Google has created MusicLM, AI which is impressive but has copyright problems

[highplainsdem]

Cross-post from General Discussion.


No kidding. Copyright problems serious enough to keep from releasing it so far.

Problems which other AI like ChatGPT and Midjourney also have, and I hope artists who have been or will be affected by them can stop them with lawsuits. Tempting as these tools are for businesses that don't want to pay real artists, and individuals who don't want take the time to learn skills and create something of their own when they can ask AI to do it and pretend they're being creative.

Google's new AI can create music from someone humming a melody, giving a text description, or even asking for music to go with a painting.

The research paper that introduced it: https://google-research.github.io/seanet/musiclm/examples/

Fortune article I ran across first: https://fortune.com/2023/01/29/chatgpt-for-music-google-ai-musiclm-generates-music-from-hums-and-text-prompts/ and Google News link to it: https://news.google.com/articles/CBMibmh0dHBzOi8vZm9ydHVuZS5jb20vMjAyMy8wMS8yOS9jaGF0Z3B0LWZvci1tdXNpYy1nb29nbGUtYWktbXVzaWNsbS1nZW5lcmF0ZXMtbXVzaWMtZnJvbS1odW1zLWFuZC10ZXh0LXByb21wdHMv0gFyaHR0cHM6Ly9mb3J0dW5lLmNvbS8yMDIzLzAxLzI5L2NoYXRncHQtZm9yLW11c2ljLWdvb2dsZS1haS1tdXNpY2xtLWdlbmVyYXRlcy1tdXNpYy1mcm9tLWh1bXMtYW5kLXRleHQtcHJvbXB0cy9hbXAv?hl=en-US&gl=US&ceid=US%3Aen

“Think of MusicLM as the ChatGPT for music,” tweeted entrepreneur Martin Uetz, adding, “I can’t wait for this to go mainstream.”

Less eager might be musicians who’ve spent decades mastering their instruments, just as illustrators and graphic artists have been angered by A.I. tools that create impressive images from mere text prompts.

Among those A.I. art tools are Midjourney, Stable Diffusion, and DALL-E 2. One man recently used Midjourney to illustrate a children’s book. Impressed with the tool, he shared his experience on social media—and was stunned by the backlash from illustrators. And last year, an image generated with Midjourney won a prize at an art festival, which also angered artists.

The problem artists have with such tools is that they train themselves on a massive collection of digitized artworks without consent. A lawsuit recently filed in San Francisco by working artists describes Stable Diffusion and Midjourney as “collage tools that violate the rights of millions of artists.”

Twitter thread with music samples starts with the tweet below. Thread Reader page at https://threadreaderapp.com/thread/1619081383477137408.html .

Link to tweet

Two-part Forbes article on AI and the music industry:

https://www.forbes.com/sites/cindygordon/2023/01/29/is-google-displacing-musicians-with-its-new-generative-ai-system-music-lm-part-1-of-a-2-part-article-series/?sh=459566f01859

https://news.google.com/articles/CBMic2h0dHBzOi8vd3d3LmZvcmJlcy5jb20vc2l0ZXMvY2luZHlnb3Jkb24vMjAyMy8wMS8yOS9pbXBhY3Qtb2YtYWktb24tdGhlLW11c2ljLWluZHVzdHJ5LXBhcnQtMi1vZi1hLXR3by1wYXJ0LXNlcmllcy_SAXdodHRwczovL3d3dy5mb3JiZXMuY29tL3NpdGVzL2NpbmR5Z29yZG9uLzIwMjMvMDEvMjkvaW1wYWN0LW9mLWFpLW9uLXRoZS1tdXNpYy1pbmR1c3RyeS1wYXJ0LTItb2YtYS10d28tcGFydC1zZXJpZXMvYW1wLw?hl=en-US&gl=US&ceid=US%3Aen

From Part 1:

Although not yet released due to copyright issues, MusicLM is a signal of what’s to come in using AI to generate music from text descriptions.

What is the impact of taking all the musical know-how and ingesting into an AI brain with only asking questions expressing your musical tastes? Who needs to be a musician to entertain when we will soon be able to create our own music even more easily and tap into the human genius of every muscian that has a recording?

-snip-

Although Google, Meta, Microsoft, OpenAI and many other AI market leaders will continue to advance the frontiers of every industry using AI. We as humans have an ethical responsibility to think harder about the world we want to create and leave for future generations.

If you are a Board Director or CEO or C-Level in the Musical Industry, learning more about AI is a business imperative to understand the long term effects of AI in the musical industry and shaping the world you want to protect as “human brains and musical talent” have value and we are rapidly commoditizing precious creative DNA into bits and bytes that will have major impacts to our musicians’ creative ownership entitlements.

