AI Lab – AI In Action | Episode 03: AI Tokenization AI Lab By Information Labs podcast

16:51

🔍 In this TL;DR episode, Kevin Frazier (University of Texas at Austin school of Law) outlines a proposal to realign U.S. copyright law with its original goal of spreading knowledge. The discussion introduces three key reforms—an AI training presumption, research safe harbors, and data commons—to help innovators access data more easily. By reducing legal ambiguity, the proposals aim to support responsible AI development and level the playing field for startups and researchers. 📌 TL;DR Highlights ⏲️[00:00] Intro ⏲️[00:40] Q1-How do copyright laws limit AI’s training data and knowledge diffusion? ⏲️[04:05] Q2-How does the original intent of the Intellectual Property Clause conflict with current copyright rules? ⏲️[10:35] Q3-What reforms could align copyright with promoting science? ⏲️[15:36] Wrap-up & Outro 💭 Q1 - How do copyright laws limit AI’s training data and knowledge diffusion? 🗣️ "Copyright law was designed to promote knowledge creation. Now it functions as a bottleneck in that knowledge ecosystem." 🗣️ "Today's AI systems face a paradoxical constraint: they require vast oceans of human-created content to learn, and yet our copyright framework increasingly cordons off those essential waters." 🗣️ "Early publishers faced guild restrictions and royal monopolies that limited the dissemination of knowledge. Today's AI developers are navigating a similarly restrictive landscape through the barriers of copyright law." 🗣️ "When there's ambiguity in the law, that hinders innovation. It leads to litigation and poses a real threat to startups trying to determine whether they can use copyrighted information for training." 💭 Q2 - How does the original intent of the Intellectual Property Clause conflict with current copyright rules? 🗣️ "The IP clause starts off with a mandate that Congress spread knowledge. If something isn’t promoting the progress of science, then it can’t be interpreted as constitutional." 🗣️ "Copyright began as a 14-year term. Now it's expanded to more than 70 years — a huge restriction on the ability to spread knowledge." 🗣️ "The founders hated monopolies. They’d seen how royal prerogatives were used to quash innovation — and tried to create a better system for incentivizing knowledge." 🗣️ "AI tools are unparalleled in their ability to create knowledge. The question now is: can we spread that knowledge?" 🗣️ "The Constitution’s goal wasn’t just to reward creators — it was to spread science and useful arts as far and wide as possible." 💭 Q3 - What reforms could align copyright with promoting science? 🗣️ "We need a clear statutory presumption that using works for machine learning constitutes fair use — that sort of clarity is essential for startups and research institutions to compete." 🗣️ "Without robust datasets, the positive use cases of AI — from public health breakthroughs to AI tutors for differently-abled students — simply aren’t possible." 🗣️ "Imagine if all that data your smartwatch gathers went toward training AI models tailored to the public good — that’s the promise of data commons." 🗣️ "AI is like fire: it can spread and fuel incredible progress — but if we smother it with fire extinguishers too soon, only the biggest players will be able to benefit." 🗣️ "We must make sure this isn’t a world where only OpenAI, Anthropic, and Google build the models — we need a future with many options and many positive use cases of AI." 📌 About Our Guest 🎙️ Kevin Frazier | the University of Texas at Austin School of Law 🌐 Article | Progress Interrupted: The Constitutional Crisis in Copyright Law https://jolt.law.harvard.edu/digest/progress-interrupted-the-constitutional-crisis-in-copyright-law 🌐 Kevin Frazier https://www.linkedin.com/in/kevin-frazier-51811737/ Kevin Frazier is an AI Innovation and Law Fellow at the Austin School of Law of the University of Texas. He is also a Contributing Editor at Lawfare, a non-profit publication and he developed the first open-source Law and AI syllabus. #AI #ArtificialIntelligence #GenerativeAI…

AI lab by information labs

1
AI lab TL;DR | Paul Keller - A Vocabulary for Opting Out of AI Training and TDM 15:06

16天之前15:06

15:06

🔍 In this TL;DR episode, Paul Keller (The Open Future Foundation) outlines a proposal for a common opt-out vocabulary to improve how EU copyright rules apply to AI training. The discussion introduces three clear use cases—TDM, AI training, and generative AI training—to help rights holders express their preferences more precisely. By standardizing terminology across the value chain, the proposal aims to bring legal clarity, promote interoperability, and support responsible AI development. 📌 TL;DR Highlights ⏲️[00:00] Intro ⏲️[00:41] Q1-Why is this vocabulary needed for AI training opt-outs? ⏲️[04:17] Q2-How does it help creators, AI developers, and policymakers and what are some of the concepts? ⏲️[11:55] Q3-What are its limitations, and how could it evolve? ⏲️[14:35] Wrap-up & Outro 💭 Q1 - Why is this vocabulary needed for AI training opt-outs? 🗣️ "At the core of the EU copyright framework is... the TDM exceptions – the exceptions for text and data mining that were introduced in the 2019 Copyright Directive." 🗣️ "It ensures that rights holders have some level of control over their works, and it makes sure that the majority of publicly available works are available to innovate on top of, to build new things." 🗣️ "The purpose of such a vocabulary is to provide a common language for expressing rights reservations and opt-outs that are understood in the same way along the entire value chain." 🗣️ "This vocabulary proposal is the outcome of discussions that we had with many stakeholders, including rights holders, AI companies, policymakers, academics, and public interest technologists." 💭 Q2 - How does it help creators, AI developers, and policymakers and what are some of the concepts? 🗣️ "At the very core, the idea of vocabulary is that you have some common understanding of language... that terms you use mean the same to other people that you deal with." 🗣️ "We offer these three use cases for people to target their opt-outs from... like sort of the Russian dolls: the wide TDM category that is AI training, and in that is generative AI training." 🗣️ "If all of these technologies sort of use the same definition of what they are opting out, it becomes interoperable and it becomes also relatively simple to understand on the rights holder side." 💭 Q3 - What are its limitations, and how could it evolve? 🗣️ "The biggest limitation is... we need to see if this lands in reality and stakeholders start working with this." 🗣️ "These information intermediaries... essentially convey the information from rights holders to model providers—then it has a chance to become something that structures this field." 🗣️ "It is designed as a sort of very simple, relatively flexible approach that makes it expandable." 📌 About Our Guest 🎙️ Paul Keller | The Open Future Foundation 🌐 Article | A Vocabulary for opting out of AI training and other forms of TDM https://openfuture.eu/wp-content/uploads/2025/03/250307_Vocabulary_for_opting_out_of_AI_training_and_other_forms_of_TDM.pdf 🌐 Paul Keller https://www.linkedin.com/in/paulkeller/ Paul Keller is the co-Founder and Director of Policy at the Open Future Foundation, a European nonprofit organization. He has extensive experience as a media activist, open policy advocate and systems architect striving to improve access to knowledge and culture. #AI #ArtificialIntelligence #GenerativeAI…

