Since I haven't found any good training data for my university project, I want to use pictures and videos from public Instagram profiles. Am I allowed to do that?
$\begingroup$
$\endgroup$
2
-
3$\begingroup$ I'm answering this here, then posting my answer on Law for review. This is a common question, and critical for AI researchers. $\endgroup$– DukeZhouSep 22, 2021 at 23:42
-
$\begingroup$ Informal precedent, as it seems to have been scrubbed from the internet: Old timers will remember when Mark Zuckerberg tried to assert copyright over all content published on facebook. The backlash was so great, he quickly retreated, tail between his slimy legs. Many legal questions are "untested". My understanding is that the IP rights for post reside with the creators of those posts. They'd have no way of knowing you're using their content, and, in any case, put that content in the public domain. $\endgroup$– DukeZhouSep 30, 2021 at 2:22
Add a comment
|
3 Answers
$\begingroup$
$\endgroup$
1
Under US copyright law, this is probably fair use
...but beware of memorization. You may run into more trouble if the AI outputs things very similar to the original work.
Also, consult a lawyer to help you apply the law to your specific situation. This is just information on general legal principles, not any specific situation, and also I'm not a lawyer.
First of all, the vast majority of images on Instagram are almost certainly subject to copyright protection, as any creative work is automatically subject to such protection (with exceptions for things like government works). You'd either need a license to use it, or an exception such as fair use.
Fair use is a four-factor balancing test
It requires consideration of four factors by a judge:
- the purpose and character of your use
- the nature of the copyrighted work
- the amount and substantiality of the portion taken
- the effect of the use upon the potential market.
Let's go through them for this.
The purpose and character of your use
This is probably best known as whether the work is transformative. Exactly how transformative it is depends on exact what type of AI you're training (an AI to do object recognition is probably more transformative than an AI that outputs Instagram-esque photos, for instance), but I'd expect that this would generally lean in favor of the AI trainer.
The nature of the copyrighted work
It's a creative work, not a factual work. Probably leans against the AI trainer. It would lean further against the AI trainer if the works were unpublished rather than publicly viewable on Instagram.
The amount and substantiality of the portion taken
This one's complicated. In some sense, you've used the whole work to train the AI. In another sense, you've only taken some general information about what's in the work. This may also be affected by how much is reproduced from the work by the AI. Probably leans in favor of the AI trainer.
The effect of the use upon the potential market
If you're using the AI to copy someone's artistic style and compete with them in that manner, that might lean against you. Otherwise, it seems unlikely that training an AI with Instagram data would affect the market for the original images. Probably leans in favor of the AI trainer.
Overall, these four factors lean in favor of using such data for training AI/ML algorithms.
Postscript: there is also a relevant legal case, but as it involves my employer, I won't be commenting on it beyond providing that link.
-
$\begingroup$ Relevant legal case is "Authors Guild v. Google". I don't see that it comments directly on the question b/c this case involves potential financial harm (a market.) The amount I would worry about training an algorithm on data scraped from the internet is exactly 0. $\endgroup$– DukeZhouSep 30, 2021 at 2:26
$\begingroup$
$\endgroup$
2
Disclaimer that every attorney will give unless formally engaged: This does not constitute formal legal advice.
- This data was published with the expectation of public view
Viewing that data is a form of utilization—taking in information which may be used to make a decision, or just as divertment (here consumption of entertainment media by humans.)
If any observer can input that data, you can surely use that data for analysis by your learning algorithm.
- It's research
- It's not for profit
- There is no infringement
Selling that dataset might constitute infringement, in that copyright in inherent to natural persons when they create novel content in any memory medium (paper, film, clay tablets, high-level representations of machine instructions, etc.)¹
Re-publishing that data without permission is definitely infringement, except in cases of fair use.
https://en.wikipedia.org/wiki/Fair_use#File_sharing
We in the AI community are on much firmer ground than file sharers, documentarians, internet publishers, and other professional communities because all we are doing is accessing a public dataset for the purpose of scientific research.
The internet itself constitutes a database,² freely available except in the case of firewalls.
My feeling is that these public databases (instagram, et al) are meant to be processed, analyzed, and utilized by any networked users, just as a mathematical formula might.
The internet and all servers and nodes are part of a computational process. Compiling into a dataset is just an extension of that computational process. That data is reproduced and mirrored locally when it is put into packets and transmitted to any system requesting the data.
- This content is all in the public domain
They may not have a creative commons statement, but, my feeling is, in this case, so long as you don't re-publish for profit, you're not infringing—there is no financial harm, thus no damages³
Even if an Instagram user believes you are infringing, they must first engage an attorney to send you a cease-and-desist letter. So must Instagram.
In that likely instance, simply comply or challenge by refusing.
- Sociologists use datasets like this for research, and I haven't head of any sociologists being sued for studying Facebook!
The analysis of those datasets, even by sociologists, requires some form of automated analytics, but which we can mean intelligence.
[1] I'd have to research further to determine this b/c many are surely doing it or have tried. But not your question, so will leave unanswered.
[2] Web pages and websites are a form of database, typically with open read access to any user, human or algorithmic. The internet is a structured database of structured data elements (pages and sites).
[3] I'm taking some graduate IT courses now and my sense is the legal advice schools are being given is to require citation even where citation is not necessary because it's fuzzy. It creates a lot of unnecessary citations of rewordings of common knowledge, but they are choosing strict minimax (paranoia). But this may be b/c non-traditional programs are teaching students to google then reword summaries to avoid detection by current gen plagiarism checkers, and thereby successfully simulate research, even where all the content is generic.
The NFL says specifically "not reproduce without express consent of the NFL", but that's broadcast/satellite/cable, which is robustly regulated.
Damages are very hard to prove, and it's a very expensive proposition, with no guarantee of success. It's a losing proposition for Instagram no matter how you slice it. (These are the mechanics of law.)
-
3$\begingroup$ "so long as you don't re-publish for profit, you're not infringing" is a very mistaken statement of at least US copyright law. It is quite possible to infringe without republishing for profit (see, e.g., people being sued for seeding torrents). $\endgroup$– Ryan MSep 23, 2021 at 0:14
-
$\begingroup$ @RyanM Please post! I can't see any legal jeopardy in this use for a dataset, but I'd like to improve my answer. (Plus, aren't we in data piracy territory in the area of torrents? What is the data being used for in those scenarios? Does it create harm by reducing potential revenues that could have been generated by that content? I can't see any financial damage arising from use as a machine learning dataset to any party.) $\endgroup$– DukeZhouSep 23, 2021 at 0:16
$\begingroup$
$\endgroup$
I'd argue that the legality depends strongly on how you access the data, and that automating this may put you in breach of facebook/instagram's terms of use for their data.
It's not that accessing a set of images wouldn't count as fair use, it's that automatically scraping them probably breaches the terms of service.
The most likely consequence, though, is that something clever on instagram's servers blocks you, and, in fact, this does happen when you try things like this.
