In the past few months, there have been a surge of AI projects that allow generating images and text:
- Stable Diffusion
- GPT-3 and earlier
- DALL-E 2 and previous
- GitHub Copilot
These AI programs are amazing – but they were also trained with publicly-available material; and the owners of that material almost certainly did not opt-in to having their material used for AI training, and have occasionally managed to get these services to repeat things very close to copyrighted material. This is almost certainly going to come up in future legal cases.
Putting the current US concept of fair use aside, I think that at this point, AI companies have a vested interest in doing everything they can to get these algorithms entrenched as an industry, because that may actually ensure their legal survival.
Consider a broader view of the US and technology:
- VCRs upset movie studios tremendously, but were declared legal even if some people would abuse them or copy tapes. Format-shifting was declared officially legal from this decision, whereas before it had been legally grey, much like AI is now. However, there’s another side to the story: In 1983, according to the New York Times, there were approximately 1.2 million VCRs sold that year, with the decision decided in January 1984. (Basically) outlaw the industry? Nah.
- Photoshop came out, and allowed for the manipulation of images in ways that were unprecedented. Users could also abuse Photoshop to make very… interesting… images of celebrities. Nonetheless, Photoshop was never sued for being liable for anything their users did.
- CD Drives allowed copying CDs which did not have DRM, and made it easy to share the ripped discs online. This did not ultimately make CD drives, CD ripping, Online File Sharing, BitTorrent, The Internet, or any of the technologies involved illegal despite all of them being abused for copyright infringement. It also didn’t legalize internet censorship of DNS and packets to prevent copyright infringement despite the MPAA’s lawsuits and failed laws (SOPA/PIPA).
If there seems to be a pattern, I would quantify it as this:
US Courts do not enjoy clamping down on any new technology, even if said new technology can and is being used in copyright-infringing ways.
Now, one could make the argument, that none of these really have much to do with AI, or AI’s propensity to regurgitate information it learned with sometimes. I think, however, that this is a “hindsight is 20/20” moment. It’s obvious now, but it wasn’t obvious then. If we had CD ripping declared illegal, or lost VCR recording being legal, or had SOPA/PIPA enforced, our precedent for new technologies and copyright infringement would be very, very different.
Thus, in a weird way, it would seem to my unlawyered thoughts that the more AI can entrench itself, become accepted, widespread, diverse in function, the stronger the legal case will become. If it was just GitHub Copilot, it may be banned. But will courts be interested in hurting Copilot, Midjourney, DALL-E, GPT-3, etc.? I think they would punt the question to Congress before they would dare make a change to the status quo or declare that it isn’t “fair use,” if previous technology/copyright conflicts are anything to go by.