AI lab | Podcast Season: 1 - Episode: 3 / Release date: 27-9-2023 - Recording date: 8-9-2023

1:1 with Teresa Nobre
COMMUNIA

Episode Highlights

00:00 Intro

01:10 The TL;DR Perspective

08:11 The Deepdive: AI decrypted
Q1 | COMMUNIA’s Policy Paper #15 states that: “The use of copyrighted works as part of the training data is exactly the type of use that was foreseen when the TDM exception was drafted and this has recently been conﬁrmed by the EC in response to a parliamentary question”. Can you clarify that?

11:04 Q2 | COMMUNIA clearly favours transparency when it comes to AI models, but also points out that when it comes to copyrighted material: “Policy makers should not forget that the copyright ecosystem itself suffers from a lack of transparency”. What do you mean by that?

16:41 Q3 | COMMUNIA sees a need to operationalise the TDM opt-out mechanism. You recommend the EC should play an active role to encourage a fair and balanced approach to opt-out and transparency through a broad stakeholder dialogue. What could that entail?

21:39 Outro

Key Quotes

"Everything would be easier if there was more transparency across the copyright ecosystem itself. (...) There's no place that you can consult that will tell you who are the owners, who are the creators, the title of the work."

1 / 9

"Machine learning developers (...) will not be able to provide this [copyright] information, because this information is simply not publicly available for the vast majority of works."

2 / 9

"To demonstrate compliance with copyright law, machine learning developers only need to show that they have respected machine-readable rights reservations."

3 / 9

"Our recommendation: European Commission, do something that's more towards involving everyone in the solution to the problem."

4 / 9

"[Our paper intended] to remind everyone that this [text and data mining (TDM)] framework provides legal certainty to machine learning training in Europe and, in our opinion, is a sensible and forward-looking approach."

5 / 9

"In order to preserve the commercial text and data mining (TDM) exception, two things need to happen. First, machine learning development needs to become more transparent. Second, we need clarity on how to make the opt-outs effective."

6 / 9

"A training data transparency requirement should be introduced in the EU in order to uphold the opt-out system. This should be a very reasonable and very proportional requirement to avoid placing disproportionate burden on machine learning developers."

7 / 9

"The European Commission has a crucial role to play: it should lead discussions [with stakeholders] to provide guidance on how the opt-out is supposed to work in practice, and the same with the transparency requirements"

8 / 9

"The European Commission clarified that the [Copyright Directive] exceptions facilitate text and data mining (TDM), including by AI developers."

9 / 9

❮ ❯

Available On

Watch Episode & Highlights

About Our Guest

Teresa Nobre | Legal Director - COMMUNIA

Teresa Nobre is the Legal Director of COMMUNIA, an international association that advocates for policies that expand the Public Domain and increase access to and reuse of culture and knowledge. She is an attorney-at-law and is involved in policy work both at the EU level and at the international level, representing COMMUNIA at the World Intellectual Property Organization.

Relevant Resources

The AI Act and the quest for transparency (COMMUNIA blog post)

Policy Paper #15 on Using Copyrighted Works for Teaching the Machine (COMMUNIA)

Answer given by Commissioner Thierry Breton on behalf of the European Commission to the Parliamentary Question by MEP Emmanuel Maurel (The Left, France)

Teresa Nobre

1:1 with Teresa NobreCOMMUNIA