and an examination of these provides a good indication of the practical consequence of these cases.
Pursuant to the Australian Patent Office guidelines for examination, the Manual of Practice and Procedure, as amended in March 1994 (“Guidelines”),[72] examiners are required to apply the following test in determining whether an algorithmrelated invention is patentable subject matter:
Does the invention claimed involve the production of some commercially useful effect? [73]
This test supersedes the Freeman-Walter-Abele two-step test.[74] In elaborating on this new test, the guidelines state that a mathematical algorithm alone will not be patentable, because it does not produce a commercially useful effect; it must be implemented to produce such a result.[75] Some examples of commercially useful effects are an improved image (from IBM), or a more efficient operation of a computer.[76] If the mathematical algorithm is used in a computer program, it is patentable as long as some commercially useful effect is produced,[77] with one exception – the invention will not be patentable if the only novel element is the mathematical algorithm.[78]
How do the post-IBM Guidelines accord with the established case law? This section of the paper contains some consideration of the practical impact of the new guidelines, and the way in which issues and concepts which have been central to this area, such as the “laws of nature” exclusion, have been affected by the developments.
3.4.1 Commercial Use, and the Fine Arts v Useful Arts Distinction The test of patentability applied by the Patent Office now requires the production of a commercially useful effect.[79] Does this requirement of a “commercially useful effect” correspond to the High Court’s assertion in NRDC that, to be patentable, a process must belong “to a useful art as distinct from a fine art ... [ie] its value is in the field of economic endeavour”?[80] This comment was quoted by Burchett J in IBM,[81] which tends to suggest that the tests are identical; if an invention has a commercially useful effect, its value must lie in economic endeavour, and it is consequently a useful art.[82] While the categorisation by Cooper J in Ccom of the commercially promising computer word-processing apparatus as part of the fine arts might suggest that there is some distinction between the tests,[83] this approach was rejected by the Full Court of the Federal Court on appeal.[84] Hence, it appears that the fine arts and useful arts distinction is really just another way of phrasing the requirement of a commercially useful effect, not a separate requirement.
Prior to the introduction of the new Guidelines, the Patent Office adopted the Freeman-Walter-Abele two-step test for the inherent patentability of algorithms. Under that approach, algorithms were often regarded as inherently unpatentable.[85] Now that the test for inherent patentability centres on whether the claimed invention produces a commercially useful effect, what types of inventions will be excluded from patentability? It seems unlikely that a patent will be sought for a product with no commercially useful effect, or at least for a product for which no commercially useful effect could at least be alleged.[86] Either these developments have rendered trivial the test for inherently patentable subject matter, or the requirement of “commercially useful effect” is more stringent than it might at first appear.
In practical terms, the Patent Office is likely to follow the plain wording of the Guidelines, resulting in the rejection of very few patent applications on the ground that they recite unpatentable subject matter. Indeed, it is understood from informal discussions with the Patent Office that far fewer applications have been rejected on the ground of inherent unpatentability since the IBM case. The current test appears to be one of the least stringent tests in the developed world.
3.4.2 The Case of Mathematical Algorithms
The Guidelines state that a mathematical algorithm per se is not patentable. It is submitted that this guideline is an accurate (indeed redundant) statement of inherent patentability, so long as the phrase “mathematical algorithm” is given the narrow meaning proposed previously[87] and the significance of the per se limitation is noted. This is because a mathematical algorithm in its narrow sense per se would not produce a commercially useful effect. It could only produce a commercially useful effect once given some form of tangible embodiment.[88]
The Guidelines also indicate that where an algorithm is used in a computer, it will be patentable as long as a commercially useful effect is produced.[89] This guideline again goes no further than the general test of inherent patentability. Thus far, there is no indication that it is necessary to be able to distinguish a mathematical algorithm from a non-mathematical algorithm, other than for the purpose of applying the general principles of patentable subject matter.
However, the Guidelines then proceed to specify that a mathematical algorithm cannot confer novelty on an invention that is not otherwise novel.[90] This is an intriguing provision; while it is included in a section discussing patentable subject matter, it appears more relevant to the question of novelty.[91] Such considerations aside, it is probably only in this paragraph of the Guidelines that the legal principles applicable to mathematical algorithms differ from the principles applicable to algorithms generally. Consequently, it continues to be necessary to identify mathematical algorithms, in order to ensure that novelty resides outside of them for the claimed subject matter to be patentable.
How has the concept of “mathematical algorithm” been interpreted in Australia? In the IBM case, Burchett J, referring to the first US Supreme Court case on this topic, Gottschalk v Benson,[92] defined an algorithm as a “procedure for solving a given type of mathematical problem”.[93] His Honour also referred to the indicia presented in the later US case of Re Walter,[94] the most crucial being whether the “end-product of the claimed invention is a pure number”.[95] Read literally, it would appear that his Honour was making the same mistake as his American counterparts, in that he was wrongly treating mathematical algorithms as definitive of all algorithms. The better view, however, is that Burchett J was in fact defining the sub-set of algorithms which deal solely with mathematical concepts, and hence which are properly regarded as “mathematical algorithms”. If this is accepted as correct, it follows that a mathematical algorithm would be regarded by the Australian courts as a method for solving a mathematical problem and which produces a numerical solution.[96] Accordingly, an invention will be unpatentable if its only novelty lies in this type of mathematical relationship.[97]
Nevertheless, the definition of “mathematical algorithm” remains incomplete. No Australian court has yet sought to define a “mathematical problem”. As discussed previously, the most likely definition excludes general problems that happen to involve some mathematical concepts, but includes narrow mathematical problems that are applied to non-mathematical situations. The inclusion or exclusion of the latter category is crucial, as most mathematical algorithms for which patents are sought will be of this type.
