Live data from Hacker News

Google's advanced music generation model and two new AI experiments

deepmind.google

221–230 of 361 posts

Re: Google's advanced music generation model and two new AI experiments

#221

Earlier quoted context omitted.

Most of today's popular music is heavily computer based. Previous musicians said the same things, electric guitars are not musical instruments, synthesizers have no soul, music made on computer DAWs is artificial and of no value, .... The kids will embrace 100% AI generated music, the parents will be "they don't make music like they used to", forgetting that their parents said the same things about their music.

> Most of today's popular music is heavily computer based. Hence my worry that I am gatekeeping. For example in 2007, making good sounding music required at least a studio. If you wanted to record 4 microphones at the same time, you needed expensive hardware. Doing any kind of music editing required a huge machine and expensive software with expensive plugins. Now I can mix and apply effects in real time to 24 tracks…

> For example in 2007, making good sounding music required at least a studio. If you wanted to record 4 microphones at the same time, you needed expensive hardware. Doing any kind of music editing required a huge machine and expensive software with expensive plugins.

A lot of people point to Garageband (released 2004) as having a big democratizing effect on digital audio production. I think v2 in 2005 supported recording 8 tracks simultaneously. Not an especially sophisticated tool (and arguably responsible for the built-in loops being heard in way too many popular songs), but a big change at a time when the alternative might have been a computer with ProTools and dedicated PCI cards.

https://www.rollingstone.com/pro/features/apple-garageband-m...

Re: Google's advanced music generation model and two new AI experiments

#222

Earlier quoted context omitted.

You are not being deprived of any rights. You just don't have them exclusively. Make music, or don't. Up to you. Either way, there will be machines also making music. So, what?

How are people going to become professional musicians if it stops being a career?

You could make the same argument against recordings (and people did). It would take away jobs from the musicians. Yet, 136 years later and people are still creating music and playing instruments.

Re: Google's advanced music generation model and two new AI experiments

#223
post #50

Earlier quoted context omitted.

But isn't music more than just the music track? People like to listen to the same music, go to concerts, share it with others, share the experience of audience/artist performance. Moments where music is truly important in one's life will never be replaced by meaningless bit fiddling. I cannot imagine people caring about books, music, art created by bit fiddling. They will always care about things created by other peo…

Well, this could have the potential to make infinite covers of your favorite song. But think about it, people used to watch the same movies in DVD. The movies they loved. Now people watch and infinity of short random videos (tik tok, instagram)... Technology changes the habits of people, specially in young people. Think about it, one friend sends a song auto-generated from some text, it is "a song made by someone", a…

I still don't see the pull. Take a band like The National. I'm 100% certain most of the listeners absolutely do not care about covers at all.

This musical world where we generate random stuff that will be consumed equivalently to current music just does not feel real. What will probably happen is that people just won't listen to music or get passionate about it at all. That might already be happening because gaming is something what music was for the generation when pull of the games was not as strong.

Young people today will probably never care about Beatles or Bob Dylan or contemporary bands. The music is just not the main interest.

Just imagine AI making games and everyone in the industry losing their job. It just does not work that way. Product is not the end. Music does not work that way. Music is more than the product.

Any creative history is focused on a group of people and their audience, not on the songs/books/poems in general, but the whole experience of creativity and its outputs.

Re: Google's advanced music generation model and two new AI experiments

#224
There is an extremely weird juxtaposition between the idea of watermarking the outputs so AI-generated samples can be detected if they make up even part of a song (which is arguably considerable as fair use of even copyrighted material if only a single bar is sampled), and the statement:

> Within the experiment, a limited set of creators will be able to use Dream Track for producing a unique soundtrack with the AI-generated voice and musical style of artists including Alec Benjamin, Charlie Puth, Charli XCX, Demi Lovato, John Legend, Sia, T-Pain, Troye Sivan, and Papoose*.

So... we're watching you and we'll know if you so much as sample one bar to create something of your own. But our preferred creators are having so much fun using AI exclusively to make the entire composition, directly ripping off the personal style of these artists who said it's OK!

I also found it hilarious that they claim the idea of embedding a watermark in the audio by adding the image to the 2d spectrogram is a "novel method ... unlike anything that exists today, especially in the context of audio." Considering Aphex Twin did it in 1999...

