8 ms·

Training algorithms on copyrighted data not illegal: US Supreme Court

327 points by alok-g 7y ago

MrStonedOne 7y ago
From a 2016 refusal to review a case
Enginerrrd 7y ago
This isn't nearly as broad a precedent as the title sounds. They used some pretty reasonable tests of copyright infringement to conclude that no such infringement occured.
breck 7y ago
Anyone know of any organizations working to repeal Intellectual Monopoly laws? I know there are organizations that try to counter the influence of the IM industry like EFF and FSF, but I’m looking for groups that have come out and said Intellectual Monopoly laws need to go, period.
- bitwize 7y ago
  Pirate Parties worldwide. Sometimes they even get seats in parliament.
herniatedeel 7y ago
SCOTUS denied the petition for writ of certiorari, thereby leaving the 2nd Circuit's ruling in Google's favor intact. However, the 2nd Circuit's ruling is not binding on any other federal circuits. Also, as Enginerrd stated, the holding is not nearly as broad as the article makes it out to be. The holding was: 1. Google’s unauthorized digitizing of copyright-protected works, creation of a search functionality, and display of snippets from those works are non-infringing fair uses. The purpose of the copying is highly transformative, the public display of text is limited, and the revelations do not provide a significant market substitute for the protected aspects of the originals. Google’s commercial nature and profit motivation do not justify denial of fair use. 2. Google’s provision of digitized copies to the libraries that supplied the books, on the understanding that the libraries will use the copies in a manner consistent with the copyright law, also does not constitute infringement. Based on the above holding, I think the article's conclusion is a stretch for general training algorithms using copyrighted data because: (1) there would not be a library supplying the information to the training algorithm, (2) there would be no similar display of snippets, and (3) we do not know if a training algorithm would provide a market substitute for the copyrighted data.
- bluGill 7y ago
  While the decision is only binding in the 2nd circuit, the precedent is admissible in other courts. If this goes to trial in a different circuit you can bring the finding to the judge who will consider it - it won't be binding, but he will consider it. If it goes to appeal in a different circuit the next circuit will reference this in their decision - if they decide the 2nd is wrong they will be very clear why the they think the 2nd circuit is wrong when they make ruling (and this can in turn be re-submitted to the 2nd circuit who might change their mind if the reasoning is good enough). If this goes to the supreme court in the future they will read this decision and it will influence them - again they can decide either way.
  - herniatedeel 7y ago
    Yes, the 2nd Circuit's decision is persuasive authority for other circuits. However, that's not what the article claims. The article claims that SCOTUS ruled when it, in fact, did not.
    - dragonwriter 7y ago
      SCOTUS ruled on the cert petition; what people may not understand is that while that is a ruling, it is (and there is explicit precedent on this point) not one which has precedential weight (even as persuasive authority) as regards the merits of the issues addressed in the lower court ruling.
      - daveFNbuck 7y ago
        It's not hard to see why people are not understanding this correctly. The HN link text: > Training algorithms on copyrighted data not illegal: US Supreme Court The sub-heading for the article: > Training algorithms on copyrighted data is not illegal, according to the United States Supreme Court.
      - angry_octet 7y ago
        Isn't it unlikely that a case will be granted cert if appeals courts in different circuits are in agreement? I.e. not a circuit split? So while not legally binding, it might be in practice indicative. I wonder, does this mean I can scrape Instagram/Facebook for photos and use them for face recognition? Is that 'fair use'? Is an Instagram post a publication?
prepend 7y ago
If this legal theory is correct and upheld, possession of data will be pretty important. So all those data you licensed for one reason or another might be able to be used to train. I think the logic makes sense because imagine if humans were prevented from getting ideas from watching movies. It seems similar to not letting AI watch every movie ever and learn.
- jacquesm 7y ago
  I don't think the judgment supports the conclusion in the article.
- tzs 7y ago
  It sounds like to get the data into their AI, Google had to make copies. They digitized the physical books, and then trained using those copies they made. Also, their system included excerpts from the books which it could retrieve and show users. Hence, they had to use fair use to justify it. I think if you could train the AI without having to make a copy first, such as having the AI read the physical books directly, or in the case of your movie example having the AI watch the movie on a TV hooked up to a DVD player playing a copy of the movie on DVD that you bought from a retailer authorized by the copyright owner to sell such DVDs, you might not even need to make a fair use argument. The definitions section of the US copyright statutes, 17 USC 101 [1], defines "copies" like this: > “Copies” are material objects, other than phonorecords, in which a work is fixed by any method now known or later developed, and from which the work can be perceived, reproduced, or otherwise communicated, either directly or with the aid of a machine or device. The term “copies” includes the material object, other than a phonorecord, in which the work is first fixed. and "fixed" is defined like this: > A work is “fixed” in a tangible medium of expression when its embodiment in a copy or phonorecord, by or under the authority of the author, is sufficiently permanent or stable to permit it to be perceived, reproduced, or otherwise communicated for a period of more than transitory duration. An AI reading or watching the work as one of many many works in order to learn weights for a neural net does not result in a material object from which the work can be perceived, reproduced, or otherwise communicated. Thus, there is no copy, and hence no copyright issue. [1] https://www.law.cornell.edu/uscode/text/17/101 https://www.law.cornell.edu/uscode/text/17/101