3.4.3 The “Laws of Nature” Exclusion
Under US patent law and the law of the European Patent Convention 1973 (EPC), various types of subject matter are excluded from patentability by virtue of falling within one of a number of established categories or “pigeon-holes” of inherently unpatentable subject matter. Examples of such pigeon-holes, enunciated by the courts and the USPTO Examination Guidelines in the case of the US, and by the express provisions of the legislation and the interpretations thereunder in the case of the EPC, are methods of mathematical calculation and other mental processes,[98] business and financial schemes,[99] and scientific theories and laws of nature.[100] Inventions involving algorithms have often been rejected by courts in the US and under the EPC on the ground that algorithms are like “laws of nature”.
The exclusion of “laws of nature”, along with the conceptually related exclusions of mental processes and business schemes, from inherent patentability has traditionally been part of Australian patent law.[101] Now, however, it appears that the requirement of a commercially useful effect may have removed the need for the “laws of nature” exclusion in Australian law. In the Ccom case, the Full Court of the Federal Court drew a distinction between “the discovery of laws or principles of nature and the application thereof to produce a particular practical and useful result”.[102] In theory this statement maintains the “laws of nature” exclusion, but in practice it goes no further than to require a commercially useful effect. As long as there is such an effect, the necessary application exists and the laws of nature exclusion will not apply. This view receives support from Burchett J’s statement in IBM that it is by “the production of some useful effect that patent law has distinguished ... between the discovery of a principle of science and the making of an invention”.[103] Consequently, it can be assumed that where a commercially useful effect can be discerned from the terms of the claim, the claim concerns an invention as opposed to a principle of science or law of nature, and is therefore inherently patentable subject matter.
So, despite the fact that the “laws of nature” exclusion is still recited in all of the relevant judgments, practically it seems to be subsumed within the requirement of a commercially useful effect, as set out in the Guidelines. These developments allow the rejection of certain undesirably broad claims, without requiring the forced classification central to the “pigeon-hole” approach. Not only is this a more conceptually honest basis, it removes the need to maintain the “laws of nature” exclusion as a separate principle of Australian law, at least as far as algorithms are concerned.
As was noted above, however, it is arguable that the “commercially useful effect” test has now become so broad that it will exclude very little.[104] Consequently, most if not all of the “screening” of patent applications is left to the requirements of novelty, obviousness, utility and disclosure, which have not previously been of great importance in cases involving algorithms. The application of these requirements to inventions involving algorithms is discussed below.
There are several key requirements of a patentable invention in addition to the requirement that it constitute inherently patentable subject matter. The claimed subject matter must also be novel,[105] it must involve an inventive step,[106] and it must be capable of adequate disclosure to the public in the patent specification.[108] These requirements have rarely received much consideration by courts in their rulings on the validity of a claim to an algorithm; if a claim has been found to be invalid, invariably it has been on the ground that it does not recite a manner of manufacture (ie, is not inherently patentable). Thus there is very little discussion of these other requirements in the relevant literature. But now that rejection of a claim to an algorithm on the ground of lack of inherent patentability appears less likely, it is necessary to investigate the issues that may arise concerning these other requirements.
The assessment of novelty requires an investigation as to whether the invention has already been made publicly available. Put in general terms, an invention will lack novelty if, when compared against the state of the art at the time of the application for a patent, it is apparent that the subject matter of the invention has already been disclosed in some manner to the public.[109]
4.1.1 Establishing the Prior Art
The ability to establish the state of the art at the date of the application for the patent (“the prior art”) is central to the test of novelty. However, maintaining a record of those algorithms which form part of the prior art is a great problem in practice. The problems are manifold. For fear of having their patent application rejected, it seems applicants in earlier times tended to avoid using the words “algorithm” or “computer program” in their claims, and attempted to “disguise” their computer programs in “hard-wired” form as integrated circuits.[110] Furthermore, in computer science, a large amount of the prior art is contained only in textbooks or on electronic bulletin boards and newsgroups, which are not generally part of the Patent Office’s search procedures.[111] The US Commissioner of Patents and Trademarks has referred to much of the prior art as “folklore”.[112]
These problems seem to be one of the reasons why US courts were reluctant to grant patents for software-related inventions for so long.[113] The Australian Patent Office has acknowledged unofficially the difficulty of determining the prior art for the purposes of judging the novelty of an algorithmic invention, which difficulty is due to the diversity of sources of prior art in the computer science field and the rapid nature of development in this area. An additional problem is that of identifying the algorithms in any software that is considered part of the prior art.
Due to the combined difficulties of determining what software is in the prior art and identifying the algorithms in that software, the test of novelty tends to be used as a filter to eliminate clearly invalid patent applications, rather than as a definitive analysis.
4.1.2 Possible Changes to Searching the Prior Art
A number of measures have been suggested to ameliorate the difficulties of discovering the relevant prior art software and identifying the algorithms therein. The US Patent & Trademark Office (“USPTO”) has proposed access to commercial and public databases of software technology, assistance from the private sector in compiling the USPTO internal database, and provision for the submission of prior art by individuals.[114]
A further way of establishing the prior art was raised at the Public Hearing conducted by the USPTO in January 1994. Patent attorney Robert Sabath suggested that patents should be granted more freely, to encourage applications which would form part of the prior art for the future.[115] But given the acknowledged inadequacies of the current examination process, this seems undesirable. It is hard to think of an approach that would be more restrictive to the software industry than to tie up all the prior art with liberal patent grants. Needless to say, none of the computer scientists who appeared at the Hearing espoused this view.