AI lab by information labs

1
AI lab TL;DR | João Pedro Quintais - Untangling AI Copyright and Data Mining in EU Compliance 25:10

5 weeks之前25:10

25:10

🔍 In this TL;DR episode, João Quintais (Institute for Information Law) explains the interaction between the AI Act and EU copyright law, focusing on text and data mining (TDM). He unpacks key issues like lawful access, opt-out mechanisms, and transparency obligations for AI developers. João explores challenges such as extraterritoriality and trade secrets, offering insights into how voluntary codes of practice and contractual diligence could help align AI innovation with EU copyright rules. #AI #ArtificialIntelligence #GenerativeAI 📌 TL;DR Highlights ⏲️[00:00] Intro ⏲️[00:51] Q1-What are the key interactions between the AI Act and EU copyright law, particularly concerning text and data mining (TDM) practices? ⏲️[7:12] Q2-How do you see the transparency obligations of the AI Act shaping the relationship between AI developers and rightsholders? ⏲️[15:22] Q3-Is the extraterritorial reference of Recital 106 of the AI Act enforceable? ⏲️[24:35] Wrap-up & Outro 💭 Q1 - What are the key interactions between the AI Act and EU copyright law, particularly concerning text and data mining (TDM) practices 🗣️ "The AI Act links to Article 4 by obliging general-purpose AI (GPAI) model providers to identify and respect rights reservation mechanisms and disclose a sufficiently detailed summary about the training data used." 🗣️ "The Copyright Directive of 2019 introduces exceptions and limitations for text and data mining (TDM), with Article 3 aimed at research and Article 4 applying broadly but with additional requirements like rights reservation or opt-out mechanisms." 🗣️ "The concept of TDM is so broad that it applies to activities involved in pre-training and training AI models, impacting entities across the value chain, not just model providers." 🗣️ "Entities like Common Crawl or LAION that perform upstream activities like web scraping are not directly regulated by the AI Act but are part of the broader TDM definition under the Copyright Directive." 🗣️ "One debated requirement is the rights reservation or opt-out mechanism for publicly accessible online content." 💭 Q2 - How do you see the transparency obligations of the AI Act shaping the relationship between AI developers and rightsholders? 🗣️ "The transparency provision in Article 53(1)(d) requires GPAI model providers to make publicly available a sufficiently detailed summary of training data, balancing interests like copyright, privacy, and fair competition." 🗣️ "If the summary is too vague, it becomes meaningless; if too detailed, it might infringe on trade secrets—so finding a balance is critical." 🗣️ "The usefulness of the training data summary lies in clarifying whether TDM exception requirements, such as lawful access and respect for opt-outs, have been met." 🗣️ "A significant challenge is ensuring compliance when data sets are obtained from upstream providers not subject to the AI Act, raising questions about responsibility and enforcement." 🗣️ "The Act balances interests, acknowledging the impossibility of listing all copyrighted works used in training due to territorial fragmentation and low originality thresholds." 💭 Q3 - Is the extraterritorial reference of Recital 106 of the AI Act enforceable? 🗣️ "Recital 106 aims to prevent regulatory arbitrage by requiring compliance with EU copyright standards, even for AI models trained outside the EU." 🗣️ "The principle of territoriality in copyright law conflicts with the extraterritorial implications of the AI Act, as copyright rules are typically governed by the location of the activity." 🗣️ "Using contractual obligations and voluntary meta-regulation, such as commitments from upstream providers, offers a more consistent way to enforce compliance than extending the law extraterritorially." 🗣️ "The Act's compliance incentives might still push GPAI providers to align with EU standards to avoid severe sanctions, even if extraterritorial enforcement remains uncertain." 🗣️ "Some suggest contractual obligations or meta-regulation as more practical solutions to ensure upstream compliance with EU law." 📌 About Our Guest 🎙️ João Pedro Quintais | Assistant Professor, Institute for Information Law (IViR) 🌐 Article | Generative AI, Copyright, and the AI Act https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4912701 🌐 Joao Pedro Quintais https://www.ivir.nl/employee/quintais/ 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.…

AI lab by information labs

1
AI lab TL;DR | Anna Tumadóttir - Rethinking Creator Consent in the Age of AI 19:41

8 weeks之前19:41

19:41

🔍 In this TL;DR episode, Anna Tumadóttir (Creative Commons) discusses how the evolution of creator consent and AI has reshaped perspectives on openness, highlighting the challenges of balancing creator choice with the risks of misuse. Examines the limitations of blunt opt-out mechanisms like those in the EU AI Act, the implications for marginalized communities and open access, and explores the need for nuanced preference signals to preserve openness while respecting creators' intentions in the age of generative AI. 📌 TL;DR Highlights ⏲️[00:00] Intro ⏲️[00:38] Q1-How has the conversation around creator consent and AI evolved in the past few years? ⏲️[05:42] Q2-How has the tension between openness and creator choice played out so far, and what lessons can be learned from this tension? ⏲️[11:02] Q3-can we ensure that marginalized or underrepresented communities maintain agency over their contributions to the commons? ⏲️[19:03] Wrap-up & Outro 💭 Q1 - How has the conversation around creator consent and AI evolved in the past few years? 🗣️ "When AI became mainstream, most creators couldn’t have anticipated how their works might one day be used by machines—some uses align with their intentions, but others do not." 🗣️ "Individuals want more choice over how their work is used, and without tailored options, some are considering paywalls or not publishing at all." 🗣️ "The EU AI Act’s opt-out mechanism is a blunt instrument—it’s just a 'no,' not a nuanced reflection of creators’ varied preferences." 🗣️ "Creators may object to large companies using their works for AI training but be fine with nonprofits or research-focused uses, showing the need for more nuanced tools." 🗣️ "We’re focusing on developing 'preference signals'—mechanisms that let creators communicate specific preferences for how their work is used in AI models." 💭 Q2 - How has the tension between openness and creator choice played out so far, and what lessons can be learned from this tension? 🗣️ "Scientists and researchers who traditionally embraced open access are now reconsidering, fearing that commercial AI providers are exploiting their work." 🗣️ "Creators who once freely shared their work under CC licenses are now hesitant, either because they misunderstand AI training risks or feel exposed." 🗣️ "The worst outcome of this tension is less openness overall—creators retreating behind paywalls or choosing not to publish at all." 🗣️ "A perception persists in the open-source AI community that CC-licensed works are 'safe' to use, but creators’ motivations for sharing openly years ago don’t always align with today’s AI landscape." 🗣️ "To preserve openness while respecting creator intentions, we need mechanisms that enable a 'no unless' approach—minimizing restrictions while maximizing use." 💭 Q3 - can we ensure that marginalized or underrepresented communities maintain agency over their contributions to the commons? 🗣️ "Generative AI amplifies existing inequalities because it demands infrastructure like electricity, internet, and computing power—resources many regions lack." 🗣️ "Even if everyone had equal internet access, a one-size-fits-all approach to technology wouldn’t work due to local contexts and different needs." 🗣️ "Traditional knowledge should be exempt from broad data mining rights, allowing communities to explicitly give or revoke permissions for its use in AI training." 🗣️ "We need public AI infrastructures that ensure diversity and regional perspectives while maintaining communities’ agency over their contributions." 🗣️ "To prevent lopsided development, policies must go beyond tools like preference signals and address broader governance and societal frameworks." 📌 About Our Guest 🎙️ Anna Tumadóttir | Creative Commons 🌐 Article | Questions for Consideration on AI & the Commons https://creativecommons.org/2024/07/24/preferencesignals/ 🌐 Anna Tumadóttir https://creativecommons.org/person/annacreativecommons-org/ Anna is the CEO of Creative Commons, an international nonprofit organization that empowers people to grow and sustain the thriving commons of shared knowledge and culture. #AI #ArtificialIntelligence #GenerativeAI…