fyp 7y ago
We probably need a technical rather than legal solution for this. The problem is that generative algorithms are susceptible to accidental memorization so you can't guarantee that the output will be transformative. For example, play with https://talktotransformer.com https://talktotransformer.com to see how many well known pieces of text it can spew out verbatim. It is very prone to derailing into a harry potter fan fiction regardless of prompt. Other than copyright there are privacy considerations too. For example Gmail's Smart Compose is trained on users' private messages so you don't want it to memorize "private" details (such as credit card numbers): https://arxiv.org/abs/1802.08232 https://arxiv.org/abs/1802.08232 Is it possible to solve this by adversarially checking if the output is "original" enough or not? Or is that intractable, given how much resource our society already pours into making the same classification in court?
- tedivm 7y ago
  This point is extremely important, particularly in the healthcare field (which I happen to be in at the moment). We have to be very, very positive that our deidentification process is thorough and accurate to prevent a HIPAA violation from occurring.
  - jacquesm 7y ago
    If you start from real data that's a very tough nut to crack. Depending on your purpose you may be better off with generating data that has the same statistical properties.
    - nradov 7y ago
      That's true, but in healthcare generating data that matches the statistical properties of real data is also a very tough nut to crack. Essentially you need experienced clinicians who can document a fake patient's care journey through multiple years with thousands of concept codes and narrative text blocks. The level of effort is on the same order of magnitude as writing a novel. Of course if you don't need a complete longitudinal patient chart then generating realistic data for just one aspect can be a lot simpler.
jacquesm 7y ago
Towardsdatascience.com is rapidly rising on my irritation meter. Tons of submissions to HN of questionable value. This article is worse than most, the person interpreting the ruling does not appear to have a legal background and has essentially twisted the ruling to support his foregone conclusion. I'd love for an actual lawyer (paging Rayiner) to interpret this and to see whether or not any such far reaching conclusions are supported by the ruling the article is about. I've read the ruling and I've come to the conclusion that it makes no such statement but since I'm also not a lawyer you should assign as much value to that opinion as to the article itself.
- mchen076 7y ago
  Medium hosted articles in general tend to have both low quality and value. It's a very shiny platform which hosts a massive amount of really poorly reasoned articles.
qwerty456127 7y ago
Is using pirated or otherwise illegally acquired data to train an algorithm legal? If yes then why is it illegal to use it for other purposes? Is it legal to use the data to train an algorithm if the license disallows that explicitly?
- jacquesm 7y ago
  The end decider of that question will be another judge, but what I've been told by counsel retained for that express purpose so far was that the model (you don't really train an algorithm) that resulted from the use of that data would count as an automatic transformation of the data leaving it very likely that the model itself would be classed as a derivative work without deserving copyright protection in its own right. In other words: you could not have come up with the model without the original data.
- TallGuyShort 7y ago
  Does copyright law actually make it illegal (as opposed to just frowned-upon, and maybe hindered by one's ISP) to receive pirated material? I was under the impression that the illegal part was the distribution of it.
  - dragonwriter 7y ago
    > Does copyright law actually make it illegal (as opposed to just frowned-upon, and maybe hindered by one's ISP) to receive pirated material? Receiving, strictly, no. However with digital material, most use involves copying; for legitimate copies that copying is covered by an implied license for the normal use of the work, for copies which are not themselves authorized, there is likewise no implied license for use. Also, receipt of digital copies itself often involves copying directed by the receiver, which is prohibited, and may even involve a request from the receiver to the originator to make the copy under circumstances which would be viewed as knowing that it was unauthorized, which, may often, as a solicitation of an unlawful act, itself be illegal.
watzon 7y ago
I mean, why would it be? We train ourselves on copyrighted material all the time.
therealmarv 7y ago
Somebody know the situation in EU for that?
karussell 7y ago
Could this be used against Google? E.g. train an algorithm from their road traffic information (fetched legally via their API or web UI) to improve rough time estimates based on OpenStreetMap data.
- dragonwriter 7y ago
  > Could this be used against Google? E.g. train an algorithm from their road traffic information (fetched legally via their API or web UI) to improve rough time estimates based on OpenStreetMap data. No, because then contractual ToS, not naked copyright law, will be at issue. Even if Google doesn't have the right terms to forestall this now, it's a trivial change for them to adopt.
anonytrary 7y ago
We should be surprised that nine non-technical people are making technical decisions that impact 300m people. It seems that legal decisions are increasingly scope-creeping into domains where domain experts are necessary. This is even more evident after seeing the Zuck-Congress hearings, where Congress proved to the people that they aren't really the best people to work on technological issues.
sgjohnson 7y ago
Misleading title. SCOTUS refused to hear the case. They didn’t rule in Google’s favour.
ALBERT300 7y ago
HIRE A PROFESSIONAL HACKER FOR ANY HACKING RELATED ISSUES: OUR SERVICES INCLUDE: CELLPHONE MONITORING SOCIAL MEDIA HACKING DATABASE ALTERATION SPYING ON ALL SORT OF DEVICES CLONING ( WEBSITES, DEVICES, CELLULAR etc. ) HACKING SPOUSE PHONE UNIVERSITIES GRADE CHANGE RECOVER LOST CRYPTO CURRENCIES GET IN TOUCH WITH THE BEST HACKER VIA: EMAIL:THEREDHACKERGROUP AT GMAIL DOTCOM PHONE/WHATSAPP: 5 7 1 3 1 8 9 4 9 8