The objective behind the requirement of inventive step is to screen out new products and processes which differ from the prior art in only a relatively insignificant way. An invention will not involve an inventive step if it would have been obvious to a person skilled in the relevant art at the time of the application for the patent.[116]
What types of inventions may be excluded from patentability by the requirement of inventive step? Generally, this requirement excludes inventions that are mere “workshop improvements” (non-inventive or routine improvements),[117] and those developed as a result of an approach which was “obvious to try”.[118] In determining what these terms mean in the context of algorithms, it is helpful to divide the inventions in this field into two categories:
(i) inventions which build on or improve algorithms previously used in the field to which the claim relates; and
(ii) inventions using algorithms which are new to the field to which the claim relates.
Analysis of the first category is relatively straightforward. It seems likely that if an algorithmic invention which builds on or improves a previously used algorithm is to be found obvious, it will be on the basis that it is just a “workshop improvement”, involving only basic or insignificant alterations to the previously known and used algorithm.
What types of algorithmic invention within the second category may be considered to lack an inventive step? This category of invention, involving as it does the concept of a “field” of knowledge, is more problematic. This is because the requirement of inventive step is determined by considering what would be obvious to “a person skilled in the relevant art in the light of the common general knowledge” in Australia before the priority date of the claim.[119] This formulation of the inventive step test limits in two ways the range of information against which the obviousness of the invention is determined. First, the information must be part of the “relevant art”; and secondly, the information must be part of the “common general knowledge” of a person skilled in that art in Australia. That is to say, apart from the limited further information which may be taken into account pursuant to section 7(3), the test of inventive step under section 7(2) of the Patents Act mirrors that under the previous legislation as interpreted by the High Court in Minnesota Mining & Manufacturing Co v Beiersdorf (Aust) Ltd (hereinafter 3M).[120]
It can be appreciated that the critical issue in determining whether an algorithmic invention involves an inventive step is the “relevant art” of the invention, since it is largely only information forming part of the common general knowledge of a person skilled in that art which is relevant to the determination. In general terms, the courts determine the art in which the relevant person must be skilled by asking to whom the patent specification is addressed. This in turn is a matter of “construing” the specification and the claims, which is a task patents courts undertake on the facts of each patent specification. It is, accordingly, an issue about which it is difficult to generalise.
It is possible, however, to point out the difficulties a court may face in undertaking the task of determining the relevant field of art in relation to algorithmic inventions. Consider, once again, Euclid’s algorithm. The “mathematical” form of Euclid’s algorithm presented in the Appendix[121] has been used for many hundreds of years. With the development of computing, this algorithm was “translated” to computer-automated form by the simple expedient of expressing it in computer code. The mathematical algorithm, whilst undoubtedly well known by mathematicians, may not have been part of the common general knowledge of a computer scientist at the time it was first embodied in a computer-automated form. If claims to an invention being the computer code embodying the algorithm, properly construed, are found to be addressed to a person skilled in the art of computer science rather than mathematics, the computer code may well be found to be not obvious to such a person.
The most recent Australian cases dealing with algorithmic inventions, IBM[122] both concern algorithms of this type – namely, computerautomations of an algorithm arguably well known in a field outside computer science. In both these cases, inventive step was not in issue before the court.[124] It is submitted that the likely reason it was not in issue is that the relevant art would be considered to be that of computer-automation of the area dealt with by the invention (graphics, in the case of IBM; character language processing, in the case of Ccom), and that it was accepted by the parties that the common general knowledge in the respective art did not include the relevant algorithm. Accordingly, it seems to be the case in Australia that a claim to the computerautomated form of an algorithm that is well known outside computer science will usually be found to satisfy the inventive step requirement for patentability.
The Patents Act 1990 (Cth) expressly requires a patentable invention to be “useful”,[125] but does not indicate what constitutes utility. It is clear from case law on the predecessors to this provision, however, that this is not merely a requirement that the invention be susceptible to some practical use, as is the case with the requirement of usefulness under the EPC and the US patent legislation. The requirement of utility under the Patents Act 1990 (Cth) is whether the invention as claimed achieves the result asserted by the patentee in the specification. An invention will not satisfy the requirement of utility if it is possible for another person to come within the claims of the patent by the making of a product or the carrying out of a process, and yet not achieve the benefit of the invention as promised by the patentee in the specification.[126]
It has already been established that an algorithm has the characteristics of definiteness, finiteness and effectiveness. Thus, where an algorithm is presented as the method for solving a particular problem, we can assume that the algorithm will be clearly defined, and that it will reach an appropriate solution within a finite period of time. If the steps of the algorithm are outlined in the specification in sufficient detail, and the alleged benefit is the solution to the problem, it seems unlikely that a claim embodying the algorithm will fail under the Patents Act 1990 (Cth) for inutility.
One of the features of patent protection is that, in return for the grant of intellectual property rights to the inventor, the community receives the benefit of information about the invention by virtue of the publication of the patent specification. The patent legislation requires that the complete patent specification must fully describe the invention[127] – that is, in a manner sufficiently clear to enable the invention to be produced or performed by a person skilled in the art. With some of the newer, very complex technologies for which patents may be sought, such as the genetic engineering of life forms, it can be very difficult to make a sufficiently full disclosure of the invention created by those technologies.[128]
Does a similar difficulty arise for algorithms? On the face of it, this difficulty will not arise. Algorithms are highly definite, which means they are clearly specified. They are also ordered, and so can be fully expressed in prose.[129] Thus algorithms seem ideally suited to full description in a written patent application.