AI lab by information labs

1
AI lab TL;DR | Carys J. Craig - The Copyright Trap and AI Policy 29:23

10 weeks之前29:23

29:23

🔍 In this TL;DR episode, Carys J Craig (Osgoode Professional Development) explains the "copyright trap" in AI regulation, where relying on copyright favors corporate interests over creativity. She challenges misconceptions about copying and property rights, showing how this approach harms innovation and access. Carys offers alternative ways to protect human creativity without falling into this trap. 📌 TL;DR Highlights ⏲️[00:00] Intro ⏲️[00:46] Q1-What is the "Copyright Trap," and why could it harm AI and creativity? ⏲️[10:05] Q2-Can you explain the three routes that lead into the copyright trap and their relevance to AI? ⏲️[22:08] Q3-What alternatives should policymakers consider to protect creators and manage AI? ⏲️[28:45] Wrap-up & Outro 💭 Q1 - What is the "Copyright Trap," and why could it harm AI and creativity? 🗣️ “To turn to copyright law is to turn to really a false friend. The idea that copyright is going to be our friend, is going to help us in this situation,(...) it's likely to do more harm than good." 🗣️ “We are imagining increasingly in these policy debates that copyright and protection of copyright owners will be a kind of counterweight to corporate power and to the sort of extractive logics of Big Tech and AI development. I think that that is misguided. And in fact, we're playing into the interests of both the entertainment industries and big tech ” 🗣️ "When we run into the copyright trap, this sort of conviction that copyright is going to be the right regulatory tool, we are sort of defining how this technology is going to evolve in a way that I think will backfire and will actually undermine the political objectives of those who are pointing to the inequities and the unfairness behind the technology and the way that it's being developed.” 🗣️ "AI industry, big tech industry and the creative industry stakeholders are all, I think, perfectly happy to approach these larger policy questions through the sort of logic of copyright, sort of proprietary logic of ownership, control, exchange in the free market, licencing structures that we're already seeing taking hold" 🗣️ "What we're going to see, I think, if we run into the copyright trap is that certainly smaller developers, but really everyone will be training the technology on incomplete data sets, the data sets that reflect the sort of big packaged data products that have been exchanged for value between the main market actors. So that's going to lessen the quality really of what's going in generally by making it more exclusive and less inclusive." 💭 Q2 - Can you explain the three routes that lead into the copyright trap and their relevance to AI? 🗣️ ""The first route that I identify is what's sometimes called the if-value-then-right fallacy. So that's the assumption that if something has value, then there should be or must be some right over it.“ 🗣️ "Because something has value, whether economic or social, doesn't mean we should turn it into property that can be owned and controlled through these exclusive rights that we find in copyright law." 🗣️ "The second route that I identify is a sort of obsession with copying and the idea that copying is inherently just a wrongful activity. (...) The reality is that there's nothing inherently wrongful about copying. And in fact, this is how we learn. This is how we create. 🗣️ "One of the clearest routes into the copyright trap is saying, well, you know, you have to make copies of texts in order to train AI. So of course, copyright is implicated. And of course, we have to prevent that from happening without permission.. (...) But our obsession with the individuated sort of discrete copies of works behind the scenes is now an anachronism that we really need to let go.” 🗣️ "Using the figure of the artist as a reason to expand copyright control, and assuming that that's going to magically turn into lining the pockets of artists and creators seems to me to be a fallacy and a route into the copyright trap." 💭 Q3 - Why is output-based remuneration better for creators, AI developers, and society? 🗣️ "The health of our cultural environment (..) [should be] the biggest concern and not simply or only protecting creators as a separate class of sort of professional actors." 🗣️ "I think what we could do is shift our copyright policy focus to protecting and encouraging human authorship by refusing to protect AI generated outputs. 🗣️ "If the outputs of generative AI are substantially similar to works on which the AI was trained, then those are infringing outputs and copyright law will apply to them such that to distribute those infringing copies would produce liability under the system as it currently exists.“ 🗣️ "There are privacy experts who might be much better placed to say how should we curate or ensure that we regulate the data on which the machines are trained and I would be supportive of those kinds of interventions at the input stage. 🗣️ “Copyright seems like a tempting way to do it but that's not what it does. And so maybe rather than some of the big collective licencing solutions that are being imagined in this context, we'd be better off thinking about tax solutions, where we properly tax big tech and then we use that tax in a way that actually supports the things that we as a society care about, including funding culture and the arts." 📌 About Our Guest 🎙️ Carys J Craig | Osgoode Hall Law School 🌐 Article | The AI-Copyright Trap https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4905118 🌐 Carys J Craig https://www.osgoode.yorku.ca/faculty-and-staff/craig-carys-j/ Carys is the Academic Director of the Osgoode Professional Development LLM Program in Intellectual Property Law, and recently served as Osgoode’s Associate Dean. A recipient of multiple teaching awards, Carys researches and publishes widely on intellectual property law and policy, with an emphasis on authorship, users’ rights and the public domain. #AI #ArtificialIntelligence #GenerativeAI…

