In most of the discussions about this fiasco, I have yet to see anyone point out the following:
- There is a finite number of distinct human voices.
- The # of desirable human voices (D) < The # of distinct human voices (A)
Where D := A - (All unwanted / undesirable human voices)
- Two voices that sound alike are treated as if they are alike
The fact that people say that it sounds like Scarlett Johansson means that there IS a region of multi-dimensional vector space that is claimed to be "owned" by the actress and that actress alone.
I HATE the end conclusion that came resulting from this mob spectacle: That people with voices that sound like hers have effectively no agency with what they can do with their own voices. That they are eternally marked as "that person that sounds like Scarlett Johansson", and that whatever they do with their own voice (including consent to be used for ML training) is met with the ire of the online mob.
Isn't the main problem that they wanted to explicitly use Johansen's voice, didn't get permission and then (maybe[1]) circumvented it by using a voice imitator, while still publicly entertaining the idea that it's her voice?
Yes, OpenAI did not explicitly say it's her voice, but Altman tweeting "her", it being called "Sky", them asking her for permission twice, they definitely wanted this connection (and her real voice), IMO, and like so many tech companies, they just did not accept "no".
They could have used the same voice, called it "Moon" or whatever and stopped Altman from tweeting, and it probably would have been fine.
In this context, I understand the result. Without this context, I would not, and would agree with your points.
[1] Or maybe just used her voice samples anyway in some form, that's hard to proof / disproof without an external audit - depends on whether or not you give OpenAI the benefit of the doubt
The legal issue here would be intent. Of course it isn't illegal to have a voice that sounds like someone else, but Sam seems to have been arrogant enough to actually make their intent pretty easy to piece together with a small amount of public information. A lawsuit would allow discovery, and I wouldn't be surprised at all if internal emails even more explicitly spell out the intent to copy the voice from Her.
Proving intent is extremely hard and I always wonder how anyone that has enough power to pay attention let's their intent be documented. At least for Sam it sure seems to be due to a massive amount if arrogance.
There are only so many combinations of 160x160 pixel arrangements but Adobe will still protect theirs.
Context matters. Adobe cannot defend its logo in all contexts. The laws make this clear. You are ignoring the context of Scarlet Johansen portraying an AI character in a movie then her likeness being used in another AI character without her consent. Had she never starred in Her, had she never been approached by OpenAI, this would not be a story.
Legality vs morality, the problem is as old as the humanity. Legally they can perfectly use a sound alike, morally it's a shit move to ask the actress who voiced a very similar product, get a "no" and still decide to find a sound alike voice actress. If it was a non profit health care company you can be sure people would be much less outraged, but silicon valley's tech companies burned the little trust people had in them a long time ago, it's hard to have empathy for sociopath
Legally it becomes a problem if it can be shown that the intent was to copy the voice. In this case, it seems like Sam made that pretty clear with only a few bits of public info. If a case was brought against them I can only assume discovery would find internal emails or chats that spell out they were absolutely trying to replicate Scarlett Johansson's voice after she turned down their offer.
- There is a finite number of distinct human voices.
- The # of desirable human voices (D) < The # of distinct human voices (A)
Where D := A - (All unwanted / undesirable human voices)
- Two voices that sound alike are treated as if they are alike
The fact that people say that it sounds like Scarlett Johansson means that there IS a region of multi-dimensional vector space that is claimed to be "owned" by the actress and that actress alone.
I HATE the end conclusion that came resulting from this mob spectacle: That people with voices that sound like hers have effectively no agency with what they can do with their own voices. That they are eternally marked as "that person that sounds like Scarlett Johansson", and that whatever they do with their own voice (including consent to be used for ML training) is met with the ire of the online mob.