There may be a difficulty, however, in isolating, for disclosure purposes, the algorithm from the embodiment of it which forms the invention. It was noted above that there are many different forms, of varying complexity, in which an algorithm can be embodied. Is the specification of any embodiment of an algorithm sufficient disclosure of the algorithm itself? While disclosure of the embodiment makes the algorithm available to the public for the practical purpose of using the algorithm in that embodiment, it may not make available to the public the algorithm per se. That this may be so is illustrated by the decision of the Technical Board of Appeal of the European Patent Office in Microchip (T 461/ 88).[130] That case considered whether disclosure to the public of an integrated circuit embodying a computer program in machine language that implemented a control procedure (an algorithm) was disclosure to the public of the procedure, so as to rob that procedure of novelty for patent purposes. The procedure itself could not be discerned merely by inspecting the integrated circuit, and there had been no provision to the public of logic diagrams illustrating the procedure. Although the possibility existed for ascertaining the contents of the computer program by complex technical analysis of the circuit, the tribunal found that this was not likely to have occurred given the cost-benefit considerations of doing so. In these circumstances, the Technical Board of Appeal held there was no disclosure to the public of the procedure.
If it is found that disclosure of an embodiment of an algorithm is not sufficient disclosure of the algorithm itself, the consequence is that the public is deprived of the full benefit of the invention, in that it does not learn of the algorithm, and thus is not able to make use of the algorithm for other purposes. It would therefore be necessary to require specification of the algorithm contained in an embodiment of the invention in a sufficiently clear form – ideally in human language – before patent protection could be afforded to that algorithm.
This section of the paper looks at what actions do or should constitute infringement of an algorithm patent. Interestingly, as a result of the understandable preoccupation in the relevant literature and case law with the inherent patentability of algorithms, to date there has been very little considered discussion of this issue.[131]
A patent is infringed when someone, acting without the authorisation of the patentee, does an act in relation to the invention which falls within the patentee’s exclusive rights. The exclusive rights of the patentee under the Australian patent legislation are the rights to exploit, and to authorise or induce another to exploit, the invention.[132] “Exploit” for this purpose generally consists of the right to use a patented process and the rights to make, to hire, to sell and to import a patented product or a product of a patented process.[133]
Because the acts of infringement differ depending on whether the invention is a product or a process, the first issue to consider is whether it is more appropriate to regard an algorithm as a product or a process. The definition of an algorithm presented in Section 2 of this paper describes an algorithm as a method for solving a problem. This suggests that an algorithm is probably most appropriately regarded as a process rather than as a product, for infringement purposes. In support of this view it is noted that whereas a product would usually be commercially exploited of itself, an algorithm is likely to be exploited only as the “back-bone” of an embodiment.[134] That is to say, it is not the set of instructions or commands composing the algorithm which would be commercially exploited (for example, sold), but the program or machine enabling those instructions to be performed. These factors appear to make it clear that an algorithm should not be regarded as a product, but rather as a process.
If the algorithm is properly regarded as a process, there are, in general terms, two types of acts which will infringe a patented algorithm – (i) use of the algorithm, and (ii) making, hiring, selling or importing any product which results from the use of the algorithm. Both of these types of infringing act are considered below.
It is clear that the patentee of an algorithmic invention has the exclusive right to use the algorithm or to authorise someone else to use it. Thus infringement will occur if an unauthorised person undertakes one or other of these activities. What does it mean to use an algorithm? In most cases, use of an algorithm will occur by the running of the machinery or the computer program which embodies the algorithm, so that the steps of the algorithm are performed. The mere act of giving tangible embodiment to the algorithm, however, such as by writing a computer program incorporating the algorithm, usually would not constitute use of the algorithm. This is illustrated by an algorithm for an industrial process, whereby embodying the algorithm requires constructing a special plant. It is clear that construction of the plant does not, of itself, amount to a use of the algorithm; it is only when the plant is used to produce the desired product that the algorithm has been used. It follows that the writing of a computer program would not be a use of the algorithm embodied therein, but that the execution of that computer program would constitute use of the algorithm.
Mere possession of an invention does not, of itself, constitute use of that invention. So to establish infringement of the patentee’s exclusive right to use an algorithmic invention, it would be necessary to prove that there was an actual performing of the algorithm, such as by running an embodiment of it.
It is clear from the provisions of the patent legislation on infringement that the patentee has significant exclusive rights over the use of the product of a patented process. Infringement will be established if it can be proved that there has been an unauthorised production, hire, sale, disposal, use, or importation of such a product.[135] It follows that it is very important to properly identify the product of the patented process. But while it may be easy to determine the product of most patented processes, it may be much harder to do this in the particular instance where the patented process is an algorithm.
What is the product of an algorithm? There appears to be three possible answers to this question – (i) the product of an algorithm is the output of the algorithm; (ii) there is no product of an algorithm (for instance, it is “productless”); or (iii) the product of an algorithm should be deemed to be some entity other than its output. Each of these possibilities is now considered in turn.