AI lab by information labs

1
AI lab TL;DR | Ariadna Matas - Should Institutions Enable or Prevent Cultural Data Mining? 16:18

12 weeks之前16:18

16:18

🔍 In this TL;DR episode, Ariadna Matas (Europeana Foundation) discusses how the 2019 Copyright Directive has influenced text and data mining practices in cultural heritage institutions, highlighting the tension between public interest missions and restrictive approaches, and explores the broader implications of opt-outs on access, research, and the role of AI in the cultural sector. 📌 TL;DR Highlights ⏲️[00:00] Intro ⏲️[00:53] Q1-How did the 2019 Copyright Directive change the landscape for cultural heritage institutions in terms of text and data mining? ⏲️[05:07] Q2-Why are some cultural heritage institutions choosing to opt-out of text and data mining and what are the challenges involved? ⏲️[11:27] Q3-What are the broader implications of these opt-outs for research, smaller institutions, and open access to cultural content? ⏲️[14:53] Wrap-up & Outro 💭 Q1 - How did the 2019 Copyright Directive change the landscape for cultural heritage institutions in terms of text and data mining? 🗣️ "The 2019 Directive was expected to open up possibilities for cultural heritage institutions to continue their public interest mission with the help of technology." 🗣️ "The first big use of text and data mining techniques is to facilitate cultural heritage institutions’ day-to-day work.It’s rare to see cultural heritage institutions preparing datasets for public text and data mining activities." 🗣️ "More institutions are leaning toward putting barriers on data use rather than encouraging it.Instead of embracing possibilities, there’s unnecessary caution in the cultural heritage sector around AI." 💭 Q2 - Why are some cultural heritage institutions choosing to opt-out of text and data mining and what are the challenges involved? 🗣️ "The only fully legitimate reason for opting out is when the rights holder explicitly requests it." 🗣️ "Cultural heritage institutions rarely own the copyright for the materials they hold, making enforcement of opt-outs challenging." 🗣️ "Confusion about the legal framework leads some institutions to fear they must protect data from misuse." 🗣️ "By opting out, institutions risk missing out on positive uses of their data due to fear of negative outcomes." 🗣️ "Cultural heritage institutions have a public interest mission to safeguard access and encourage the use of their information." 💭 Q3 - What are the broader implications of these opt-outs for research, smaller institutions, and open access to cultural content? 🗣️ "Some organisations block access to avoid supporting big players in activities perceived as unethical." 🗣️ "Opting out doesn’t weaken monopolistic practices but harms smaller players who can’t access the data." 🗣️ "Institutions must balance the implications of their decisions on access with the potential for positive uses." 🗣️ "Aggressive crawling that disrupts public services may justify access restrictions in certain cases." 🗣️ "Overly broad decisions could limit the positive applications of text and data mining techniques on cultural heritage data." 📌 About Our Guest 🎙️ Ariadna Matas | Europeana Foundation 🌐 Article | AI ‘opt-outs’: should cultural heritage institutions (dis)allow the mining of cultural heritage data? AI ‘opt-outs’: should cultural heritage institutions (dis)allow the mining of cultural heritage data? 🌐 Ariadna Matas https://pro.europeana.eu/person/ariadna-matas Ariadna is Policy Advisor at the Europeana Foundation, an independent, non-profit organisation that stewards the common European data space for cultural heritage and contributes to other digital initiatives that put cultural heritage to good use in the world. #AI #ArtificialIntelligence #GenerativeAI…

AI lab by information labs

1
AI lab TL;DR | Martin Senftleben - How Copyright Challenges AI Innovation and Creativity 10:09

16 weeks之前10:09

10:09

🔍 In this TL;DR episode, Martin Senftleben (Institute for Information Law (IViR) & University of Amsterdam) discusses how EU regulations, including the AI Act and copyright frameworks, impose heavy burdens on AI training and development. The discussion highlights concerns about bias, quality, and fairness due to opt-outs and complex rights management systems, questioning whether these rules truly benefit individual creators. A proposal is made to focus regulatory efforts on the market exploitation phase of AI systems, ensuring compensation flows back to creative industries and authors through well-managed redistribution mechanisms. 📌 TL;DR Highlights ⏲️[00:00] Intro ⏲️[01:04] Q1-How does the EU's current approach to AI training and copyright hinder innovation and fair pay for authors? ⏲️[04:53] Q2-What’s your alternative to balance author compensation and AI development? ⏲️[06:50] Q3-Why is output-based remuneration better for creators, AI developers, and society? ⏲️[09:23] Wrap-up & Outro 💭 Q1 - How does the EU's current approach to AI training and copyright hinder innovation and fair pay for authors? 🗣️ "What policymakers try to do in this space where we try to reconcile AI innovation with traditional copyright goals is: first of all, of course we want the best AI systems and we want the least biassed AI systems." 🗣️ "The regulation puts a heavy, heavy burden on AI training by requiring to take into account rights reservations, the so-called opts-outs." 🗣️ "You might not get the best AI systems if you put all these burdens on the AI training process." 🗣️ "The moment you allow rights holders to opt out and to remove certain resources from training, then of course you no longer know whether you get the least biassed AI systems." 🗣️ "The simple fact that a big creative industry right holder receives some extra money doesn't mean that this money is passed on to the individual authors really doing the creative work." 💭 Q2 - What’s your alternative to balance author compensation and AI development? 🗣️ "If we imagine a regulation that leaves this development phase totally unencumbered by copyright burdens, you give lots of freedom for AI developers to use all the resources they think are necessary." 🗣️ "Once we have these fully developed, high potential AI systems and these systems are brought to the market, (...) You place a tax, a burden on the AI systems, not at the development stage, but at the moment where they are exploited in the marketplace.” 🗣️ "The money finally flows back in the form of compensation to the creative industry and individual authors." 💭 Q3 - Why is output-based remuneration better for creators, AI developers, and society? 🗣️ "From a European perspective, it's quite easy to propose that this should be collecting societies because we have a very well developed system of collective rights management in the area of copyright." 🗣️ "In the case of AI output, you can also use data from the systems itself: to which extent is a certain style, a certain genre prominent in prompts that users enter? What type of AI output is generated and to which extent does it resemble certain pre-existing human works and creations? What is the market share on the more general market for literary, artistic expression and so on?" 🗣️ "Traditionally, repartitioning schemes have a split between money that is directly given to individual authors and money that is given to the industry.We have a guarantee that a certain percentage of the money will directly reach the individual authors and performers, and will not stay at industry level exclusively." 📌 About Our Guest 🎙️ Martin Senftleben | Institute for Information Law (IViR) and University of Amsterdam 🌐 Article | Win-win: How to Remove Copyright Obstacles to AI Training While Ensuring Author Remuneration (and Why the European AI Act Fails to Do the Magic) Win-win: How to Remove Copyright Obstacles to AI Training While Ensuring Author Remuneration (and Why the European AI Act Fails to Do the Magic) 🌐 Martin Senftleben linkedin.com/in/martin-senftleben-2430aa5b Martin Senftleben is Professor of Intellectual Property Law and Director, Institute for Information Law (IViR), University of Amsterdam. His activities focus on the reconciliation of private intellectual property rights with competing public interests of a social, cultural or economic nature. He publishes extensively on these topics and lectures across the globe. #AI #ArtificialIntelligence #GenerativeAI…

