9 ms·

Prediction: regulation will require AI companies to have copyright licenses granted for training data. There will be legal models trained on datasets that have

Prediction: regulation will require AI companies to have copyright licenses granted for training data. There will be legal models trained on datasets that have copyright granted, and illegal black market models trained on copyrighted materials. However, it will be really difficult to enforce and prove what data was used for training in each.

kneebonian 3y ago
Further predicition: There will be a certification process that will be required to prove you only used data you have legally obtained. Running or hosting a model will be grounds for a fine, or possible criminal penalties.
- mock-possum 3y ago
  I think it’ll be more like using samples in music - probably not a big deal until you’re making a bunch of money, and then people come after you for licensing fees.
- mtillman 3y ago
  A certification would happen in the EU first.
  - InCityDreams 3y ago
    ...with the states and the uk to follow, rather like the gdpr-like privacy setup in Cali.
metry 3y ago
This could be true.
oh_sigh 3y ago
Prediction: clean room AI will forever lag behind AI that hoovers up whatever data it can get it's hands on.
- CharlesW 3y ago
  It won't matter if using the output has sufficient risk. Go ahead and train that model on Disney content and sell the results, then see what happens.
  - TobyTheDog123 3y ago
    No one will be able to prove the source of the data - was it Disney content, or was it someone copying the animation style of Disney content?
    - ChatGTP 3y ago
      Eventually you will need to show sources and I’m sure we’ll have ways to extract sources from the models. ma these things are not magic after all.
      - amanaplanacanal 3y ago
        I’m not sure why sources matter at all. In copyright you compare the end result, not how you got there.
        blibble 3y ago
        this isn't true at all the process the work was generated through is of primary importance for copyright the existence of the concept of "derivative work" should make this obvious see the reverse engineering of the IBM BIOS for another example
        CuriouslyC 3y ago
        There are specific criterion for fair use: https://fairuse.stanford.edu/overview/fair-use/four-factors/ https://fairuse.stanford.edu/overview/fair-use/four-factors/. None of them are process specific. Maybe bone up on the law rather than getting yourself in any deeper.
rvz 3y ago
I'm from the future and this is true. There will certainly be AI regulations and licensed AI models and it had already happened with the Shutterstock and O̶p̶e̶n̶AI.com partnership. Stable Diffusion and the Getty Images lawsuit will end with a settlement and licensing will be the option to go with.
- jquery 3y ago
  I'm also from the future (2028 to be precise), and enforcing copyright on AI-generated content has become near-impossible. It's working about as well as DRM has in stopping video and music piracy. There's now BitTorrent-like software for AI, which allows groups of individuals to train and run their own AI on whatever they please, and they seem to have little regard for copyright. In fact, underground music from these creators is taking off, and despite the law requiring all AI-assisted songs to cite their influences for royalty payments, they're simply refusing to comply. The music is so good that the public has no appetite for prosecuting the individuals making it. In fact, most people are using these AIs to create their own custom music stations. The old guard is stomping its gold-encrusted shoes in anger, but the youth just don't care. There's a cool song called 'Neural Harmony.' The creators used a mixture of classical and electronic music as input, but they never disclosed the specific songs they were influenced by. This has made it impossible for the original artists to claim royalties, and yet the public can't get enough of its sick beatz. There's even an AI-generated album called 'Digital Renaissance.' The creators claim to have used thousands of songs from various genres as inspiration, but they never provided a list of the songs or artists. The album has gained a massive following and has even been featured in several popular playlists. There was briefly an attempt to prosecute them but the case was dropped after public outcry.
  - blibble 3y ago
    > It's working about as well as DRM has in stopping video and music piracy. quite well then? streaming services have essentially replaced music and video piracy and steam (essentially streaming games) vastly reduced video game piracy
    - 3y ago
mrbungie 3y ago
Prediction: Some nation-state that cares nothing about IP (i.e. China) wins the AI race in this domain making these regulations totally useless.
- EntrePrescott 3y ago
  Another prediction would be: maybe Western states start blocking Chinese businesses that don't comply with their laws. After all, China already blocks the Western businesses that don't implement the spying/surveillance measures the Chinese communist party wants. The more the Chinese communist government continues pushing a war course against Taiwan makes that even more likely, e.g. in the context of sanctions.
  - mrbungie 3y ago
    Good luck blocking such a big market. Money simply speaks louder.
    - vkou 3y ago
      Blocking imports is actually very easy.
  - londons_explore 3y ago
    But it won't matter. If an integrated economy can be more efficient by copying each other internally, then it will eventually be more competitive at business worldwide, despite trade levies aimed against it.
- gumballindie 3y ago
  Yeah China hasn't been respecting IP for a long time and it's still well behind. We shouldn't become China just to win a race - there are rules to the game you know? If everyone's content is stolen and resold by an AI then why would anyone want to create content?
  - hammyhavoc 3y ago
mrtksn 3y ago
I think we will settle on proprietary AI required to obtain licenses on trained data. Any model trained on public domain data will be completely public domain too. So I imagine, someone will train a model to do high end law work and they will have to train it on the data produced by the people they hire to create it. This of course assumes that the society will have the same structure as of today. What I actually think it will happen is, we will completely delegate all our work to machines and the concepts of ownership will vanish as everything for exception of land will be in abundance. I imagine in the future we will fight each other over apartments in cool areas or trade it for some kind of social credit which we generate by impressing other humans. No apartment will be worse than the other but the proximity natural wonders, cultural centres and networks will be the paramount. After all, it's all about how we pick our sexual mates and social status. I'm sorry you don't like it but There's no way the society functions the same once resources and servants are in abundance. Your bank account is relevant only when there's a scarcity and the current scarcity can be gone once machines are autonomous in enough areas to convert the material all around us into things we need using the practically limitless energy from the sun.
rvba 3y ago
I wonder if some sort of copyright traps cannot be created? Fake music pieces. Map-makers use arificial "trap streets" ( https://en.m.wikipedia.org/wiki/Trap_street https://en.m.wikipedia.org/wiki/Trap_street ) to show that their maps were copied. Dictionaries add fictious entries ( https://en.m.wikipedia.org/wiki/Fictitious_entry https://en.m.wikipedia.org/wiki/Fictitious_entry ). Similar idea could be used with music?
- SMAAART 3y ago
  Related (and funny): https://twitter.com/brdskggs/status/1637114268876144640 https://twitter.com/brdskggs/status/1637114268876144640
oytis 3y ago
AI leaders will be located in China or India then.
wefarrell 3y ago
This could be a boon for publishers since they'll be able to act as the gatekeepers of almost all professionally produced content.
kouteiheika 3y ago
Or companies will just move to a country where training on copyrighted data is 100% legal and allowed (like e.g. in Japan which had an explicit amendment to its copyright law to allow it), although it'll be interesting to see how such models would be treated internationally.
- kmeisthax 3y ago
  The EU also had its copyright law amended to allow data mining on copyrighted data sets. And I would not be surprised if US courts decide that training an AI is fair use. The goal, after all, is to create a machine that draws new images.
nomoreusernames 3y ago
[dead]
MikusR 3y ago
Prediction: that will lead to a couple huge copyright guzzlers that will sue everyone trying to enter their market.