5.3.1 The Output of an Algorithm as its Product
The most obvious candidate for the product of an algorithm is its output, as this results from its use. In relation to non-mathematical algorithms, this approach is not problematic. Thus, for example, in relation to the facts of the US Supreme Court case Diamond v Diehr Commissioner of Patents and Trademarks,[136] the product of the process patented therein would be cured rubber. As all manufacturing processes involve algorithms in the broad sense of the term, numerous illustrations could be drawn demonstrating the production of a physical output. In these situations, the algorithm’s output can be appropriately regarded as its product.
There is a difficulty with this analysis, however, in relation to mathematical algorithms. With these algorithms, the output is usually intangible – such as a number, a curved image on a screen or a sorted list of words. To regard the output of a mathematical algorithm as its product would make the commercial exploitation of the outputted number, image or list an infringement of a patent for that algorithm. Is it meaningful to refer to the production, hire, sale, disposal, use, or importation of a number, an image or a list of words? It is submitted that it may not be meaningful to do so, and hence that it may be inappropriate to regard the output of a mathematical algorithm as its product for the purposes of patent protection.
5.3.2 The Mathematical Algorithm as “Productless”
Perhaps, therefore, a mathematical algorithm should be regarded as productless. There is certainly a precedent in Australian patent law standing for the principle that not all patentable processes produce a “product” as that term is used in the definition of “exploit” in Schedule 1 of the Patents Act 1990 (Cth). The case is Rescare Ltd v Anaesthetic Supplies Pty Ltd,[137] where the Federal Court at first instance held that “a method for treating snoring and/or obstructive sleep apnoea in a patient comprising ...” certain steps did not produce a product.[138]
If a mathematical algorithm is regarded as a productless process, there is no product to which the exclusive rights of the patentee extend. Infringement will therefore only occur if an unauthorised person uses the patented mathematical algorithm or authorises someone else to use it. This potentially covers a reasonably broad range of activity, because selling or hiring the program in certain circumstances would constitute authorising someone else to use the algorithm.[139] However, producing, keeping or disposing of a program embodying the algorithm would not constitute an infringement, and neither would offering to sell or hire the program or running the program without using the algorithm, if that were possible. Thus the level of protection of the algorithm is substantially reduced if it is considered to be a productless process. It seems preferable, and arguably more in keeping with the aim of the legislation (discussed below), to seek to find a product for a patented algorithm.
5.3.3 Deeming the Product of a Mathematical Algorithm to be Something Other Than its Output
Can an appropriate product be identified for mathematical algorithms? In answering this, it is helpful to look at the intention of the legislature in conferring protection on the product of a patented process. It will often be difficult, if not impossible, for a patentee to identify the existence of infringement of a process patent other than by reference to the commercial exploitation of the product of that process. Unless the exclusive rights of the patentee cover commercial dealing in the product of the process, the patentee in practice will be powerless to take action against the infringer. The rationale of the legislature in conferring protection on the product of a patented process, is thus to provide a patentee with a means to restrain an infringer from reaping a commercial reward from an act of infringement which the patentee could not prove.[140]
Where the patent relates to a mathematical algorithm, usually it will be impossible to catch an unauthorised person in the process of using the algorithm. It should, however, be possible to identify an unauthorised person who is producing or selling an embodiment of the algorithm, whether that embodiment be some piece of machinery or, as is more likely, a computer program. It may, therefore, be more appropriate to regard the embodiment of a mathematical algorithm as the product of the algorithm in the context of infringement, even though the embodiment does not strictly result from the use of the algorithm. If an embodiment of a mathematical algorithm were regarded as the product of the algorithm, then an unauthorised person who makes, hires or sells, for example, a computer program embodying a patented algorithm would infringe the patent. It should be appreciated that deeming the embodiment of a mathematical algorithm to be its product does not unduly expand the scope of patent protection for the algorithm. In the example above, the only reasonable purpose for an unauthorised person making, hiring or selling of a computer program embodying the algorithm could be to allow that or another person to use the algorithm. The right to prevent such use is precisely the purpose for granting a patent in relation to the algorithm.[141]
The definition of “algorithm” discussed in Section 2 of this paper included problem-solving methods for use by both human and non-human operators. The reality is, however, that because of the tremendous versatility, power and speed of computers, virtually all algorithms of commercial value are for use by computers, and hence the algorithm is embodied in a computer program. As such, a survey of the implications of patenting algorithms would be incomplete without an evaluation of the practical implications of patenting computer programs.
The principal aim of the patent system as it applies to all industry is “to stimulate innovation and inventiveness”.[142] While there has always been some contention about the ability of the patent system to achieve this goal, there has been an especially large amount of debate with regard to the effectiveness of the patent system in relation to the software industry.
It must be accepted that the patent system is now quite entrenched in our society, both generally[143] and specifically in relation to software.[144] Accordingly, the burden is on the opponents of software patents to demonstrate that there is a special case which sets it apart, such that the patent system should not apply to them as it does to all other industries.
A number of features distinguish the software industry from other industries in relation to which the patent system applies. One notable feature is that much less investment capital is required to design and distribute the products of the software industry than is usually the case for hardware-oriented industries.[145] For instance, a complex software application may be designed and created by two people in a year. Once completed, the costs of duplication and distribution (such as via the internet) are negligible. Compare this to the automobile industry, which requires a team of dozens of people working for several years to design and build a new model of car, an expensive factory to produce “copies” of the new model, and a large retail dealership network to display and sell those “copies” to the public. The computer industry is perhaps most distinctive because of its tremendous pace of development. The pace of software development is driven by hardware development. Faster computer hardware demands software which can fully utilise it, which in turn drives further hardware development by encouraging users to upgrade their equipment. Improvements in computer hardware occur almost on a monthly basis, and, consequently, the commercial life of a software product is rarely more than three years.[146] This, of course, is significantly less than most other industries, which lack this rapid development phenomenon.