AI lab by information labs

1
AI lab TL;DR | Mark Lemley - How Generative AI Disrupts Traditional Copyright Law 8:40

19 weeks之前8:40

8:40

🔍 In this TL;DR episode, Mark Lemley (Stanford Law School) discusses how generative AI challenges traditional copyright doctrines, such as the idea-expression dichotomy and substantial similarity test, and explores the evolving role of human creativity in the age of AI. 📌 TL;DR Highlights ⏲️[00:00] Intro ⏲️[00:54] Q1-How does genAI challenge traditional copyright doctrines and will this lead to an evolution of copyright? ⏲️[03:58] Q2-Can we expect new forms of legal recognition or protection for prompts? ⏲️[06:13] Q3-Are current copyright rules able to address authorship in genAI works or do we need new legal categories? ⏲️[08:00] Wrap-up & Outro 💭 Q1 - How does genAI challenge traditional copyright doctrines and will this lead to an evolution of copyright? 🗣️ "Copyright law has always tried to protect creative expression but is careful not to protect the idea behind a work." 🗣️ "Generative AI changes the normal economics and dynamics of creation by doing the hard work for us, like making the painting or doing the actual brushstrokes." 🗣️ "If copyright law doesn’t protect the expression created by AI rather than by a person, the question is, what, if anything, is there to copyright?" 🗣️ "Generative AI blows up the substantial similarity test because it’s unclear whether two similar works came from the same prompt or if the AI just made the same thing." 🗣️ "I might copy your prompt, input it into generative AI, and get a different output—making similarity no longer the evident marker of copying." 💭 Q2 - Can we expect new forms of legal recognition or protection for prompts? 🗣️ "We're still litigating whether the material generated by AI can be copyrighted, but we may ultimately say yes, as with photography 150 years ago." 🗣️ "Courts may get comfortable with the idea that structuring the prompt and iterating it is a form of creativity that leads to the final output." 🗣️ "In early photography, we gave copyright protection even though the machine made the image, because human judgment helped determine the outcome." 🗣️ "Prompt engineering could become more sophisticated, leading courts to see creativity in how prompts are structured and refined." 🗣️ "Sometimes I just ask a very simple question, and if that's all I contribute, I’m not sure there’s any protection." 💭 Q3 - Are current copyright rules able to address authorship in genAI works or do we need new legal categories? 🗣️ "There may be something around the creativity of prompts that will matter, but we're not there yet in terms of case law." 🗣️ "The assumption that 'I made a movie, I wrote text, so I get copyright in that work' is going to be called into question in the generative AI context." 🗣️ "Movie studios or video game companies that use AI to save money might be shocked when other people are free to copy AI-generated backgrounds." 🗣️ "Even if we get copyright protection for AI outputs, it will occupy a weird middle ground that feels different from what we’re used to." 🗣️ "There’s going to be pressure to change the law to make it align more with what copyright industries have been comfortable with, but it won’t be easy." 📌 About Our Guest 🎙️ Mark Lemley | Stanford Law School 🌐 Article | How Generative AI Turns Copyright Upside Down https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4517702 🌐 Mark Lemley https://law.stanford.edu/mark-a-lemley/ Mark is William H. Neukom Professor of Law at Stanford Law School and the Director of the Stanford Program in Law, Science and Technology. He teaches intellectual property, patent law, trademark law, antitrust, the law of robotics and AI, video game law, and remedies and he is the author of 11 books and 218 articles.…

AI lab by information labs

1
AI lab TL;DR | Jacob Mchangama - Are AI Chatbot Restrictions Threatening Free Speech? 16:10

22 weeks之前16:10

16:10

🔍 In this TL;DR episode, Jacob Mchangama (The Future of Free Speech & Vanderbilt University) discusses the high rate of AI chatbot refusals to generate content for controversial prompts, examining how this may conflict with the principles of free speech and access to diverse information. 📌 TL;DR Highlights ⏲️[00:00] Intro ⏲️[00:51] Q1-How does the high rate of refusal by chatbots to generate content conflict with the principles of free speech and access to information? ⏲️[06:53] Q2-Could AI chatbot self-censorship conflict with the systemic risk provisions of the Digital Services Act (DSA)? ⏲️[10:20] Q3-What changes would you recommend to better align chatbot moderation policies with free speech protections? ⏲️[15:18] Wrap-up & Outro 💭 Q1 - How does the high rate of refusal by chatbots to generate content conflict with the principles of free speech and access to information? 🗣️ "This is the first time in human history that new communications technology does not solely depend on human input, like the printing press or radio." 🗣️ "Limiting or restricting the output and even the ability to make prompts will necessarily affect the underlying capability to reinforce free speech, and especially access to information." 🗣️ "If I interact with an AI chatbot, it's me and the AI system, so it seems counterintuitive that the restrictions on AI chatbots are more wide-ranging than those on social media." 🗣️ "Would it be acceptable to ordinary users to say, you're writing a document on blasphemy, and then Word says, 'I can't complete that sentence because it violates our policies'?" 🗣️ "The boundary between freedom of speech being in danger and freedom of thought being affected is a very narrow one." 🗣️ "Under international human rights law, freedom of thought is absolute, but algorithmic restrictions risk subtly interfering with that freedom.(...) These restrictions risk being tentacles into freedom of thought, subtly guiding us in ways we might not even notice." 💭 Q2 - Could AI chatbot self-censorship conflict with the systemic risk provisions of the Digital Services Act (DSA)? 🗣️ "The AI act includes an obligation to assess and mitigate systemic risk, which could be relevant here regarding generative AI’s impact on free expression." 🗣️ "The AI act defines systemic risk as a risk that is specific to the high-impact capabilities of general-purpose AI models that could affect public health, safety, or fundamental rights." 🗣️ "The question is whether the interpretation under the AI act would lean more in a speech protective or a speech restrictive manner." 🗣️ "Overly broad restrictions could undermine freedom of expression in the Charter of Fundamental Rights, which is part of EU law." 🗣️ "My instinct is that the AI act would likely lean in a more speech-restrictive way, but it's too early to say for certain." 💭 Q3 - What changes would you recommend to better align chatbot moderation policies with free speech protections? 🗣️ "Let’s use international human rights law as a benchmark—something most major social media platforms commit to on paper but don’t live up to in practice." 🗣️ "We showed that major social media platforms' hate speech policies have undergone extensive scope creep over the past decade, which does not align with international human rights standards." 🗣️ "It's conceptually more difficult to apply international human rights standards to an AI chatbot because my interaction is private, unlike public speech." 🗣️ "We should avoid adopting a 'harm-oriented' principle to AI chatbots, especially when dealing with disinformation and misinformation, which is often protected under freedom of expression." 🗣️ "It's important to maintain an iterative process with AI systems, where humans remain responsible for how we use and share information, rather than placing all the responsibility on the chatbot." 📌 About Our Guest 🎙️ Jacob Mchangama | The Future of Free Speech & Vanderbilt University 𝕏 https://x.com/@JMchangama 🌐 Article | AI chatbots refuse to produce ‘controversial’ output − why that’s a free speech problem https://theconversation.com/ai-chatbots-refuse-to-produce-controversial-output-why-thats-a-free-speech-problem-226596 🌐 The Future of Free Speech https://futurefreespeech.org 🌐 Jacob Mchangama http://jacobmchangama.com Jacob Mchangama is the Executive Director of The Future of Free Speech and a Research Professor at Vanderbilt University. He is also a Senior Fellow at The Foundation for Individual Rights and Expression (FIRE) and author of “Free Speech: A History From Socrates to Social Media”.…

