Notice of the EPO dated 7 April 2025 concerning the preparation of the minutes of OP held by ViCo with the assistance of AI will be issued in the next OJ.
A corresponding decision of the president dated 7 April 2025 is as well to be issued in the next OJ.
In May 2025, the EPO will start a pilot for the drawing up of the minutes of OP held by ViCo before EDs and ODs with the assistance of AI.
Following a human-centric approach to using AI, the goal is to further improve the quality of the minutes of OP and to provide them more efficiently.
Over time, and following a review of the pilot, the EPO will gradually extend this practice to cover all OP before EDs and ODs as well as those before the Receiving Section and the Legal Division.
To enable examiners to make full use of the capabilities of AI tools when drafting minutes, the EPO will audio record OP in their entirety. The division will then use the transcriptions of these recordings to draw up the minutes of OP with the assistance of AI technology.
The recording and its transcription are preparatory documents within the meaning of R 144(b).
They will neither be public, i.e. in the register, nor made available to the parties.
The recording and its transcription will be deleted once the minutes of OP are issued to the parties to proceedings.
The pilot will start with OP before selected EDs and ODs scheduled to take place in May 2025. It is planned to run until the end of 2025.
Comments
That the EPO makes use of the development of technology is to be welcomed. It should however not go on a par with an increased pression on examiners.
When the EPO speaks about a “pilot”, it is more or less certain that its benefits will be judged positively when reviewed. It is thus a euphemism to speak about a “pilot”.
As there is the possibility for the parties to request a correction of the minutes, the recording and its transcription should not be deleted once the minutes of OP are issued to the parties.
The possibility of requesting a correction of the minutes is explicitly provided for in the Guidelines E-III, 10.4.
In T 1679/17 it was held that, according to the established case law of the BA, the parties and their representatives are expected to carefully examine the content of the minutes, in particular their completeness and correctness, immediately after receipt and, if necessary, to object to them promptly. Here the opponent took 4 months to request a correction. This was way too long, and the board considered that the opponent had actually withdrawn the objections under Art 83 and 123(2).
Without a request for a correction of the minutes, it is more or less impossible to claim that a Substantial Procedural Violation has occurred. See for instance T 1481/19.
A reasonable delay, let’s say of one month, should thus be observed before deleting the recording and its transcription.
The quality of the minutes could be improved, even without AI, in that they correspond to the decision and vice versa. In all recent SPV it could be seen that the minute writer did not care about the decision and vice versa, and the chair of the OD signed both documents without batting an eyelid. See for instance T 2396/22 or T 1753/21 in examination, and T 1870/21 or T 21/24 in opposition.
Whether AI will really improve the quality of the minutes will primarily depend on the quality of the training data. If the training data are biased, the result of the AI will also be biased. It would be interesting to know what were the training data.
Comments
2 replies on “AI and minutes of OP – Advance publication for the OJ”
It would be surprising – and hugely inefficient and wasteful of users’ fees – if the EPO develops its own speech transcription software. They will surely use a third-party system – perhaps installed on premise to keep data within their control – of which there are many to choose. As such, I doubt we will ever see the training data.
I can somehow agree with you that it “would be surprising – and hugely inefficient and wasteful of users’ fees – if the EPO develops its own speech transcription software”. I do however doubt that any off the shelf speech transcription software will do the job.
Although not belonging to AI programs, when the EPO has introduced WiseFlow as EQE examination software, the candidates where informed and training sessions were offered.
If a third party software is actually used, then it would be important and useful to inform users of the EPO which third party software has been acquired and how it is used and/or if it has been adapted to the needs of the EPO. Simply transcribing the voice recordings on paper would not bring much in matters of efficiency for the minute writer.
In order to gain time, witnesses’ hearings have been recorded for many years, and the transcript was then used as minutes. During such a recording, it used to be necessary to identify each speaker before it starts speaking. It is to be hoped that witness hearings will not undergo AI treatment and that a full transcript will still be delivered. I cannot imagine AI manipulated witnesses’ hearings.
As the aim is to improve the quality of minutes, but at the same time relieve examiners from some work, the AI manipulated recordings will result in some summarizing of the discussions. And there it will be necessary to watch what the AI is actually doing. Do you know any off the shelf speech transcription software doing this in a reliable way?
Last but not lest, the use of AI for the minutes should not hinder the normal progress of OP.