Search Unity

  1. We are migrating the Unity Forums to Unity Discussions. On July 12, the Unity Forums will become read-only.

    Please, do not make any changes to your username or email addresses at id.unity.com during this transition time.

    It's still possible to reply to existing private message conversations during the migration, but any new replies you post will be missing after the main migration is complete. We'll do our best to migrate these messages in a follow-up step.

    On July 15, Unity Discussions will become read-only until July 18, when the new design and the migrated forum contents will go live.


    Read our full announcement for more information and let us know if you have any questions.

Official Updating Unity’s guiding principles for ethical AI

Discussion in 'Announcements' started by TreyK-47, Jun 13, 2023.

  1. TreyK-47

    TreyK-47

    Unity Technologies

    Joined:
    Oct 22, 2019
    Posts:
    1,843
    Howdy, everyone!

    As AI technology continues to play a significant role in our society, we understand the importance of maintaining transparency, fairness, and accountability in its development and deployment. We are dedicated to adhering to government regulations worldwide and ensuring that our AI solutions are aligned with ethical standards.

    In our pursuit of responsible AI usage, we have recently revised our guiding principles for ethical AI. We extend an invitation to our community of creators to join us in promoting the responsible and safe utilization of AI within the Unity community. Together, we can foster an environment where AI is used productively while upholding fairness and integrity.

    You can learn more in our blog post. BUT we want to hear from you! Please share your thoughts, concerns, suggestions, and more with us below!
     
    Last edited: Jun 13, 2023
  2. makaka-org

    makaka-org

    Joined:
    Dec 1, 2013
    Posts:
    1,093
    Please expand the topic. What is the potential harm?
     
  3. AcidArrow

    AcidArrow

    Joined:
    May 20, 2010
    Posts:
    12,015
    Is this a roundabout way to tell us you won't be disclosing your sources and an attempt to frame it as some sort of positive or ethical thing?
     
  4. JiRo_Dev

    JiRo_Dev

    Joined:
    Jun 8, 2018
    Posts:
    53
    You're focusing on something irrelevant, they said "unauthorized disclosure", so no point in opposing that as the opposite would mean no data given to Unity to make the tools that would eventually trickle down to us, but I get that hiding the source could give them a way to have whatever source they want, so what I think Unity should do is disclose the category of the sources and not the sources individually, that'll protect the sources while assuring developers that the data is not taken from say a black market although highly unlikely as the data is probably from games and videos...
     
  5. AcidArrow

    AcidArrow

    Joined:
    May 20, 2010
    Posts:
    12,015
    Am I? They are painting disclosure as a bad thing, therefore when they don't disclose anything it's somehow a good thing.

    If they had gone the extra mile of making sure they use data where there are no ethical issues and no potential litigation to be brought against them, they would have been touting their own horn so much.

    But they aren't. Prove me wrong Unity.
    What's the point of that?
     
    TheOnePoly, LazloBonin and Ryiah like this.
  6. Ryiah

    Ryiah

    Joined:
    Oct 11, 2012
    Posts:
    21,937
    What do you mean by "category"? Because I can't think of any reason that categorizing them would prevent them from coming from disreputable sources.
     
  7. superpig

    superpig

    Drink more water! Unity Technologies

    Joined:
    Jan 16, 2011
    Posts:
    4,670
    Hey folks!

    No. What we're saying there is: we will be treating all the data we use for training AI - which may include, for example, data that you have chosen to share with us - with appropriate care. We will keep it secure, we will not use it for things you didn't agree to when you shared it with us, etc.

    I agree that we could be more explicit about our approach to sources more generally. I discussed it with Legal and here is the position we plan to incorporate more clearly into the next iteration of the AI Guiding Principles:

    We will uphold the rights of creators in any datasets that we construct ourselves, and will continually seek to work with partners who also share our belief in the importance of respecting the rights of creators.
     
    HyperBrid, DragonCoder and AcidArrow like this.
  8. Unifikation

    Unifikation

    Joined:
    Jan 4, 2023
    Posts:
    1,122
    Very interesting. Have you hired anyone with a demonstrated understanding of ethics (and history), or is this some sort of posturing?