AI lab by information labs

1
AI lab TL;DR | Jurgen Gravestein - The Intelligence Paradox 14:49

24 weeks之前14:49

14:49

🔍 In this TL;DR episode, Jurgen Gravestein (Conversation Design Institute) discusses his Substack blog post delving into the ‘Intelligence Paradox’ with the AI lab 📌 TL;DR Highlights ⏲️[00:00] Intro ⏲️[01:08] Q1-The ‘Intelligence Paradox’: How does the language used to describe AI lead to misconceptions and the so-called ‘Intelligence Paradox’? ⏲️[05:36] Q2-‘Conceptual Borrowing’: What is ‘conceptual borrowing’ and how does it impact public perception and understanding of AI? ⏲️[10:04] Q3-Human vs AI ‘Learning’: Why is it misleading to use the term ‘learning’ for AI processes and what this means for the future of AI development? ⏲️[14:11] Wrap-up & Outro 💭 Q1-The ‘Intelligence Paradox’ 🗣️ What’s really interesting about chatbots and AI is that for the first time in human history, we have technology talking back at us, and that's doing a lot of interesting things to our brains. 🗣️ In the 1960s, there was an experiment with Chatbot Eliza, which was a very simple, pre-programmed chatbot (...) And it showed that when people are talking to technology, and technology talks back, we’re quite easily fooled by that technology. And that has to do with language fluency and how we perceive language. 🗣️ Language is a very powerful tool (...) there’s a correlation between perceived intelligence and language fluency (...) a social phenomenon that I like to call the ‘Intelligence Paradox’. (...) people perceive you as less smart, just because you are less fluent in how you’re able to express yourself. 🗣️ That also works the other way around with AI and chatbots (...). We saw that chatbots can now respond in extremely fluent language very flexibly. (...) And as a result of that, we perceive them as pretty smart. Smarter than they actually are, in fact. 🗣️ We tend to overestimate the capabilities of [AI] systems because of their language fluency, and we perceive them as smarter than they really are, and it leads to confusion (...) about how the technology actually works. 💭 Q2-‘Conceptual Borrowing’ 🗣️ A research article (...) from two professors, Luciano Floridi and Anna Nobre, (...) explaining (...) conceptual borrowing [states]: “through extensive conceptual borrowing, AI has ended up describing computers anthropomorphically, as computational brains with psychological properties, while brain and cognitive sciences have ended up describing brains and minds computationally and informationally, as biological computers." 🗣️ Similar to the Intelligence Paradox, it can lead to confusion (...) about whether we underestimate or overestimate the impact of a certain technology. And that, in turn, informs how we make policies or regulate certain technologies now or in the future. 🗣️ A small example of conceptual borrowing would be the term “hallucinations”. (...) a common term to describe when systems like chatGPT say something that sounds very authoritative and sounds very correct and precise, but is actually made up, or partly confabulated. (...) this actually has nothing to do with real hallucinations [but] with statistical patterns that don’t match up with the question that’s being asked. 💭 Q3-Human vs AI ‘Learning’ 🗣️ If you talk about conceptual borrowing, “machine learning” is a great example of that, too. (...) there's a very (...) big discrepancy between what learning is in the psychological terms and the biological terms when we talk about learning, and then when it comes to these systems. 🗣️ So if you actually start to be convinced that LLMs are as smart and learn as quickly as people or children (...) you could be over attributing qualities to these systems. 🗣️ [ARC-AGI challenge:] a $1 million USD prize pool for the first person that can build an AI to solve a new benchmark that (...) consists of very simple puzzles that a five-year old (...) could basically solve. (...) it hasn't been solved yet. 🗣️ That’s, again, an interesting way to look at learning, and especially where these systems fall short. [AI] can reason based on (...) the data that they've seen, but as soon as it (..) goes out of (...) what they've seen in their data set, they will struggle with whatever task they are being asked to perform. 📌 About Our Guest 🎙️ Jurgen Gravestein | Sr Conversation Designer, Conversation Design Institute (CDI) 𝕏 https://x.com/@gravestein1989 🌐 Blog Post | The Intelligence Paradox https://jurgengravestein.substack.com/p/the-intelligence-paradox 🌐 Newsletter https://jurgengravestein.substack.com 🌐 CDI https://www.conversationdesigninstitute.com 🌐 Profs. Floridi & Nobre's article http://dx.doi.org/10.2139/ssrn.4738331 🌐 Jurgen Gravestein https://www.linkedin.com/in/jurgen-gravestein Jurgen Gravestein is a writer, conversation designer and AI consultant. He works at the CDI, the world’s leading training and certification institute in conversational AI. He also runs a successful Substack newsletter “Teaching computers how to talk”.…