327 points by alok-g 7y ago

From a 2016 refusal to review a case

This isn't nearly as broad a precedent as the title sounds. They used some pretty reasonable tests of copyright infringement to conclude that no such infringement occured.

breck 7y ago

Anyone know of any organizations working to repeal Intellectual Monopoly laws? I know there are organizations that try to counter the influence of the IM industry like EFF and FSF, but I’m looking for groups that have come out and said Intellectual Monopoly laws need to go, period.

bitwize 7y ago
Pirate Parties worldwide. Sometimes they even get seats in parliament.

herniatedeel 7y ago

SCOTUS denied the petition for writ of certiorari, thereby leaving the 2nd Circuit's ruling in Google's favor intact. However, the 2nd Circuit's ruling is not binding on any other federal circuits. Also, as Enginerrd stated, the holding is not nearly as broad as the article makes it out to be. The holding was: 1. Google’s unauthorized digitizing of copyright-protected works, creation of a search functionality, and display of snippets from those works are non-infringing fair uses. The purpose of the copying is highly transformative, the public display of text is limited, and the revelations do not provide a significant market substitute for the protected aspects of the originals. Google’s commercial nature and profit motivation do not justify denial of fair use. 2. Google’s provision of digitized copies to the libraries that supplied the books, on the understanding that the libraries will use the copies in a manner consistent with the copyright law, also does not constitute infringement. Based on the above holding, I think the article's conclusion is a stretch for general training algorithms using copyrighted data because: (1) there would not be a library supplying the information to the training algorithm, (2) there would be no similar display of snippets, and (3) we do not know if a training algorithm would provide a market substitute for the copyrighted data.