kneebonian 3y ago

Further predicition: There will be a certification process that will be required to prove you only used data you have legally obtained. Running or hosting a model will be grounds for a fine, or possible criminal penalties.

mock-possum 3y ago
I think it’ll be more like using samples in music - probably not a big deal until you’re making a bunch of money, and then people come after you for licensing fees.
mtillman 3y ago
A certification would happen in the EU first.
- InCityDreams 3y ago
  ...with the states and the uk to follow, rather like the gdpr-like privacy setup in Cali.

metry 3y ago

This could be true.

oh_sigh 3y ago

Prediction: clean room AI will forever lag behind AI that hoovers up whatever data it can get it's hands on.

CharlesW 3y ago
It won't matter if using the output has sufficient risk. Go ahead and train that model on Disney content and sell the results, then see what happens.
- TobyTheDog123 3y ago
  No one will be able to prove the source of the data - was it Disney content, or was it someone copying the animation style of Disney content?
  - ChatGTP 3y ago
    Eventually you will need to show sources and I’m sure we’ll have ways to extract sources from the models. ma these things are not magic after all.
    - amanaplanacanal 3y ago
      I’m not sure why sources matter at all. In copyright you compare the end result, not how you got there.
      - blibble 3y ago
        this isn't true at all the process the work was generated through is of primary importance for copyright the existence of the concept of "derivative work" should make this obvious see the reverse engineering of the IBM BIOS for another example
        CuriouslyC 3y ago
        There are specific criterion for fair use: https://fairuse.stanford.edu/overview/fair-use/four-factors/ https://fairuse.stanford.edu/overview/fair-use/four-factors/. None of them are process specific. Maybe bone up on the law rather than getting yourself in any deeper.

rvz 3y ago

I'm from the future and this is true. There will certainly be AI regulations and licensed AI models and it had already happened with the Shutterstock and O̶p̶e̶n̶AI.com partnership. Stable Diffusion and the Getty Images lawsuit will end with a settlement and licensing will be the option to go with.

