nukacolaholic's recent activity

  1. Comment on Quizzle – Can you guess the word in fewer than twenty questions? in ~games

    nukacolaholic
    Link Parent
    For me, it was larger than a house cat, so it is really interesting the difference in the response there.

    For me, it was larger than a house cat, so it is really interesting the difference in the response there.

  2. Comment on Making or using generative ‘AI’ is, all else being equal, a dick move in ~tech

    nukacolaholic
    Link Parent
    I think what you are saying is persuasive at the output level. If an AI generated work goes through the same tests a human generated work would go through and fails, I think it should be...

    I think what you are saying is persuasive at the output level. If an AI generated work goes through the same tests a human generated work would go through and fails, I think it should be considered infringement. However, this led me to take a closer look at Sobel and I am really struggling to square his arguments about infringement during the training of AI with the Grimmelmann Copyright for Literate Robots article he cited, which seems to argue that machine reading and copies only for machine consumption cannot be infringement. Grimmelmann seems to argue that should not necessarily be the case, but that it is the state of the law.

    And ultimately, I tend to be skeptical of copyright law because it in particular seems to be captured by the greed of large corporations against the interest of the public (see Disney and the ever increasing length of time it takes a work to enter public domain). Fundamentally, I am suspicious that any attempt to modify copyright law specifically for the issue of AI would likely be weaponized against small creators and the public good.

    3 votes

  3. Comment on Making or using generative ‘AI’ is, all else being equal, a dick move in ~tech

    nukacolaholic
    Link Parent
    I think Sobel is ultimately unsuccessful at disambiguating AI training from humans interacting with copyrighted material in cases where there is not intermediate copying. Particularly in the...

    I think Sobel is ultimately unsuccessful at disambiguating AI training from humans interacting with copyrighted material in cases where there is not intermediate copying. Particularly in the section you quoted, how is that distinct from a human artist doing the same thing? I don't think a human artist doing a work in the style of another artist after hearing their music would be considered infringement.

    On speed and scale, I think my problem lies in that if the fundamental premise (i.e. work trained on the work of other artists is necessarily infringing) is incorrect, the speed and scale should be irrelevant in a legal framework. Once you say that is okay to draw the line somewhere, it creates a legal precedent that the fundamental premise has some validity, and people are going to start pushing at where that line should be drawn. If a human artist is inspired by the work of another artist and then starts putting out work in that same style but vastly outproducing the inspiring artist, does the inspiring artist have a claim against the one they inspired? I can definitely see big companies trying to push that boundary at the expense of smaller creators.

    In general, I don't know that we are super far off from each other. There probably needs to be some legal framework for the use of AI, but copyright law just seems like a really bad approach to me.

    Edited to add: At least, copyright law at the training level. At the output level, if it's putting out a note for note replication of a song, that's problematic, and preventing AI output from being copyrightable does seem like a beneficial approach.

    4 votes

  4. Comment on Making or using generative ‘AI’ is, all else being equal, a dick move in ~tech

    nukacolaholic
    Link Parent
    As much as I agree that AI should not be used to replace human created work, I really struggle with the copyright arguments against AI. It's very hard for me to distinguish between the arguments...

    As much as I agree that AI should not be used to replace human created work, I really struggle with the copyright arguments against AI. It's very hard for me to distinguish between the arguments against AI training and the process of creation by a human. Humans creating artistic works are going to be inspired and influenced by the works that they have seen/read/heard in their lifetimes. It seems nonsensical to claim that a human should be expected to obtain a special training license for every book they have read in their life in order to write. The difference with AI is scale and speed, but the underlying process doesn't seem different and that makes it problematic to me to try to use the same legal framework.

    I do hope that copyright law ends up settling on the side of AI works being non-copyrightable, because I think that will help ensure that humans don't end up entirely replaced, even if the tech continues to improve.

    17 votes