Generative AI copyright problems continue - Weekly News Roundup - Issue #449
Plus: EU checks Microsoft-OpenAI partnership; bioengineered air-purifying plant; plant-based 3D-printed eel; digital resurrection consent; Elvis is coming back thanks to AI and VR; and more!
Welcome to Weekly News Roundup Issue #449. This week, the main story focuses on the ongoing problems caused by copyright infringement claims in the generative AI industry.
In other news, the European Union is set to take a closer look at the relationship between OpenAI and Microsoft. Bioengineers have created an air-purifying plant as well as a plant-based 3D-printed eel, and Elvis Presley is making a comeback thanks to AI and VR.
The new year did not start great for the generative AI industry. Before 2023 ended, The New York Times filed a lawsuit against OpenAI and Microsoft for copyright infringement, claiming that OpenAI’s large language models GPT-3.5 and GPT-4 were trained using the newspaper's content without authorization, potentially causing billions in damages. In the lawsuit, The New York Times provides evidence that ChatGPT outputs verbatim text from the newspaper’s articles and makes a strong case that OpenAI used copyrighted material in training their large language model.
That’s not the first time ChatGPT was caught generating answers using copyrighted materials. Last year, researchers found that if you prompt ChatGPT to repeat one word indefinitely, the chatbot at some point will start outputting its own training data, which included sensitive information such as email addresses and phone numbers. Using only $200 worth of OpenAI API tokens, the researchers were able to extract over 10,000 unique verbatim memorized training examples. OpenAI “fixed” the issue by making these kinds of prompts a violation of the terms of service.
Behind the scenes, OpenAI started to lobby for copyright exemption, claiming that ChatGPT would not be possible without copyrighted materials. In evidence submitted to the UK House of Lords communications and digital committee, OpenAI said: “Because copyright today covers virtually every sort of human expression – including blog posts, photographs, forum posts, scraps of software code, and government documents – it would be impossible to train today’s leading AI models without using copyrighted materials. Limiting training data to public domain books and drawings created more than a century ago might yield an interesting experiment, but would not provide AI systems that meet the needs of today’s citizens.”
OpenAI is not the only generative AI company dealing with copyright issues. In an article published in IEEE Spectrum, Gary Marcus and Reid Southen show multiple examples of instances where Midjourney v6 and DALL-E 3 generate copyrighted images (I recommend at least checking out the images in their article, the similarities are evident). They have shown that not only do these text-to-image generators produce copyrighted images when directly prompted (example prompt: “Thanos infinity war, 2018, screenshot from a movie, movie scene, 4k, bluray --ar 16:9 --v 6.0”) but also when the prompt does not mention any titles or character names. According to Marcus and Southen, this is strong evidence that Midjourney and DALL-E 3 were trained on copyrighted materials.
The evidence of copyrighted materials used in the training of generative AI systems, as shown by The New York Times and Marcus’ and Southen’s research, could mean serious problems not only for the companies involved but also for the entire generative AI industry. When these systems were in the research phase, copyright infringement was not a big issue. But now, when OpenAI is valued at $100 billion and Midjourney charges $10 to $120 per month for their services, the conversation is different. Without proper credit and compensation, what these companies do is theft, some people argue. OpenAI replies by saying it complies with all copyright laws when it trains its models and that legally copyright law does not forbid training.
So, what can be done to make everyone happy? In an article on his personal Substack, Gary Marcus proposes that OpenAI and other companies just pay licensing fees to the publishers, just like streaming services such as Netflix or Spotify pay the copyright owners to stream their content. According to a report published in The Information, OpenAI is trying to do that by offering publishers between $1 million to $5 million per year to use their content. But as the report notes, this is a tiny amount even for small publishers.
As Marcus and Southen point out in their article in IEEE Spectrum, the situation in which generative AI companies have found themselves in is similar to the early days of Napster and peer-to-peer file-sharing networks. After Napster was brought down by massive copyright lawsuits, a new crop of streaming services like Spotify emerged. These streaming services now share the revenue with publishers and artists (but we need to acknowledge this model is still not ideal). Something similar could happen with generative AI, creating a fairer relationship between AI companies, artists and publishers in which copyrights are respected. In any case, it looks like the future of generative AI will be shaped not only by brilliant engineers and researchers developing more sophisticated algorithms but also in the courtrooms by lawyers and judges deciding if these companies violated copyright laws.
If you enjoy this post, please click the ❤️ button or share it.
I warmly welcome all new subscribers to the newsletter this week. I’m happy to have you here and I hope you’ll enjoy my work. A heartfelt thank you goes to everyone who joined as paid subscribers this week.
The best way to support the Humanity Redefined newsletter is by becoming a paid subscriber.
If you enjoy and find value in my writing, please hit the like button and share your thoughts in the comments. Additionally, please consider sharing this newsletter with others who might also find it valuable.
For those who prefer to make a one-off donation, you can 'buy me a coffee' via Ko-fi. Every coffee bought is a generous support towards the work put into this newsletter.
Your support, in any form, is deeply appreciated and goes a long way in keeping this newsletter alive and thriving.
🦾 More than a human
Resurrection consent: It's time to talk about our digital afterlives
A new study, which investigated public attitudes towards digitally resurrecting the deceased using AI, focusing on the importance of consent, has revealed a significant difference in acceptance based on whether the deceased had consented, with 58% approval for resurrection with consent versus 3% without. The study highlights the need for legal discussions and potential consent-based regulations for digital resurrection. This issue is becoming more pressing as AI chatbots capable of replicating personalities are increasingly used, raising ethical and legal questions about posthumous digital personas. If you want to learn more about the digital afterlife and how it can change how we view death, check out my article, Digital Immortality: An Afterlife in Digital Clouds, which explores this topic in detail.