AI lab by information labs

1
AI lab TL;DR | Stefaan G. Verhulst - Are we entering a Data Winter? 12:55

27 weeks之前12:55

12:55

🔍 In this TL;DR episode, Dr. Stefaan G. Verhulst (The GovLab & The Data Tank) discusses his Frontiers Policy Labs contribution on the urgent need to preserve data access for the public interest with the AI lab 📌 TL;DR Highlights ⏲️[00:00] Intro ⏲️[01:13] Q1-‘Data Winter’: Can you provide a brief overview of your concept of 'Data Winter' and why you believe we are on the brink of entering one? ⏲️[05:05] Q2-Generative AI-nxiety: What are some of the most significant challenges currently hindering public access to social media and climate data, and the effects of Generative AI-nxiety? ⏲️[07:49] Q3-‘Decade for Data’: Could you outline what the “Decade for Data” initiative entails and how it could transform data stewardship and collaboration? ⏲️[12:25] Wrap-up & Outro 💭 Q1-‘Data Winter’ 🗣️ At the time of an AI summer, when everyone suddenly is excited about the potential of generative AI (...) for public interest purposes, (...) we are actually entering a data winter. 🗣️ What I’ve witnessed the last few months, and that’s mainly as a result of advances in artificial intelligence, is that we actually see a backtracking of the progress that we’ve made in society as it relates to opening up data for public interest purposes. 🗣️ Social media platforms such as X, but also Facebook, have closed down access to some of their data for research and for data journalism purposes as well. 🗣️ Science data, such as climate science data, which was typically open science, has now become commercialised and is becoming proprietary data enclosed for many in society. 🗣️ The initial data that was available for training data has now also become much harder to access, a result of concerns that some of that data has been extracted without a return to the data holder. 💭 Q2-Generative AI-nxiety 🗣️ Some of the data that typically was available through APIs has now been closed off, and so some are calling this the post-API environment that we're currently in, where data was easily available through an API now is actually much harder to access unless one pays for it. 🗣️ New licensing is being used to actually shield off the data for public interest purposes as well. So there are a whole range of vehicles that exist to enclose data that actually makes it much harder to access it for reuse. 🗣️ We see a decline in access to Wikipedia, a decline in people accessing Wikipedia, and a decline in people contributing to Wikipedia, mainly because they fear that whatever they contribute will be used as training fodder for generative AI purposes. 🗣️ Initiatives like Wikipedia, which are to a large extent the main source of a lot of the training data of generative AI services, are currently also suffering from AI extraction because they are dependent on voluntary contributions by the audience and the participants. 🗣️ As a result, we are entering a data winter, which if we are not careful (...) may actually affect the AI summer that we currently have as well. 💭 Q3-‘Decade for Data’ 🗣️ I’ve been calling for, together with others, such as the United Nations University, a Decade for Data, which is a typical way the United Nations often operates, to feature a problem and then have a well-defined strategy to address that problem. 🗣️ A Decade for Data would have multiple components, one being advancing data collaboration, where you actually have new models of data being shared, including data commons, which can be updated in the current AI environment. 🗣️ We need a new reimagined profession of data stewards that are individuals or teams who have the sophistication and competencies to provide access to data in a systematic, sustainable, and responsible manner. 🗣️ A Decade for Data would also involve rethinking data governance and embedding digital self-determination in data governance to go beyond the current paradox of consent, facilitating access in a way that aligns with perceptions, expectations, and preferences of communities. 🗣️ Establishing a social license for reuse is key, where you understand the preferences and expectations of communities and individuals, translating that into a social license so that data can be reused in a way that is trusted and aligned with community expectations. 📌 About Our Guest 🎙️ Dr. Stefaan G. Verhulst | Co-Founder, The GovLab & The Data Tank 🌐 Frontiers Policy Labs | Are We Entering a Data Winter? https://policylabs.frontiersin.org/content/commentary-are-we-entering-a-data-winter 🌐 The Data Tank https://datatank.org 🌐 GovLab https://thegovlab.org 🌐 Dr. Stefaan G. Verhulst https://www.linkedin.com/in/stefaan-verhulst Dr. Stefaan G. Verhulst co-founded several research organisations, including the GovLab (New York) and The DataTank (Brussels). He focuses on using advances in science and technology, including data and AI, to improve decision-making and problem-solving and has been recognized as one of the 10 Most Influential Academics in Digital Government globally.…

AI lab by information labs

1
AI lab – AI in Action | Episode 03: AI Tokenization 7:28

29 weeks之前7:28

7:28

Let’s talk about AI tokenization in this third episode of our AI in Action series. Tokenization is actually pretty interesting, especially if you ever wondered how these fancy AI machines understand the stuff we type and say and produce things when we give them prompts. Next time you're marvelling at an AI-generated text, remember it's all about those tiny tokens, dancing together in a complex symphony of language and prediction.…