jquery 3y ago
I'm also from the future (2028 to be precise), and enforcing copyright on AI-generated content has become near-impossible. It's working about as well as DRM has in stopping video and music piracy. There's now BitTorrent-like software for AI, which allows groups of individuals to train and run their own AI on whatever they please, and they seem to have little regard for copyright. In fact, underground music from these creators is taking off, and despite the law requiring all AI-assisted songs to cite their influences for royalty payments, they're simply refusing to comply. The music is so good that the public has no appetite for prosecuting the individuals making it. In fact, most people are using these AIs to create their own custom music stations. The old guard is stomping its gold-encrusted shoes in anger, but the youth just don't care. There's a cool song called 'Neural Harmony.' The creators used a mixture of classical and electronic music as input, but they never disclosed the specific songs they were influenced by. This has made it impossible for the original artists to claim royalties, and yet the public can't get enough of its sick beatz. There's even an AI-generated album called 'Digital Renaissance.' The creators claim to have used thousands of songs from various genres as inspiration, but they never provided a list of the songs or artists. The album has gained a massive following and has even been featured in several popular playlists. There was briefly an attempt to prosecute them but the case was dropped after public outcry.
- blibble 3y ago
  > It's working about as well as DRM has in stopping video and music piracy. quite well then? streaming services have essentially replaced music and video piracy and steam (essentially streaming games) vastly reduced video game piracy
  - 3y ago

mrbungie 3y ago

Prediction: Some nation-state that cares nothing about IP (i.e. China) wins the AI race in this domain making these regulations totally useless.

EntrePrescott 3y ago
Another prediction would be: maybe Western states start blocking Chinese businesses that don't comply with their laws. After all, China already blocks the Western businesses that don't implement the spying/surveillance measures the Chinese communist party wants. The more the Chinese communist government continues pushing a war course against Taiwan makes that even more likely, e.g. in the context of sanctions.
- mrbungie 3y ago
  Good luck blocking such a big market. Money simply speaks louder.
  - vkou 3y ago
    Blocking imports is actually very easy.
- londons_explore 3y ago
  But it won't matter. If an integrated economy can be more efficient by copying each other internally, then it will eventually be more competitive at business worldwide, despite trade levies aimed against it.
gumballindie 3y ago
Yeah China hasn't been respecting IP for a long time and it's still well behind. We shouldn't become China just to win a race - there are rules to the game you know? If everyone's content is stolen and resold by an AI then why would anyone want to create content?
- hammyhavoc 3y ago

mrtksn 3y ago

I think we will settle on proprietary AI required to obtain licenses on trained data. Any model trained on public domain data will be completely public domain too. So I imagine, someone will train a model to do high end law work and they will have to train it on the data produced by the people they hire to create it. This of course assumes that the society will have the same structure as of today. What I actually think it will happen is, we will completely delegate all our work to machines and the concepts of ownership will vanish as everything for exception of land will be in abundance. I imagine in the future we will fight each other over apartments in cool areas or trade it for some kind of social credit which we generate by impressing other humans. No apartment will be worse than the other but the proximity natural wonders, cultural centres and networks will be the paramount. After all, it's all about how we pick our sexual mates and social status. I'm sorry you don't like it but There's no way the society functions the same once resources and servants are in abundance. Your bank account is relevant only when there's a scarcity and the current scarcity can be gone once machines are autonomous in enough areas to convert the material all around us into things we need using the practically limitless energy from the sun.

rvba 3y ago

I wonder if some sort of copyright traps cannot be created? Fake music pieces. Map-makers use arificial "trap streets" ( https://en.m.wikipedia.org/wiki/Trap_street https://en.m.wikipedia.org/wiki/Trap_street ) to show that their maps were copied. Dictionaries add fictious entries ( https://en.m.wikipedia.org/wiki/Fictitious_entry https://en.m.wikipedia.org/wiki/Fictitious_entry ). Similar idea could be used with music?

SMAAART 3y ago
Related (and funny): https://twitter.com/brdskggs/status/1637114268876144640 https://twitter.com/brdskggs/status/1637114268876144640

oytis 3y ago

AI leaders will be located in China or India then.

wefarrell 3y ago

This could be a boon for publishers since they'll be able to act as the gatekeepers of almost all professionally produced content.

kouteiheika 3y ago

Or companies will just move to a country where training on copyrighted data is 100% legal and allowed (like e.g. in Japan which had an explicit amendment to its copyright law to allow it), although it'll be interesting to see how such models would be treated internationally.

kmeisthax 3y ago
The EU also had its copyright law amended to allow data mining on copyrighted data sets. And I would not be surprised if US courts decide that training an AI is fair use. The goal, after all, is to create a machine that draws new images.

nomoreusernames 3y ago

[dead]

MikusR 3y ago

Prediction: that will lead to a couple huge copyright guzzlers that will sue everyone trying to enter their market.