Soft robotic, wearable device improves walking for individual with Parkinson’s disease
Researchers have developed a soft, wearable robot to aid people living with Parkinson’s disease in walking without freezing, a common and debilitating symptom. This exosuit, designed to be worn around the hips and thighs, assists with leg swings, allowing for longer strides, effectively eliminating freezing during indoor walking and enabling the participant to walk faster and further.
🧠 Artificial Intelligence
EU checking if Microsoft’s OpenAI investment falls under merger rules
The partnership between Microsoft and OpenAI is very close. The tech giant invested over $10B into OpenAI and serves as its main technological partner, providing vast computing resources through Azure. Microsoft also has the exclusive right to offer OpenAI models as part of their Azure offering. This partnership is so close that the European Union is examining it under its merger regulations. This scrutiny by the EU follows similar inquiries by the UK's competition authority and Germany's Federal Cartel Office, with concerns about Microsoft's influence on OpenAI and the broader impact on market competition.
AI Elvis to make virtual reality comeback in London show
Elvis Presley is coming back to the stage, thanks to virtual reality and AI. The project, named Elvis Evolution, will use AI, holographic projection, augmented reality, and live theatre to recreate events in Presley's life and music, said Layered Reality, an immersive entertainment company developing the show. “It’s going to be a joyous celebration of Elvis's life: the man, the music, and his cultural legacy,” Layered Reality founder and chief executive Andrew McGuinness told Reuters. The first show will take place in London later this year before going on a world tour in cities such as Las Vegas, Tokyo, and Berlin.
Gaming voice actors blindsided by 'garbage' union AI deal
The US actors' union SAG-AFTRA has allowed AI firm Replica Studios to use AI-generated voices in video games, but the deal was not welcomed by voice actors. This deal, which SAG-AFTRA claims ensures "fully informed consent and fair compensation" for its members, was signed following a months-long strike in 2023 over AI use in film and television. However, many voice actors feel betrayed, arguing that the deal contradicts the goals of the strike and fearing AI could replace them. The agreement requires consent from actors to use voices based on their likeness and allows them to deny perpetual use without consent.
If you're enjoying the insights and perspectives shared in the Humanity Redefined newsletter, why not spread the word?
🤖 Robotics
Knightscope receives Authority to Operate within the U.S. government
Knightscope Inc., a security robot provider, has received an Authority to Operate from the Federal Risk and Authorization Management Program, endorsed by the U.S. Department of Veterans Affairs. This certification enables Knightscope to deploy its K5 Autonomous Security Robot in federal settings and list it on the FedRAMP Marketplace for federal agencies. Knightscope’s robots are already patrolling some malls in the United States. Recently, one of them was leased to the New York Police Department to patrol the Times Square subway station. Knightscope robots have also become a meme after one of them drove itself into a fountain.
1X, robotic startup backed by OpenAI, receives $100M in funding
1X Technologies, a Norwegian startup robotics startup backed by OpenAI, among others, announced it has raised $100 million in its series B round of funding from a long list of other big names, including Samsung. 1X is one of many robotics startups working on humanoid robots. The investment is adding more hype to the prospect of seeing commercial humanoid robots being available within years.
The Global Project to Make a General Robotic Brain
Unlike generative AI tools like ChatGPT, which are trained on vast internet data, robots require specific robotic-interaction data. Data that is scarce and task-specific. The RT-X project, a collaboration between Google, the University of California, Berkeley, and 32 other robotics labs, was created to address this problem and to explore the potential of general-purpose robots. RT-X aims to overcome the data problem by pooling data from various robots to enable a single neural network to control different types of robots, a concept known as cross-embodiment. The project has already assembled a large dataset from nearly a million robotic trials involving 22 types of robots, making it the largest open-source dataset of real robotic actions. Early results show that robots trained on this diverse data can perform a range of tasks more successfully than those using lab-specific methods.
🧬 Biotechnology
Neoplants shows off its Neo P1 bio-engineered air-purifying plant
At CES 2024, the French startup Neoplants showcased its bioengineered houseplants, designed to function as effective air purifiers for home and office spaces. These plants, known as Neo P1, are claimed to be 20 times more efficient than traditional houseplants like Epipremnum aureum (also known as "devil’s vine") in removing pollutants, particularly volatile organic compounds (VOCs) like formaldehyde and toluene, from the air.
Steakholder Foods introduces plant-based 3D-printed eel
Steakholder Foods, a company specialising in cultivated meat products, has introduced what it claims is “the world’s first” plant-based 3D-printed eel. The company claims that its product “accurately replicates the complex texture of eel” through precise layering and “a unique combination of materials” in its proprietary 3D printing technology. The eel product is currently based on plant materials, with plans to incorporate cultivated eel cells in the future, “as economies of scale allow price-competitive cell development.”
Thanks for reading. If you enjoyed this post, please click the ❤️ button or share it.
Humanity Redefined sheds light on the bleeding edge of technology and how advancements in AI, robotics, and biotech can usher in abundance, expand humanity's horizons, and redefine what it means to be human.
A big thank you to my paid subscribers, to my Patrons: whmr, Florian, dux, Eric, Preppikoma and Andrew, and to everyone who supports my work on Ko-Fi. Thank you for the support!