bluGill 7y ago
While the decision is only binding in the 2nd circuit, the precedent is admissible in other courts. If this goes to trial in a different circuit you can bring the finding to the judge who will consider it - it won't be binding, but he will consider it. If it goes to appeal in a different circuit the next circuit will reference this in their decision - if they decide the 2nd is wrong they will be very clear why the they think the 2nd circuit is wrong when they make ruling (and this can in turn be re-submitted to the 2nd circuit who might change their mind if the reasoning is good enough). If this goes to the supreme court in the future they will read this decision and it will influence them - again they can decide either way.
- herniatedeel 7y ago
  Yes, the 2nd Circuit's decision is persuasive authority for other circuits. However, that's not what the article claims. The article claims that SCOTUS ruled when it, in fact, did not.
  - dragonwriter 7y ago
    SCOTUS ruled on the cert petition; what people may not understand is that while that is a ruling, it is (and there is explicit precedent on this point) not one which has precedential weight (even as persuasive authority) as regards the merits of the issues addressed in the lower court ruling.
    - daveFNbuck 7y ago
      It's not hard to see why people are not understanding this correctly. The HN link text: > Training algorithms on copyrighted data not illegal: US Supreme Court The sub-heading for the article: > Training algorithms on copyrighted data is not illegal, according to the United States Supreme Court.
    - angry_octet 7y ago
      Isn't it unlikely that a case will be granted cert if appeals courts in different circuits are in agreement? I.e. not a circuit split? So while not legally binding, it might be in practice indicative. I wonder, does this mean I can scrape Instagram/Facebook for photos and use them for face recognition? Is that 'fair use'? Is an Instagram post a publication?

prepend 7y ago

If this legal theory is correct and upheld, possession of data will be pretty important. So all those data you licensed for one reason or another might be able to be used to train. I think the logic makes sense because imagine if humans were prevented from getting ideas from watching movies. It seems similar to not letting AI watch every movie ever and learn.

jacquesm 7y ago
I don't think the judgment supports the conclusion in the article.
tzs 7y ago
It sounds like to get the data into their AI, Google had to make copies. They digitized the physical books, and then trained using those copies they made. Also, their system included excerpts from the books which it could retrieve and show users. Hence, they had to use fair use to justify it. I think if you could train the AI without having to make a copy first, such as having the AI read the physical books directly, or in the case of your movie example having the AI watch the movie on a TV hooked up to a DVD player playing a copy of the movie on DVD that you bought from a retailer authorized by the copyright owner to sell such DVDs, you might not even need to make a fair use argument. The definitions section of the US copyright statutes, 17 USC 101 [1], defines "copies" like this: > “Copies” are material objects, other than phonorecords, in which a work is fixed by any method now known or later developed, and from which the work can be perceived, reproduced, or otherwise communicated, either directly or with the aid of a machine or device. The term “copies” includes the material object, other than a phonorecord, in which the work is first fixed. and "fixed" is defined like this: > A work is “fixed” in a tangible medium of expression when its embodiment in a copy or phonorecord, by or under the authority of the author, is sufficiently permanent or stable to permit it to be perceived, reproduced, or otherwise communicated for a period of more than transitory duration. An AI reading or watching the work as one of many many works in order to learn weights for a neural net does not result in a material object from which the work can be perceived, reproduced, or otherwise communicated. Thus, there is no copy, and hence no copyright issue. [1] https://www.law.cornell.edu/uscode/text/17/101 https://www.law.cornell.edu/uscode/text/17/101

fyp 7y ago

We probably need a technical rather than legal solution for this. The problem is that generative algorithms are susceptible to accidental memorization so you can't guarantee that the output will be transformative. For example, play with https://talktotransformer.com https://talktotransformer.com to see how many well known pieces of text it can spew out verbatim. It is very prone to derailing into a harry potter fan fiction regardless of prompt. Other than copyright there are privacy considerations too. For example Gmail's Smart Compose is trained on users' private messages so you don't want it to memorize "private" details (such as credit card numbers): https://arxiv.org/abs/1802.08232 https://arxiv.org/abs/1802.08232 Is it possible to solve this by adversarially checking if the output is "original" enough or not? Or is that intractable, given how much resource our society already pours into making the same classification in court?

tedivm 7y ago
This point is extremely important, particularly in the healthcare field (which I happen to be in at the moment). We have to be very, very positive that our deidentification process is thorough and accurate to prevent a HIPAA violation from occurring.
- jacquesm 7y ago
  If you start from real data that's a very tough nut to crack. Depending on your purpose you may be better off with generating data that has the same statistical properties.
  - nradov 7y ago
    That's true, but in healthcare generating data that matches the statistical properties of real data is also a very tough nut to crack. Essentially you need experienced clinicians who can document a fake patient's care journey through multiple years with thousands of concept codes and narrative text blocks. The level of effort is on the same order of magnitude as writing a novel. Of course if you don't need a complete longitudinal patient chart then generating realistic data for just one aspect can be a lot simpler.