* Apparently it was removed from the game, and his Wikipedia doesn't even mention it, but, yes, Papoose, of the "You just mad cuz I'm stylin' on ya!" track from GTA IV.

Re: Google's advanced music generation model and two new AI experiments

#225
post #199

Earlier quoted context omitted.

I wish to subscribe to your Post-Scarcity Party newsletter. It also makes me wonder about agriculture. It's difficult to envision a post-scarcity future where food production is still manual-intensive. Mandatory picking service, as a civic obligation? And there are a lot of food types that require manual picking. UBI would therefore effectively end production of those types of food. (E.g. more grain, less fruit) Also…

UBI would likely only be enough for the barest of necessities of survival. Gardening in an empty lot would be an obvious way to supplement that income, with the same benefits of a gym membership. The point of UBI is that via automation and reduction in cheap energy, opportunities for useful economic activity will be harder to find. In the best case, most people would have a mixture of reliable but marginal income (ga…

> In the best case…

Why is that the best case? In my “best case”, all of everyone’s needs are met, people pick and grow more intense crops because they want to, artists just enjoy making art, and “income” from work just isn’t really a thing anymore. My best case is everyone has what they need and can access what they want.

Re: Google's advanced music generation model and two new AI experiments

#226
I might have a different take, but the introduction of ChatGPT last year actually prompted me to dive head first into learning an instrument aided also by a fantastic live performance by Yo-Yo Ma.

In one interview I watched, Yo-Yo said he strived to play pieces perfect until his 20s when in the middle of a concert he realized he was playing perfectly and it was boring. From that moment he decided to dedicate his musical performances to making a connection with the audience. His concert was incredible to experience a room of 1000 people all having their seats shaken by a human being bowing a string without any electronics involved. People were getting drawn in, then jumping back, and getting transported outside the physical. Hard to describe.

And that was when i realized that no matter how perfect a computer plays, humans will always want to hear other humans play. We have had player pianos for over a century and not to mention a long history of sculptures that play music and even now you can get a virtually unlimited amount of extremely high quality near perfect replication recording of the best performances out there.

But still, I just went to buy tickets to a classical music event on a monday night to a solo concert by a not-famous player that each seat cost around $100 and they were sold out.

For me personally, learning an instrument has been a tremendous experience. Even the interaction with my teacher gives me weekly human contact that has become scarce in the post-covid world filled with remote work.

Spending months of practice just to play a song poorly is actually an incredibly freeing and humbling experience. I had my recital with a bunch of people watching me, where I messed up, I played out of tune, but it was an incredible flood of human connection that i never really had prior.

I think worrying about AI taking over music is overblown. Unless the music is going to be generated to manipulate people somehow, but the reality is that in 100 years people will still crave live performances. Just like for the past 100 years, paradoxically, the access and lowered barriers of entry, has actually increased accessed to classical and live performances. People complaining that people today don’t like “music with passion” anymore are living in some fairytale land where they imagine every single person 200 years ago was skipping down to the opera house. The reality was that 200 years ago access to music was incredibly limited and most people only experienced it in church context. If that (rural churches had a lot less going on musically).

People will always crave to see other humans’ accomplishments. Music, paintings, etc are not just an end product, but the more you understand the difficulties and journey of the person producing it, the deeper your appreciation becomes.

Re: Google's advanced music generation model and two new AI experiments

#227

Earlier quoted context omitted.

But isn't music more than just the music track? People like to listen to the same music, go to concerts, share it with others, share the experience of audience/artist performance. Moments where music is truly important in one's life will never be replaced by meaningless bit fiddling. I cannot imagine people caring about books, music, art created by bit fiddling. They will always care about things created by other peo…

I think the bigger concern should be around things like backing scores, soundtracks, and other instrumental music. Things that you might have paid a composer to do, or a local orchestra to perform, can be done by AI and a synth. This has similar energy to the writer’s strike - save money by removing the first layer of human effort. For some things (like the backing music in a commercial) this is sufficient, and for o…

> I think the bigger concern should be around things like backing scores, soundtracks, and other instrumental music. Things that you might have paid a composer to do, or a local orchestra to perform, can be done by AI and a synth.

