AI Secrets is your new weekly update on how AI is changing our everyday lives. Our experts will keep it clear and simple, so you can stay ahead of the game. This week we are focussing on Music.Share this with anyone you want to keep up to date.
15 Minutes of Deepfaked Fame ⌚ ❌
Last week, a new viral music hit took TikTok by storm. Heart On My Sleeve - by the anonymous user @Ghostwriter977 - became the latest focal point for AI hype thanks to its apparent use of Canadian rapper Drake’s vocals to generate the song’s lyrics. It also became a focus for copyright lawyers thanks to a question that could potentially spell doom for the AI industry: If a song contains no direct samples of Drake’s work, can it be said to infringe his copyright?
In Heart On My Sleeve’s case, the answer was not immediately obvious, because Ghostwriter977’s identity and methods were shrouded in mystery. For a while, pundits even speculated that the song was merely an edgy promotional stunt by Drake’s label. The truth was less dramatic, but also more alarming for the record industry; the song almost certainly originated from a Discord channel called AI Hub, which offers straightforward DIY tools for budding deepfakers to roll their own AI vocalists. At the time of writing, the channel has almost 40,000 members, posting hundreds of new deepfaked recordings every day.
AI Hub’s server rules are explicit, stating “violating anyone’s intellectual property or rights will result in instant ban”, but nobody’s entirely sure what that means in practice. AI “outputs” from voice training are not direct copies of intellectual property, but the “inputs” that create them unquestionably are. Whether this is legal is a question of “fair use”, an exemption in US law that allows for limited creative use of other people’s intellectual property, and is crucial for legal arguments seeking to justify the training of AIs on copyrighted material.
To get a sense of how thorny this issue is, consider Google’s position. As the owners of YouTube, they have negotiated an uneasy legal truce with the entertainment industry over uploads of copyrighted material, and are keen to avoid upsetting this valuable applecart. On the other hand, they have staked the company’s future on generative AI products, whose legality depends heavily on fair use exemptions that justify the copyrighted material in the AI’s training data. Surprisingly, the solution to this seemingly no-win situation may come from old media; the US Supreme Court is currently ruling on whether Andy Warhol’s use of other artists’ work was “fair” or not, and the verdict could determine whether we all experience his promised “15 minutes of fame” as part of an AI’s training data.
Of course, some artists are more open to the idea of AI collaborations; Grimes has declared she will happily split royalties with any artist who deepfakes her voice, but while the legal confusion over what constitutes fair use lingers, it’s unclear whether anybody needs her permission. In the meantime, there are upsides to the confusion; don’t miss our practical guide to enlisting celebrities as your personal council of advisors in ChatGPT!
Special Limited Edition!
Puzzle lovers will adore this unique custom cube. Each of its six sides features a different design. With 43 quintillion unique positions it’s more complex than any spy story! Get ready to have fun and be challenged.
The Internet is quickly filling up with paragraphs that begin with the tell-tale phrase “As a language model”, as AI-generated text floods every service from Amazon’s bookstores to social media.
Researchers have discovered a surprising fact about GPT-3.5; it is more honest in some languages than in others, and seems particularly prone to misinformation when answering questions in Chinese.
Runway - an app that applies generative AI filters to your phone’s video recordings - has launched on iOS and kickstarted the next wave of video-to-video AI apps, but: Is it any good?
Bolting AI features onto existing popular apps is not a guaranteed success: Snapchat’s rollout of its “My AI” feature has been a disaster with users, leading to an enormous spike in one-star reviews for the product on app stores.
Google’s struggling LLM has had an unusually good week, with the release of new features including the ability to write and edit code, while Wired have written an excellent guide to writing better Bard prompts.
The latest AI to hit the Internet is a little different; HuggingChat is the new offering from open source AI underdogs Hugging Face, and promises a more transparent - if less effective - chatbot alternative.