jacquesm 7y ago

Towardsdatascience.com is rapidly rising on my irritation meter. Tons of submissions to HN of questionable value. This article is worse than most, the person interpreting the ruling does not appear to have a legal background and has essentially twisted the ruling to support his foregone conclusion. I'd love for an actual lawyer (paging Rayiner) to interpret this and to see whether or not any such far reaching conclusions are supported by the ruling the article is about. I've read the ruling and I've come to the conclusion that it makes no such statement but since I'm also not a lawyer you should assign as much value to that opinion as to the article itself.

mchen076 7y ago
Medium hosted articles in general tend to have both low quality and value. It's a very shiny platform which hosts a massive amount of really poorly reasoned articles.

qwerty456127 7y ago

Is using pirated or otherwise illegally acquired data to train an algorithm legal? If yes then why is it illegal to use it for other purposes? Is it legal to use the data to train an algorithm if the license disallows that explicitly?

jacquesm 7y ago
The end decider of that question will be another judge, but what I've been told by counsel retained for that express purpose so far was that the model (you don't really train an algorithm) that resulted from the use of that data would count as an automatic transformation of the data leaving it very likely that the model itself would be classed as a derivative work without deserving copyright protection in its own right. In other words: you could not have come up with the model without the original data.
TallGuyShort 7y ago
Does copyright law actually make it illegal (as opposed to just frowned-upon, and maybe hindered by one's ISP) to receive pirated material? I was under the impression that the illegal part was the distribution of it.
- dragonwriter 7y ago
  > Does copyright law actually make it illegal (as opposed to just frowned-upon, and maybe hindered by one's ISP) to receive pirated material? Receiving, strictly, no. However with digital material, most use involves copying; for legitimate copies that copying is covered by an implied license for the normal use of the work, for copies which are not themselves authorized, there is likewise no implied license for use. Also, receipt of digital copies itself often involves copying directed by the receiver, which is prohibited, and may even involve a request from the receiver to the originator to make the copy under circumstances which would be viewed as knowing that it was unauthorized, which, may often, as a solicitation of an unlawful act, itself be illegal.

watzon 7y ago

I mean, why would it be? We train ourselves on copyrighted material all the time.

therealmarv 7y ago

Somebody know the situation in EU for that?

karussell 7y ago

Could this be used against Google? E.g. train an algorithm from their road traffic information (fetched legally via their API or web UI) to improve rough time estimates based on OpenStreetMap data.

dragonwriter 7y ago
> Could this be used against Google? E.g. train an algorithm from their road traffic information (fetched legally via their API or web UI) to improve rough time estimates based on OpenStreetMap data. No, because then contractual ToS, not naked copyright law, will be at issue. Even if Google doesn't have the right terms to forestall this now, it's a trivial change for them to adopt.

anonytrary 7y ago

We should be surprised that nine non-technical people are making technical decisions that impact 300m people. It seems that legal decisions are increasingly scope-creeping into domains where domain experts are necessary. This is even more evident after seeing the Zuck-Congress hearings, where Congress proved to the people that they aren't really the best people to work on technological issues.

sgjohnson 7y ago

Misleading title. SCOTUS refused to hear the case. They didn’t rule in Google’s favour.

ALBERT300 7y ago

HIRE A PROFESSIONAL HACKER FOR ANY HACKING RELATED ISSUES: OUR SERVICES INCLUDE: CELLPHONE MONITORING SOCIAL MEDIA HACKING DATABASE ALTERATION SPYING ON ALL SORT OF DEVICES CLONING ( WEBSITES, DEVICES, CELLULAR etc. ) HACKING SPOUSE PHONE UNIVERSITIES GRADE CHANGE RECOVER LOST CRYPTO CURRENCIES GET IN TOUCH WITH THE BEST HACKER VIA: EMAIL:THEREDHACKERGROUP AT GMAIL DOTCOM PHONE/WHATSAPP: 5 7 1 3 1 8 9 4 9 8