    The blog post suggests the latter, strongly.
     
    TheOnePoly likes this.
  9. AcidArrow

    AcidArrow

    Joined:
    May 20, 2010
    Posts:
    12,015
    Fair enough, thanks.
     
  10. Baste

    Baste

    Joined:
    Jan 24, 2013
    Posts:
    6,371
    That blog post is probably the most corporate nothing statement I have ever read in my life. I get it, you
    , so you can't actually say anything that's binding in any way, but holy hell then maybe don't say anything at all?

    There's a lot of ethical concerns about AI, that blog post mentions none of them directly. There's a lot of concrete things you could do, that blog post mentions none of them directly. So it spends a couple of hundred words saying nothing. When BP writes a post like that, I assume that there's just been an oil leak somewhere, so this mostly makes me wonder what bad thing you have done, rather than it instilling any kind of confidence.



    If you have any courage left as a company, and are allowed to speak without double-checking with legal every third word, questions you could answer are:
    - What's your stance on using data sets where artists might not have opted in to the use of their art?
    (The blog post says "appropriate data sets" and "open and healthy expression", but that doesn't mean anything as appropriate is completely arbitrary)

    - Will you let your users know if you start using (or are using) AI services to replace actual staff? Things like bug classification and customer service or whatnot.

    - The asset store will probably (if it hasn't already) be drowned in AI-generated texture packs etc. What's your stance on that? Do those assets have to be marked as AI-generated?

    - If you sell us any AI services, do you make any kinds of guarantees about the copyright status of those services? As in, if say Unity creates some AI asset production tools and sells those, will you guarantee that the training set isn't possibly tainted in a way that a future court case might decide that all assets made by those tools are copyrighted by a third party?
     
  11. r9shackleford

    r9shackleford

    Joined:
    Jul 23, 2019
    Posts:
    23
    You guys send AI-generated email spam to your users...
     
    drcfrx, AcidArrow and Unifikation like this.
  12. LazloBonin

    LazloBonin

    Joined:
    Mar 6, 2015
    Posts:
    833
    @Baste is asking all the right questions. Kudos.

    As creators, it's our collective responsibility to demand a clear answer out of Unity about the source of the datasets before even considering using any of these AI tools.

    For example, Adobe has been using user uploaded content from Adobe Stock to train Adobe Firefly without explicit creator consent and with no ability to opt-out and with no royalties paid on images trained on their source material. Rightfully, this led to massive controversy. So what in the FAQ right now is guaranteeing that Unity Muse isn't intending to do the same?

    And to be clear, this is not only about protecting the content made by Unity creators: existing popular datasets that were scraped from the internet at large without explicit consent or usage rights are profoundly unethical, if not potentially downright illegal (laws haven't yet adapted to AI, but regulations are evolving quickly). Basic solidarity with artists from all fields should be a given for digital creators like gamedevs.

    Not to mention both gamers and critics are already frowning heavily (and rightfully) upon games that are using unethical AI datasets — remember High on Life? Myst?

    Unity has a responsibility to clearly define whatever "appropriate datasets" is supposed to mean.

    "Appropriate" is meaningless legal/comms mumbo-jumbo. It does not mean ethical.

    If Unity reps are ready to commit to it meaning "legally obtained from compensated creators who gave explicit consent", then by all means, Muse and Sentis are tremendously exciting news.

    But if it means "technically legal so far, at least none of the other big corps stealing content got sued yet"... then these tools are profoundly disrespectful of creators of all disciplines, and run an incredibly high risk for their users (us!) of copyright infringement pollution in our games as the law adapts over the next few years.
     
    Last edited: Jun 27, 2023
    NotaNaN, AcidArrow, Stealcase and 2 others like this.
  13. YondernautsGames

    YondernautsGames

    Joined:
    Nov 24, 2014
    Posts:
    356
    The asset store is the only real candidate for where they could get that training data that I can think of. I can imagine there's something from a longstanding Weta R&D project or something like that, but the asset store seems far more likely when they say things like "data that you have chosen to share with us"

    Edit: Looking through the asset store provider agreement, I can't see anything that would constitute permission for this use, so I guess not?
     
