This Science article http://science.sciencemag.org/content/362/6419/1140 describes an example of such a class of AI inventions that I believe should be patentable as a matter of principle. Please let me know if you have questions or if you would like to discuss this.
The following is a summary and a critique of the new guidelines as they apply to software-implemented inventions.
- What is technical?
A two-step test is applied to assess ‘technical effect’ or ‘technical contribution’ (the so called ‘two hurdles approach’):
Step 1: Is there at least one ‘technical means’ in the claim? If so, this overcomes Article 52 EPC (equivalent to USC 101).
For example, an apparatus claim has ‘technical means’ if there is at least one hardware device (processor, memory, sender, input/output circuit, …) or at least an electronic signal. A method has ‘technical means’ if it is ‘computer-implemented’ or has some hardware element in one of the method steps.
Step 2: Does the claimed subject matter have a ‘technical purpose’?
Embedding the underlying algorithm of the invention into a technical framework provides a basis for arguing ‘technical purpose’. The examples given in the guidelines comprise signal processing, encryption/encoding and generating a key in a cryptosystem.
On the other hand, a control method for controlling some unspecified technical process is considered to be too broad to justify a ‘technical purpose.’ Likewise, an improved machine learning method that provides an improved learning capability but without a limitation to a concrete technical application would be too abstract to confer a ‘technical purpose’.
If both hurdles, i.e. steps 1 and 2, are overcome, the mathematical/algorithmic content of the claim is taken into consideration for assessing inventive step; hence, in this case the case law on mixture of technical/non-technical features does not apply and all claim features (also those which relate to an algorithm) can support inventive step.
Critique: It is illogical that the ‘technical purpose’ depends on the broadness of the claim. The allowable claim scope must be determined in view of the prior art. If the invention is broadly applicable for technical control systems such as by providing a more stable, faster, precise or otherwise technically improved control methodology, there is no reason to restrict the scope to some arbitrary concrete technical application.
As to machine learning it is also illogical to deny ‘technical purpose’ for an improved machine learning method that makes machines smarter. Such an improved algorithm would enable to build improved machines patent protection of which is precisely the purpose of the patent system. There is no conceivable reason why a claim that contains an improved machine learning method should be limited to some concrete purpose as improved machine learning is in itself such a concrete purpose.
Practical advice: Patent applications on software-implemented inventions should contain concrete implementation examples and explanations of technical effects and purposes of the invention. On the other hand, the claims should be formulated as broad as possible only in view of the prior art without limiting the claims to some arbitrary technical purpose. If the invention resides in the algorithm and is broadly applicable in various technical context the claims should have a corresponding broadness. The updated guidelines suggest the contrary, which however seems illogical and not founded by the EPC which is there to protect innovation. Furthermore, the guidelines are just this – guidelines – and they are non-binding for the Boards of Appeal. It would not be the first time that the guidelines would have to be changed due to Boards of Appeal case law.
On the other hand, it may be a pragmatic approach to limit the claim scope to some technical purpose or to at least provide a respective fallback position if this technical purpose is the sole purpose that is within the business interest of the client in order to expedite the patent examination process.
In full: EPO Guidelines G-II, 3.3 to 3.6 – Patentability of mathematical methods, mental acts, business methods, and programs for computers
- Multiple claims in the same category
In general, Rule 43 EPC allows only one independent claim per claim category, i.e. method and apparatus, unless one of the exceptions listed in that rule is applicable. The updated guidelines provide clarification what that means for computer-implemented inventions (CIIs) which are often realized in a distributed computing environment:
‘For such distributed CIIs, the claim set may comprise claims directed to each entity of the distributed system and/or to the overall system and the corresponding methods’.
Critique: This clarification is very welcome as it eliminates endless discussions with the Examining Division as to the number of allowable claim categories and often eliminates the need to file a divisional.
Practical advice: All commercially relevant tradable entities of the invention should be claimed separately for maximum scope and enforceability. In particular, this applies for artificial intelligence inventions where a training system and a resultant technical system that includes the trained classifier should be claimed as separate entities. Also the method of training and the ensuing method of using the training result should be claimed separately.
In full: EPO Guidelines F-IV, 3.9.3 – Cases where the invention is realized in a distributed computing environment