Patenting Artificial Intelligence in Europe

Patenting AI in Europe has developed to become one of the ‘hot topics’ around because of the large expectations around this emerging and divergent technology. The EPO (European Patent Office) held a conference in Munich on 30th May to further expound on this delicate matter, with videos and e-courses on its website as proof of this.

A quick search on Espacenet Patent search with the keywords ‘neural network’, ‘EP’ and ‘2017’ generated just 78 hits. In other parts of the world, this artificial intelligence based search might not be deemed as significant considering the same search in 2017 yielded 612 hits for US patent applications and 3,679 for Chinese patent applications.

Regardless of the low number of filings, the EPO has committed itself in preparing for a large number of filings once the key players, China and America, decide to extend the application of local patent in Europe. Another key factor for the EPO’s interest is the rule of thumb that AI will majorly influence the way that the world works. Patent offices, including the EPO, should effectively be prepared to operate such filings with precision.

At the conference, the EPO talked at length about this topic, hosting a couple of representative patents in the AI field, who were provided with the USPTO and the EPO. Various aspects of the AI were disclosed by the patent, which used an array of terms in their specifications and claims.

 

The filings were shown in depth based on their appearance and how the whole process works at EPO conference. Utterances that are not normally used in claims such as artificial intelligence although they are clear and accepted could be replaced with buzzwords such as trained recognizer, which offer an expanded view of specification appearing on dependent claims only. Having learnt the above, one can deduce if a learned algorithm can be used instead and seek to enclose it in the main claim.

The main reason for this kind of approach is that a particular step-by-step claim can be conjured around by a model based on Big Data. There are different courts who all have divergent approaches when applying the function-way-result test or similar other approaches for equivalence under Article 69 of the European Patent Convention, some of which may deduce that the two ways are not quantifiably the same.

In addition, this brings up issues in valuing the novelty as a trained model cannot predict a specific detailed step-by-step process depending on the type of patent applications seen so far. To add onto this, innovative step provisions from other parts should be integrated to AI patents by the EPO.

All the more, artificial intelligence is a subtle player and designing around could rapidly change as the so-called ‘distillation algorithms’ are integrated to a trained model. These algorithms should come up with a reciprocating flowchart algorithm from the trained model, which courts could use, in the event access to the trained model has been requested for checking its equivalence. This would create an enforcing computer-implemented method claim tougher than before; however, it is far from realization, and thus necessary changes could be made before.

Conclusion

There might be difficulties in the patent system with the expanding use of non-core AI patents; however, the European patent system can depend on a firm jurisprudence that works well for more computer-implemented inventions and may be adapted to endure with such difficulties.


GUTE-URLS

Wordpress is loading infos from epo

Please wait for API server guteurls.de to collect data from
e-courses.epo.org/pluginfile.php...