AI lab by information labs

1
AI lab TL;DR | Bertin Martens - The Economics of GenAI & Copyright 13:41

30 weeks之前13:41

13:41

🔍 In this TL;DR episode, Dr. Bertin Martens (Bruegel) discusses his working paper for the Brussels-based economic think tank on the economic arguments in favour of reducing copyright protection for generative AI inputs and outputs with the AI lab * 9:44: Mr Martens intended to say "humans" instead of machines 📌 TL;DR Highlights ⏲️[00:00] Intro ⏲️[01:21] Q1-Balancing Innovation & Rights Can the TDM opt-out right hinder innovation and economic growth, and what does it mean as regards the power of copyright holders vs. the potential societal benefits of generative AI? ⏲️[05:42] Q2-Licensing Impact on EU AI Competitiveness What are the implications of licensing for genAI as regards competitiveness and quality of models and potential economic disadvantages for EU AI developers? ⏲️[09:11] Q3-GenAI's Impact on Creative Industries & Economy Looking at outputs, how could genAI impact the creative industries and the broader economy, and what are your thoughts on how policy should evolve to reflect this? ⏲️[13:08] Wrap-up & Outro 💭 Q1-Balancing Innovation & Rights 🗣️ Copyright is an economic policy tool to stimulate investment in the production of artwork, and granting an exclusive copyright to an author avoids free writing on that artwork that would undermine the incentives to invest in its production. 🗣️ The optimal scope of copyright protection should balance, on the one hand, the welfare losses from this exclusive right given to an author against the welfare gains for society from stimulating investment in new and innovative productions. 🗣️ Both [copyright] overprotection and underprotection are bad. They will hamper innovation and reduce the economic efficiency of copyright. 🗣️ Generative AI opens up new and much cheaper possibilities to produce new and innovative artwork, and also has applications in a wide variety of other sectors outside the media sector and across the economy. 🗣️ The AI Act and the copyright law in Europe give priority to the private interest of copyright holders over the wider interest of society, and I don't think that's a good thing and we should change that. 💭 Q2-Licensing Impact on EU AI Competitiveness 🗣️ Generative AI models require vast amounts of training data to develop the model and to have a high-quality model. And already today we observe that the largest and most advanced models are running out of high-quality human edited text for model training. 🗣️ There is still sufficient supply of low-quality text data, for instance from social media, or from the transposition of voice data into text, or even from synthetic data. But all these low-quality sources reduce the quality of generative AI models. 🗣️ Imposing copyright licensing requirements on text data for model training will further shrink the available supply of text data for model training, and that will further reduce the quality of these models. 🗣️ Only the biggest tech companies can actually afford to negotiate the licensing fees and pay those fees to copyright holders, while smaller AI startups cannot afford this and are pushed out of the market. 🗣️ Pushing smaller AI startups out of the market is bad for competition, bad for innovation in the AI setting, and this is not the way we want to go. 💭 Q3-GenAI's Impact on Creative Industries & Economy 🗣️ Generally, copyright law worldwide grants copyright only to human authors of artwork, not to machine-produced artwork. With the arrival of generative AI models, however, that has changed, and for the first time in human history, a machine can produce artwork output. 🗣️ From an economic perspective, there is no need to grant copyright to AI-produced artwork because the marginal cost of producing generative AI output is actually very close to zero (...) and the risk of free riding, therefore, is very limited. 🗣️ The human labor that goes into designing a prompt set that you feed into a generative AI model is costly, and this prompt set is human artwork and could indeed receive copyright protection, just like any other human design, text or computer code. 📌 About Our Guest 🎙️ Dr. Bertin Martens | Senior fellow at Bruegel and non-resident research fellow at TILEC, Tilburg University 🌐 Bruegel | Economic Arguments in Favour of Reducing Copyright Protection for Generative AI Inputs and Outputs https://www.bruegel.org/working-paper/economic-arguments-favour-reducing-copyright-protection-generative-ai-inputs-and 🌐 Bruegel https://www.bruegel.org 🌐 Tilburg Law & Economics Centre (TILEC) https://www.tilburguniversity.edu/research/institutes-and-research-groups/tilec 🌐 Dr. Bertin Martens https://www.bruegel.org/people/bertin-martens Dr. Bertin Martens is a Senior fellow at Bruegel and a non-resident research fellow at the Tilburg Law & Economics Centre (TILEC, Tilburg University). He has worked on digital economy issues as a senior economist at the European Commission's Joint Research Centre for over a decade until April 2022. Before that, he was deputy chief economist for trade policy at the EC.…

AI lab by information labs

1
AI lab TL;DR | Alexander Peukert - Copyright in the Artificial Intelligence Act–A Primer 15:36

38 weeks之前15:36

15:36

🔍 In this TL;DR episode, Prof. Dr. Alexander Peukert (Goethe University Frankfurt am Main) discusses his primer on copyright in the EU AI Act with the AI lab 📌 TL;DR Highlights ⏲️[00:00] Intro ⏲️[01:26] Q1-Merging copyright & AI regulation: What challenges arise from merging copyright law and AI regulation? How might this impact legislation, compliance, and enforcement? ⏲️[06:08] Q2-AI Act copyright targets: Who are the main targets of the AI Act's copyright-related obligations? ⏲️[09:33] Q3-AI Act copyright obligations: What key copyright-related obligations does the AI Act impose on AI model providers? How should training content summaries and TDM opt-out mechanisms be implemented? ⏲️[14:44] Wrap-up & Outro 💭 Q1 - Merging Copyright & Ai Regulation 🗣️ Any copyright infringement triggers remedies. (...) In the EU AI Act context, it’s very different because the EU AI Act establishes systemic compliance obligations . 🗣️ AI model providers have to put in place a general copyright policy . Whether that policy is sufficient or not is then a question which is pretty difficult to answer and not straightforward . 🗣️ When we merge copyright with the AI regulation, (...) this is also true for the DSA, (...) you have to ask: at what point is a systemic compliance obligation violated? Only then do you have a violation of this AI regulation . 🗣️ The AI Act is primarily enforced by public authorities (...). That might become a challenge for rightholders because they were used to enforce their rights at their will . Now they have to make sure that the [EC or national authorities act]. 🗣️ For the first time, (...) public authorities enter the copyright environment to a very significant extent through the EU AI Act . 💭 Q2 - AI Act Copyright Targets 🗣️ The specific copyright obligations are only addressed to general-purpose AI model providers . (...) AI systems that are then built upon it (...), which eventually create the output, are not subject to specific copyright obligations. 🗣️ The EU legislature (...) said: we focus on the [ general-purpose AI ] models because they are the very basis of all systems, and if we target them (...), then we make sure that any kind of system, generative AI [is] copyright-compliant . 💭 Q3 - AI Act Copyright Obligations 🗣️ The EU AI Act [ obliges ] AI model providers to program their crawlers, who crawl the Internet, to collect data for [AI] training (...) in a manner that the opt-out of copyright holders is respected . 🗣️ There’s a market for AI training data , which is based on these copyright rules in connection with the EU AI Act. 🗣️ You have to put in place a copyright policy . (...) One potential consequence (...) might be a kind of moderation obligation so that you have to make sure that not only the training (...) but also the eventual output is copyright-compliant . 🗣️ It might become difficult for the [ general-purpose] AI model provider to moderate the output of systems that another company has built on [ their ] model . (...) I see a potential problem in the implementation of these copyright obligations. 🗣️ The [ training content ] summary need not be granular so that you mention each and every URL that you have mined, (...) it suffices to describe the content in a narrative way . So what kind of databases have you searched or crawled? 🗣️ The [ training content ] summary (...) is a tool to enable rightholders to figure out whether they were mined and perhaps whether their preventive measures were circumvented and (...) potentially sue for copyright infringement . 📌 About Our Guest 🎙️ Prof. Dr. Alexander Peukert | Full Professor of Civil, Commercial and Information Law at Goethe University Frankfurt am Main 🌐 GRUR International | Copyright in the Artificial Intelligence Act – A Primer https://academic.oup.com/grurint/article-abstract/73/6/497/7675073 https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4771976 🌐 Prof. Dr. Alexander Peukert http://www.jura.uni-frankfurt.de/peukert/ http://ssrn.com/author=1244916 Alexander Peukert (pronounce as Poikert) has since 2009 been full professor of civil, commercial and information law at Goethe University Frankfurt am Main. He studied law and obtained his Dr. iur. (s.c.l.) at the University of Freiburg (1993-1999). After his second state examination (2001), he practiced law in a Berlin law firm specializing in copyright and media law. From 2002 to 2009, he was senior research fellow and head of the U.S. department at the Max Planck Institute for Intellectual Property and Competition Law in Munich.…

AI lab by information labs

1
AI lab TL;DR | Thomas Margoni - Copyright Law & the Lifecycle of Machine Learning Models 13:29

39 weeks之前13:29