This was already somewhat accomplished by synths and music production software. Take any keyboardist with knowledge of composition and you can get a very realistic sounding orchestra, played by one person. Synths make air/string instruments sound extremely realistic even when played on a keyboard.

Sounds like that kind of work is almost equivalent to clerical work, yet maybe requires a bit more training effort. Although, do you really need to be a violin maestro to play for a movie soundtrack in an orchestra, are your skills being used at their limits?

So I guess we can just see what happened with clerical work and how the society responds. I guess clerks just didn't have enough "voting power".

Re: Google's advanced music generation model and two new AI experiments

#228

Earlier quoted context omitted.

UBI would likely only be enough for the barest of necessities of survival. Gardening in an empty lot would be an obvious way to supplement that income, with the same benefits of a gym membership. The point of UBI is that via automation and reduction in cheap energy, opportunities for useful economic activity will be harder to find. In the best case, most people would have a mixture of reliable but marginal income (ga…

> In the best case… Why is that the best case? In my “best case”, all of everyone’s needs are met, people pick and grow more intense crops because they want to, artists just enjoy making art, and “income” from work just isn’t really a thing anymore. My best case is everyone has what they need and can access what they want.

It would require a lot of strong social structres. Because a lot of important tasks (like maintaining the floor cleaning equipment in the robot-tractor tyre factory) just really isnt that interesting but still needs to be done. I can see ways motivation can come from others. I'd argue my motivation to persist in my work isn't (directly) because of the money I am paid but rather due to social effects of my colleagues, and the money only contributed to building up this colleague social structure in the first place (alongside the potential value our work brings being an almost equally powerful motivator)

keeping these things going when people lose interest will be a challenge.

Re: Google's advanced music generation model and two new AI experiments

#229

Earlier quoted context omitted.

Being a professional musician is already a fraught career with low rates of success. This is even more true if you exclude people that are working with modern digital systems. Not a lot of jobs available for orchestral musicians, for example. But there still are people making art with older techniques. There is a greater variety of art being made today than ever before in history. The development of new techniques ha…

This is different though, because the end game of AI music isn't better music technology for musicians, but replacing musicians. I'm worried we're going to go into a dark age where society forgets how to make truly creative new art due to lack of incentives, and the AIs won't make anything new due to lack of anything creative to train on. We'll just regurgitate the pre-AI stuff forever.

If we're fast-forwarding the tape, why not directly go to the ultimate conclusion that the end game is AI replacing humans, period?

This is not a slippery slope; new art will be created regardless of incentives and skilled people who learn to augment their craft with AI will probably create better works of art.

Re: Google's advanced music generation model and two new AI experiments

#230

Earlier quoted context omitted.

Regarding the first point of being buried under an avalanche of crap, we can solve this problem through many ways. We can still filter by human music. Or we can use better recommendation systems (automated or p2p/word-of-mouth). Regarding whether it is "my" expression or not, I consider expression as a way of making my feelings or thoughts known. This grants me an additional way of making my feelings or thoughts know…

> we can solve this problem through many ways You realise that this is a problematic situation created by technology, that was designed as such first, right? > This grants me an additional way of making my feelings or thoughts known. Not really. Because again, this is a generator: what it outputs according to your prompt may _please_ you, or seem like it matches what you think you wanted, but this is still an output…

> You realise that this is a problematic situation created by technology

I'm saying this technology is a net positive for me. Sure, problems can arise from applying this net-positive technology but even considering them, they're a net positive. Especially in this case where we already have solutions for large amounts of crap content created by humans today. I think they'll scale well to crap content created by AI tomorrow (eg: Recommendation systems, p2p sharing etc).

Regarding your second point. I think we just have different perspectives on this. I see this technology as allowing me to do "strictly more" than I could before. You say that it's not "my expression" but I say it is because I'm able to bring forth what's in my mind (according to me!). You say it's not what I had in mind but compare against this scenario: I have an idea for a song and use a musical instrument. The output that I produce is quite far from my intent as I'm untrained. With this model, I can get a lot closer to my intent compared to using an instrument since I'm untrained. I still have the option of later learning an instrument if I'm unsatisfied.

Overall, I lose nothing and gain a nice power and still retain the option to learn an instrument if I wish. Thus it's a clear net positive for me that allows me to express more right now.

Post reply on HN