Thanks for the link to the Statutory Guidance, which provides some very helpful commentary, and I feel that points 12 and 13 are of some interest:
12. The judgment accepts an argument that moving data outside the computer system, in the form of a file that is transferred, provides an external (outside world) technical effect (see [73]-[74]). When coupled with the purpose and method of selecting the file’s contents, this fulfills the requirement for a technical effect which avoids the computer program exclusion (see [76]).
13. The judgment further holds that a trained hardware ANN is capable of being an external technical effect which prevents the computer program exclusion from applying (see [78]). The judgment holds there ought to be no difference between a hardware ANN and an emulated ANN in this respect.
If I am not mistaken (though there is a good chance I am), the judge could have made a narrow judgement in favor of the plaintiff on the basis of point 12 alone. Inventions that use a suitably-programmed computer in a process that produces an "external technical effect" are not prima facie ruled out (though they do have to satisfy a requirement of sufficiency of description, which, as the judge noted in his opening remarks, is not being ruled on here, together with (I think) non-obvious novelty and utility)
Point 13 may have much broader implications, as it asserts that any trained software ANN is, in itself, an "external technical effect", and presumably patentable if it satisfies other requirements such as sufficiency.
The first issue with this is, as you and others have pointed out, that the justification given for this claim (the possibility of a functionally-equivalent implementation in hardware) works equally well for any computer program - ANN programs are not different in this regard. If followed to its logical conclusion, this principle would seem to completely nullify the "program for a computer" exclusion.
A second point that might be made here is that the difference between a trained ANN (whether implemented in software or hardware), and itself in its untrained state, are data - specifically, the weights. This might imply that the intellectual property issue here is a matter of copyright, not invention.
Finally, and perhaps most broadly, if a trained ANN is, in itself, an "external technical effect", does that open the door to the patenting of ANN training methods? (Maybe they already are?)
12. The judgment accepts an argument that moving data outside the computer system, in the form of a file that is transferred, provides an external (outside world) technical effect (see [73]-[74]). When coupled with the purpose and method of selecting the file’s contents, this fulfills the requirement for a technical effect which avoids the computer program exclusion (see [76]).
13. The judgment further holds that a trained hardware ANN is capable of being an external technical effect which prevents the computer program exclusion from applying (see [78]). The judgment holds there ought to be no difference between a hardware ANN and an emulated ANN in this respect.
If I am not mistaken (though there is a good chance I am), the judge could have made a narrow judgement in favor of the plaintiff on the basis of point 12 alone. Inventions that use a suitably-programmed computer in a process that produces an "external technical effect" are not prima facie ruled out (though they do have to satisfy a requirement of sufficiency of description, which, as the judge noted in his opening remarks, is not being ruled on here, together with (I think) non-obvious novelty and utility)
Point 13 may have much broader implications, as it asserts that any trained software ANN is, in itself, an "external technical effect", and presumably patentable if it satisfies other requirements such as sufficiency.
The first issue with this is, as you and others have pointed out, that the justification given for this claim (the possibility of a functionally-equivalent implementation in hardware) works equally well for any computer program - ANN programs are not different in this regard. If followed to its logical conclusion, this principle would seem to completely nullify the "program for a computer" exclusion.
A second point that might be made here is that the difference between a trained ANN (whether implemented in software or hardware), and itself in its untrained state, are data - specifically, the weights. This might imply that the intellectual property issue here is a matter of copyright, not invention.
Finally, and perhaps most broadly, if a trained ANN is, in itself, an "external technical effect", does that open the door to the patenting of ANN training methods? (Maybe they already are?)