    Last edited: Jun 27, 2023
  14. Unifikation

    Unifikation

    Joined:
    Jan 4, 2023
    Posts:
    1,122
    The words used to describe the AI Muse feature are sufficiently nebulous and seemingly deliberately vague that I'd guess they're looking at user code and project structures.
     
  15. superpig

    superpig

    Drink more water! Unity Technologies

    Joined:
    Jan 16, 2011
    Posts:
    4,670
    Yes, we require publishers of AI-generated assets to disclose that they're AI-generated assets. When a publisher is listing their asset there is a checkbox for "I have used AI/ML in my package creation process" and they provide an explanation, and this is then shown in a 'Created with AI' section on the Asset listing - you can see an example here.
     
  16. Stealcase

    Stealcase

    Joined:
    Feb 26, 2018
    Posts:
    11
    In this "Correction" of Unity's stated values, there's a nasty implication.
    1. That your current AI Sollutions you've implemented in Unity do not adhere to these stated values. This post was made 2 weeks ago, before the announcement of Sentis and Muse. The updated AI Guiding Principles were conveniently NOT updated before this announcement.
    2. "Will continually seek to work with" implies that you do not require lawful or ethical datasets. You'll just give it the good ol' college try.
    3. The fact that you're checking with lawyers to give an answer on this... is telling.

    Something can be legal and also completely unethical.

    But I have my sincere doubts on the legality of any of this. Navigating Copyright laws across the world is tedious in the best of times, and you (Unity) just decided to jump headfirst into the dubious realm of content generation trained on assets you literally have no license to use. Fair Use MIGHT apply in the US, but it literally doesn't exist in Europe.
     
    AcidArrow and Unifikation like this.
  17. Coffeeseed

    Coffeeseed

    Joined:
    Jan 18, 2017
    Posts:
    6
    If you are serious about the ethics and accountability then transparency must be achieved as well or all this blog post is just PR posturing.

    You're mentioning the respect of creators rights then prove it and disclose your models and training datasets.

    No current genAI is ethical at all and without unity's transparency we can assume that your is not ethical as well.

    Instead of vague and corporate words, tell us the actual action you're taking to ensure that no copyrighted data was included in the datasets you've used (which I doubt because some of it awfully look mile stable diffusion which contain hundreds of millions of copyrighted data as well as CP) 1nd explain us exactly how you're going to protect your user's data from being used in the future.

    Also, what is the point of having genAI "tools" in a video game production pipeline when you can not protect ML output and most likely infringe other's rights? What studio will build an IP using assets that are worthless and fall into creative commons the moment they are generated? You will give a lot of studio legal departement a huge headache as they'll have to be sure that their employee/contractor do NOT use it to not create trouble for the brand and people.

    Lastly, using accessibility as a PR move is really a cheap shot. The whole blog post basically use a lot of words to say nothing, peak corporate.

    I'm not pleased with Unity's decision to jump on the latest unethical (and likely illégale) tech fad, you're not making your product better, you're hurting the creators.
     
    Unifikation likes this.
  18. Andy-Touch

    Andy-Touch

    A Moon Shaped Bool Unity Legend

    Joined:
    May 5, 2014
    Posts:
    1,515
    Curious about where Atlas falls into these guidelines as a Verified Solution:
    https://twitter.com/TheMirzaBeig/status/1673814584698843145
     
    Unifikation likes this.
  19. Taro_FFG

    Taro_FFG

    Joined:
    Jun 24, 2022
    Posts:
    57
    The fact that this Atlas "3D Asset Creator" is not actually creating 3D assets but downloads them and is being promoted with the taglines:
    "
    Build confidently with Verified Solutions
    Find professional solutions that have undergone enhanced vetting from creators who are committed to providing high-quality solutions, service, and long term support.
    "

    At best this asset is "just" missleading customers.
    This shouldn't pass the sniff test especially during a high profile marketing push for only 10 assets.