Another practical distinguishing feature of the software industry is that a commercial computer program is likely to contain thousands or millions of lines of code and possibly hundreds of algorithms,[147] each of which might be patentable or might infringe a patented algorithm. In hardware-orientated industries, there would usually be far fewer inventive, and hence potentially patentable, concepts involved in each innovative product or process.
Finally, there is the nature of software itself. The patentable subject matter of traditional industries generally operate to mirror and further our physical capabilities,[148] whereas software imitates our intellectual processes, replacing or augmenting human mental function in a given situation.[149] Algorithms are part of the armory of basic language skills and tools that computer programmers possess in specifying their requirements to computers. As one programmer has stated, “[s]oftware is, perhaps, more analogous to literature and music than it is to mechanical invention”.[150] Thus algorithms embodied in computer software should perhaps be viewed as a form of creative mental process, rather than an invention per se. It may follow from this that software falls more naturally into the realm of copyright, rather than patent, protection.[151] But the fact that algorithms are more akin to literary or creative works than to the types of inventions which have traditionally fallen within the scope of patent protection lends support to the view that permitting patenting is inappropriate in this field.
A number of implications for the patent system flow from these particular features of the computer software industry. First, it is often argued that the cost of using the patent system is disproportionate to the development costs of patentable algorithms.[152] The patent system is just as expensive for software developers as for more capital-intensive industries. The many valuable algorithms in one program multiply this expense. It is true that the rapid growth of software patents discussed above may belie this theory. It is notable, however, that most software patenting in the US is done by companies which are not part of the “software industry”. In 1995, for example, only 2 per cent of software patents were awarded to software-only companies.[153] The majority were awarded to computer hardware and automotive companies. As such, it might be thought that rather than encouraging industry growth as a whole, the patent system serves only those developers with the resources to use it.
The patent system is also criticised as sluggish.[154] The long application and examination process leaves the rest of the industry in limbo, unsure whether they are free to use the algorithm claimed by the applicant. This period from initial application to granting of the patent is commonly more than two years. This period can represent the entire development period for even the most complex software products. Competitors are faced with the choice of either using an algorithm the subject of a patent application in a new product and thereby risking substantial losses if the patent is granted, or waiting until the patent is granted or rejected and thereby allowing the applicant for the patent a substantial competitive advantage via the lead-time in adopting the algorithm. This sort of situation can hardly be said to encourage innovation.
It appears that the greatest concerns of software professionals about algorithm patents are the costs of licensing existing patents and the costs of being found to infringe an existing patent if a licence is not obtained. These concerns are exacerbated by the likelihood of inadvertent infringement by programmers.[155] The majority of programmers have no training in patent law, and require legal advice to determine whether their work constitutes an infringement. Most of those working in the software industry do not have the resources to remain up to date with developments in intellectual property nor to discover which algorithms are protected.[156] Searching for potentially relevant patents is time-consuming, and even then, the difficulties of algorithm identification make it hard to be confident that there is no infringement. Despite this, programmers report being asked to provide their clients with an indemnity stating that their work does not infringe any patents. Where an indemnity has been granted, the consequences could be disastrous for the programmer.[157]
In summary, given the practical difficulties of patent offices in properly examining applications for algorithm patents, the conceptual difficulties of applying the principles of infringement of process patents to algorithmic inventions, and the nature of the software industry, a case can be made for the view that it is generally inappropriate to grant patents for algorithms. It is therefore desirable in the final substantive section of this paper to consider the potential for, and practicality of, reform of patent law as it applies to algorithms.
The most obvious possible reform of the law to deal with the arguable inappropriateness of patenting algorithms is to exclude that subject matter from patent protection. Such a proposal, however, gives rise to a number of practical issues, each of which needs to be addressed.
7.1.1 Would there be Sufficient Incentive to Produce without Patents?
One argument against the exclusion of algorithms from patent protection is that without patents there would no longer be sufficient incentive to produce the main embodiment of algorithms, namely software. But the fact that creativity and productivity flourished in the software industry before the wide availability of patent protection suggests that the protection of algorithms may not be necessary to encourage development in this field. As the minority judges in Diamond v Diehr noted:
Notwithstanding fervent argument that patent protection is essential for the growth of the software industry, commentators have noticed that this industry is growing by leaps and bounds without it.[158]
Furthermore, it must be recalled that a substantial use of algorithms is in computer programs. Even if algorithms were to be excluded from patent protection, a computer program embodying an algorithm would nevertheless be protected by copyright.[159] This protection would, in certain circumstances, be supplemented by the right to restrain a breach of confidence (also known as trade secrets protection). Whilst the scope of protection provided by these two regimes is by no means coterminous with the protection provided by a patent, nevertheless such protection may well prove to be sufficient to promote innovation in the field of algorithms.[160]
7.1.2 How Could an Exclusion be Implemented? One possibility for implementing an exclusion of algorithms from patent protection is to introduce into the patent legislation an express exclusion of algorithms from the ranks of patentable subject matter. Such an approach would require legislative action, which may be difficult to obtain. An alternative is for the courts simply to interpret the provision of the patent legislation which identifies patentable subject matter as not including algorithms. In Australia, for example, the way is open for the court to apply the policy-oriented approach of the Statute of Monopolies, and interpret the “manner of manufacture” test of inherent patentability by reference to the best interests of the country, including the interests of its software industry. The court could use this freedom to find a claim to an algorithm inherently unpatentable, either as a general rule or in any particular case.