From Part 2:

There are many pros and cons to AI in the music industry. One may argue that we are inventing a new art form by reducing the time or expenses to create music.

Cons are many in terms of impact to a precious fabric of what makes up our humanitarian society that has been infused with music and stories for centuries. The implication of copyright is enormous as AI only can learn from music and that’s other’s music so the ability to pay for training AI models from musical pieces is a complex and a not well regulated area.

-snip-

What AI models won’t have is the next unique human voice on file to learn from. Perhaps there in lies the opportunity for future singers to ensure their digital identities are securely encoded so a future song cannot be easily produced, and also having cyber security software detect even if a hint of a Beyonce song or beat has been integrated into a future song generating a red alert for musician copyright protection.

In conclusion, it is imperative that the legal and ethical frameworks hasten to protect the music industry on the creative use of AI-enabled tools. Currently producers and musicians either using AI technology are vulnerable as laws will pass and major lawsuits will unfold, and to producers and artists in the music industry, increasing your own knowledge and lobbying for improved laws is key.

[highplainsdem]

4. It's a safe bet politicians and special interests of all types will

try using AI to try to make their messages more convincing. Combine AI and data mining and you get very carefully crafted messages.

Scammers of all types will also find AI useful. Sheer clumsiness has often alerted people to scams in the past. But AI can get rid of the grammar and spelling errors, etc.

Voice cloning, using audio deepfakes, has already been used. One of many stories, this from a couple of years ago:

https://www.forbes.com/sites/thomasbrewster/2021/10/14/huge-bank-fraud-uses-deep-fake-voice-tech-to-steal-millions/?sh=127aca6a7559

Another:

https://www.khou.com/article/news/local/pearland-father-daughter-voice-cloning-scam/285-1a7b0e9b-ccca-488c-afff-04dcec0d1e77

With enough data about a person gleaned from social media and the internet, AI can impersonate people fairly convincingly.

As for why people didn't get enough warnings or a chance to vote on it - some people and companies saw profits for themselves in it. And advances in AI have been rapid. And it was never a top issue for most people.

[rog]

2. You can't just put random sounds together and call it music.

The first example was moderately interesting, but it started with a solid melody to work from. The other examples sounded like 'monkeys with typewriters'.

This is something many non-musicians, including scientists, don't get.

I have had many students, brilliant in other fields, such as medicine, finance, history, science, etc, who assume learning music is trivial, and that it might make a nice hobby that doesn't require much effort. I'd say most of them have been surprised.

Edited to add: I have had a couple of technical 'discussions' with ChatGPT in which the 'bot' replies with a level of assurance and 'expert' tone that inspires confidence, but often strays far off the mark. The bot is generally apologetic and 'seems' to correct itself when I catch it, but it does so by using my request for clarity as the basis for its answer, as if it knew what it was talking about all along, and just 'slipped up'.

I'd have to interact with MusicLM in real time to form an opinion, but based on the examples above, I'm not impressed ... so far.

♫rog♫

[highplainsdem]

3. ChatGPT does make mistakes on facts, as CNET found out when

the site's management tried using it to write articles, sort of burying the fact the articles were AI-written, and what was supposed to be human editorial oversight didn't happen at all or was careless.

But it does sound glib. But that glibness makes it appealing to students who don't want to do the work themselves.

ChatGPT is already causing a lot of problems in education. It's been able to pass exams at business and law schools: https://www.cnn.com/2023/01/26/tech/chatgpt-passes-exams/index.html

It's also a huge threat to writers of both nonfiction and fiction, though some fiction writers are already using AI to do much of their work for them. See https://www.democraticunderground.com/100217495259 . And some scientists are using it to write papers, which is already recognized as a problem: https://www.nature.com/articles/d41586-023-00191-1

And AI art tools threaten artists, which is why they're suing, as one of the articles I quoted in the OP pointed out.

MusicLM is doing much more than putting random sounds together, or there wouldn't be this concern about it.

AI will threaten filmmakers and actors, too, as deep fakes get better and better. ChatGPT has already been used to write and direct a short film: https://nofilmschool.com/2022/12/filmmakers-use-ai-write-and-direct-short-film-and-it-actually-makes-some-sense . They used human actors and needed humans to do the filming, but as AI can better simulate human actors, that won't be necessary.

Some artists will still be able to make some money with live performances.

But imagine a future where AI plus holograms could generate a bar band with musicians as talented as any that ever lived - because their characteristics are drawn from records and video of real artists, shuffled just enough to make something "new" - and able to interact with the audience, take requests, etc. A bar band that will never show up late, won't need breaks, and might not cost the bar owner much at all.

And all of this threat to human artists will come from AI's theft of human artists' work.