AI lab | Podcast Season: 1 - Episode: 2 / Release date: 6-9-2023 - Recording date: 12-6-2023

1:1 with João Pedro Quintais
Institute for Information Law (IViR)

Episode Highlights

00:00 Intro

01:16 The TL;DR Perspective

06:27 The Deepdive: AI decrypted
Q1 | You consider that it is impossible to comply with the transparency obligation to “document and make publicly available a summary of the use of training data protected under copyright law”. Can you explain why?

09:58 Q2 | Can you explain: 1. how you connect the TDM exceptions in the Directive on Copyright in the Digital Single Market to the EU AI Act; and, 2. your views on the different schools of thoughts on this?

16:17 Q3 | You refer to the user safeguards in Article 17 of the Directive on Copyright in the Digital Single Market, e.g. exceptions and freedom of speech. Where do you make the link with the EU AI Act?

21:13 Outro

Key Quotes

"[The EP’s AI Act] documentation obligations will prove either difficult to comply [with] or ineffectual, and the safeguards obligation might require development of Content ID like tools and even lead to overfiltering."

1 / 9

"The goal of [having] generative AI providers listing all or most of the copyrighted material that they are including in their training datasets in an itemised manner with clear indication of rights ownership claims (...): it's impossible to comply with it."

2 / 9

"The low threshold of originality, the territorial fragmentation of copyright and its ownership, absence of a registration requirement for works, and in general the poor state of (...) metadata, kind of demonstrate this impossibility [of the EP’s AI Act obligations]."

3 / 9

"Focus more on improving access to datasets, incentivising cooperation between providers and rights holders, and possibly [standardising] the reservation of rights or opt-outs (...) under the [Copyright in the Digital Single Market] Directive."

4 / 9

"There has long been a tradition of unequal contractual relationships between (...) artists and their commercial exploiters (...): [it is] crucial that creators are protected against signing away these training rights absent appropriate remuneration."

5 / 9

"Rules that attempt to safeguard freedom of expression and obligations to prevent illegal content in the DSA would not necessarily directly apply [to] generative AI tools."

6 / 9

"Overfiltering or overblocking: (...) it's very easy, and you see it already in the copyright discussions, the tendency to propose blunt tools like keyword filtering at the prompt stage or really strong matching filtering at the actual output stage."

7 / 9

"Bolster the [EU AI Act] language (...): there's a real risk here that voluntary solutions (...) would lead to an outcome where a lot of output is filtered contrary to (...) freedom of expression."

8 / 9

"The real danger: (...) the lack of clear direct application of all the discussions and acquis that we have on the copyright side for filtering by OCSSPs or on the DSA (...). Clarification could come (...) by better formulation of the provision."

9 / 9

❮ ❯

Available On

Watch Episode & Highlights

About Our Guest

Dr. João Pedro Quintais | Assistant Professor - Institute for Information Law (IViR)

Dr João Pedro Quintais is Assistant Professor at the University of Amsterdam’s Law School, in the Institute for Information Law (IViR). João notably studies how intellectual property law applies to new technologies and the implications of copyright law and its enforcement by algorithms on the rights and freedoms of Internet users, on the remuneration of creators, and on technological development. João is also Co-Managing Editor of the widely read Kluwer Copyright Blog and has published extensively in the area of information law.

Relevant Resources

Kluwer Copyright Blog | Generative AI, Copyright and the AI Act

Institute for Information Law (IViR)

Dr João Pedro Quintais

1:1 with João Pedro QuintaisInstitute for Information Law (IViR)