As a general rule, a court might conclude that it is “generally inconvenient”[161] to continue to allow patents in the field of computer software. There is precedent in Australia for such an approach, in relation both to new technologies generally[162] and to information technology in particular.[163] Alternatively, in cases where the patent specification discloses that the claimed algorithm was previously used in another (for instance, non-computer automated) field, a court might hold that the claimed use of the algorithm is merely “analogous to” its previously known uses, and hence is not an “invention” as it fails to satisfy the threshold requirement of inventiveness. Again, there is recent precedent in Australia for such an approach.[164]
7.1.3 Are there International Constraints on Excluding Algorithms from Patenting?
Given the increasing international harmonisation of intellectual property laws in recent decades, it is necessary to consider whether it is practicable or even possible for individual countries to choose to exclude algorithms from patent protection. The World Trade Organization’s Agreement on Trade-Related Aspects of Intellectual Property rights (“TRIPS”) which forms part of the General Agreement on Tariffs and Trade (“GATT”) of 1994, obliges GATT members to maintain uniform intellectual property laws, so as to promote free trade.[165] In particular, article 27 of TRIPS sets out the types of invention for which patent protection shall be available. Article 27(1) provides that:
patents shall be available for any inventions, whether products or processes in all fields of technology, provided that they are new, involve an inventive step and are capable of industrial application ... without discrimination as to the ... field of technology.
Pursuant to articles 27(2) and (3), members of GATT are only allowed to exclude subject matter from patentability in two limited circumstances.[166] The first of these is where the prevention of the commercial exploitation of the invention is necessary to protect “ordre public” or morality, including to protect human, animal or plant life or health, or to avoid serious prejudice to the environment.[167] The second of these is where the subject matter is either (i) a diagnostic, therapeutic or surgical method for the treatment of humans or animals, or (ii) a plant or animal (other than a microorganism), or an essentially biological process for the production of a plant or animal (other than a non-biological or a microbiological process).[168]
At first glance, it might appear that these provisions prevent GATT members from excluding algorithms from patentability. However, the requirement that “patents be available for all inventions ...” goes no further than requiring that protection be available to all products or processes which are properly considered patentable subject matter and which satisfy the further requirements of patentability. A product or process will not be patentable subject matter if it is not an appropriate subject of patent protection – such as, for instance, if it is considered in Australia not to satisfy the threshold requirement of inventiveness.[169] Thus there may still be scope to exclude algorithms from patentability on the grounds that they are not inherently patentable subject matter, and therefore not “inventions” within the meaning of article 27 of TRIPS.
It seems, therefore, that it would be possible for an individual country in its patent legislation expressly to exclude algorithms from protection without failing in its international obligations. But would it be practicable to do so? For countries other than the US, the answer would appear to be no. The US is the undisputed world leader in the software field, and is not adverse to using trade sanctions or at least the threat of them to coerce other countries into adopting intellectual property laws that serve the interests of US trading entities. If a country such as Australia were to exclude algorithms from patentability, it is very likely the US would retaliate with trade sanctions – an outcome that would be politically and economically highly undesirable. Accordingly, in the absence of a push for a change of approach to patenting algorithms from the US, it is unrealistic to expect individual countries to implement an express and general exclusion of algorithms from patent protection.
If the current system of patentability of algorithms is to be retained in Australia, the next most significant reform possibility would be to modify the present system to address some of the various problems identified earlier. There are three areas in relation to which the patent system could be modified – (i) the examination of patents; (ii) the duration of patent protection; and (iii) the scope of patent protection.
7.2.1 Examination of Algorithm Patents
Two of the major difficulties with the examination of algorithmic inventions are the establishment of the prior art for the purposes of judging the requirements of novelty and inventive step, and the determination of what would be obvious to a person skilled in that art as part of the test for whether the claimed subject matter involves an inventive step. Unless the current procedures can be improved, the requirements of novelty and especially inventive step cannot be judged effectively, and examination of the patent application approaches mere registration.[170]
This first of these difficulties is essentially practical. The sources of the information potentially forming the prior art are more diverse than for many other areas of technology, and in any event identifying relevant algorithms within that information is not straightforward.[171] As attempts to improve the prior art database seem likely to present significant administrative problems, a full evaluation of an application is likely to take place not during a patent office’s own examination, but in the context of opposition proceedings. Thus the details of accepted applications must be widely published, so that those in the industry who are aware of relevant prior art have ample opportunity to oppose the grant. The best way of accomplishing this may be by publishing at least the abstract of each accepted application on the Internet. It can be seen that any reforms to overcome the difficulty of an incomplete prior art base must be directed to the practice of the Patent Office, rather than to the provisions of the patent legislation.[172]
This is not the case, however, in relation to the second difficulty – the test for an inventive step. The effect of section 7(2) of the Australian Patents Act is that, subject to the limited exception in section 7(3), only information which is part of the “common general knowledge” of a person skilled in the relevant art in Australia may be considered in judging if a claim involves an inventive step. This limits unduly the information which may be taken into account, especially in relation to a field of technology such as computer software which often utilises algorithms previously known in another field.[173] It would be desirable for this limitation to be removed.