    Imagine finding out post launch that your "AI" generated main character is actually from some asset store and it turns out the uploader actually did not create the asset themselves but stole it from somebody else.

    I'm all for a well done AI toolchain, but this is a problem.
     
    Unifikation likes this.
  20. YondernautsGames

    YondernautsGames

    Joined:
    Nov 24, 2014
    Posts:
    356
    A number of them are problematic tbh.

    Leonardo.ai claims you can train it with your own data. If that feature is exposed in the editor then you can guarantee that the majority of that data will be sourced from places like the asset store. Sure there's a EULA for unity users that says you can't do that, but how will that be enforced? There's a huge imbalance in terms of effort to police (all on the artists) damage done (artists work is essentially used against their will and rights, end users are unknowingly using stolen content, occasionally a user ignoring the EULA might get caught and get a slap on the wrist) and benefit (mostly on Leonado.ai). Meanwhile in their license:
    Layer.ai has the following in its ToS:
    Where "Your Content" is anything you upload to them. "Game assets by Layer are in your art style. Just upload a few reference images and hit run."

    LMNT has the following:
    Where, again, "User Content" is anything you upload. They offer the ability to "Clone a Voice" where users can upload any recording. Again, there's a ToS section to say you own the rights, but of course people will abuse it and everyone else will suffer the consequences.

    All of these come across as companies that have no interest in the legal or ethical ramifications of their work looking to crowd source data sets via unity while offloading the risk and responsibility to the end users.

    Some others of the assets make it clear that they create their own source data internally, don't train on user submitted content and don't retain the rights to user submitted content. That should be the rule. Not the exception, and not just the expectation.
     
  21. leoaa69

    leoaa69

    Joined:
    Jul 19, 2023
    Posts:
    1
    Unity Technologies had not officially released specific guiding principles for ethical AI. However, given the evolving landscape of artificial intelligence and ethical concerns surrounding its use, it's essential for any technology company, including Unity, to consider developing ethical guidelines for AI.

    If Unity Technologies decides to update or establish guiding principles for ethical AI, here are some key areas they may want to address:

    1. Transparency and Explainability: Ensure that AI systems developed using Unity's technology are transparent and provide explanations for their decisions. Users should have insights into how the AI makes decisions and understand the factors that influence its behavior.

    2. Fairness and Bias Mitigation: Strive to create AI systems that are fair and unbiased. Unity should take measures to identify and mitigate potential biases in data and algorithms that may lead to discriminatory outcomes.

    3. Data Privacy and Security: Respect user privacy by implementing robust data protection measures. Unity should handle data responsibly, ensuring it is securely stored and used only for its intended purposes.

    4. Accountability and Responsibility: Hold developers and users accountable for their AI applications. Encourage responsible AI development and use, discouraging harmful or malicious practices.

    5. Human-Centric Design: Prioritize the well-being of humans in AI development. Unity should aim to enhance human capabilities, promote inclusivity, and consider the social impact of AI applications.

    6. Collaboration and Openness: Foster collaboration within the AI community and promote open research and discussions about the ethical implications of AI.

    7. Continuous Review and Improvement: Recognize that AI ethics is an ongoing process. Unity should regularly review and update its ethical guidelines to align with evolving ethical standards and technological advancements.

    8. Compliance with Regulations: Ensure that AI applications developed using Unity's technology comply with relevant laws and regulations related to AI, data protection, and privacy.
    To develop these guiding principles, Unity may involve AI ethics experts, researchers, developers, and stakeholders from diverse backgrounds. Transparent communication with their user community and other interested parties is also crucial during the process.
     
    Unifikation likes this.
  22. LazloBonin

    LazloBonin

    Joined:
    Mar 6, 2015
    Posts:
    833
    lmao this reply was clearly generated by chatGPT, come on

    and it doesn't address the core issue: none of these points care about ethical, compensated sourcing for the trained datasets
     
    Unifikation likes this.