The Australian Patent Office has recognised the argument that patents are more easily obtained in Australia than in other developed economies such as the US and in Europe due to, amongst other things, the way in which the requirement of inventive step is tested.[174] One possibility for reform canvassed by the Patent Office is to have a test of “universal obviousness”, along the lines of that recommended by the Industrial Property Advisory Committee in 1984. Under that approach, the information which may be taken into account in determining if there is an inventive step is “publicly available documentary information anywhere in the world which a skilled artisan should reasonably have been expected to find, ascertain and regard as relevant”.[175] If the test for inventive step was amended to include such information, it is likely that many claims to a computer program embodiment of an algorithm that was previously known in a different field of technology would not satisfy the requirement.
7.2.2 Duration of Algorithm Patents
Given the fast-moving nature of the software industry and the short life cycle of any piece of software, it might be thought appropriate to reduce the term of protection from the current period (which is generally 20 years from the date of filing the application for a patent).[176] A reduction of patent term could be justified on the ground that the advantage in having the sole rights over the invention for the first few years of its commercial life would be sufficient reward for the investment of time and creative effort.[177] However, because TRIPS currently specifies that the term of patent protection must be at least 20 years from the date of filing the application,[178] a reduction in the term of protection for algorithmic inventions will not in fact be possible at this time.[179]
7.2.3 Scope of Algorithm Patents
While TRIPS currently does not allow any reduction in the duration of protection of patented algorithms, it does allow some latitude with respect to scope of protection. Pursuant to article 30:
Members may provide limited exceptions to the exclusive rights conferred by the patent, provided that such exceptions do not unreasonably conflict with a normal exploitation of the patent and do not unreasonably prejudice the legitimate interests of the patent owner, taking account of the interests of third parties.
In particular, it would be possible for GATT members to reduce the scope of the patentee’s exclusive rights by providing for non-voluntary (compulsory) licensing. Article 31 of TRIPS allows the grant of a non-voluntary licence if, prior to use, the proposed user made reasonable attempts to procure a licence from the patentee, and these efforts were not successful within a reasonable time. Where a nonvoluntary licence is granted, the patentee is to be paid adequate remuneration.
The Australian Patents Act does permit the grant of non-voluntary licences, in certain circumstances. Section 133 allows a compulsory licence to be granted if the reasonable requirements of the public with respect to the patented invention have not been satisfied and the patentee has given no satisfactory reason for failing to exploit the patent.[180] The “reasonable requirements of the public” are deemed to have not been satisfied in a number of circumstances, including where an industry in Australia is unfairly prejudiced because of the patentee’s failure to grant a licence on reasonable terms.[181]
Making available non-voluntary licences of algorithm patents might alleviate some of the concerns in the software industry that the patenting of algorithms is removing many of the most effective procedures from the programming arena. However, a programmer wishing to use a patented algorithm would still be required to attempt to procure authorisation from the patentee first, and given the number of algorithms contained in a standard-sized commercial computer program, the difficulty of doing this for each patented algorithm might still effectively exclude these algorithms from use by persons other than the patentee.
This paper has sought to determine the availability and scope of patent protection in Australia for algorithms. In doing so it has noted that the importance of algorithms in practice is that they form the backbone of a device by which the algorithm can be used. The most common form of such a device is a computer program.
There are, in the authors’ opinion, a number of difficulties with the application of the Australian patents legislation to algorithms intended for embodiment in a computer program. These are summarised as follows.
A.2 EUCLID’S ALGORITHM IN COMPUTER PROGRAM FORM A typical expression of Euclid’s algorithm for use in programming contexts is: Take two positive integers, m and n.
1. [find remainder] Divide m by n; let the remainder be denoted r (0 = r < n).For example, if we start with m = 75 and n = 30
2. [remainder = 0?] If r = 0, n is the answer.
3. [replace and repeat] Else, set m = n, n = r and repeat from step 1.
1. 75/30 = 2 remainder 15, so r = 15REPEAT
2. r ? 0
3. Let m = 30, n = 15.
1. 30/15 = 2 remainder 0, so r = 0–> The greatest common divisor of 75 and 30 is 15. The algorithm would not be entered into a computer in this form; it would first be expressed in a computing language, such as Pascal. Thus, a computer program (as opposed to an algorithm) is created. Two computer programs embodying Euclid’s algorithm are set out below.
2. r = 0, so 15 is the answer.
A.3 EUCLID’S ALGORITHM IN THE PROGRAMMING LANGUAGE PASCAL This first program is written in Pascal:
program Euclid (input, output);A.4 EUCLID’S ALGORITHM IN THE PROGRAMMING LANGUAGE C Another implementation of Euclid’s algorithm in the language C (from Sedgewick, R., Algorithms in C, 1990, Addison Wesley, at p.8):
var
m, n, r: integer
begin
write (“Please enter two positive integers:”);
readln (m,n);
r := n;
while r <> 0 do {<> means ?}
begin
if m > n then
r := m mod n {finds the remainder of m/n}
else r:= n mod m;
m:=n;
n:=r;
end;
writeln(“The greatest common divisor of numbers =”,m); {m has taken the old value of n; n is now zero}
end. {Euclid}
#include <stdio.h>
int gcd(int u, int v){gcd is a function within the program}
{
int t;
while (u>0)
{
if (u < v)
{t = u; u = v; v = t; }
u = u–v;
}
return v;
}
main() {this is the main section of the program}
{
int x,y;
while (scanf (“%d %d”, &x, &y) !=EOF)
if (x>0 && y>0)
printf(“%d %d %d\n”, x, y, gcd(x,y));
}
AustLII:
Copyright Policy
|
Disclaimers
|
Privacy Policy
|
Feedback
URL: http://www.austlii.edu.au/au/journals/SydLawRw/1998/23.html