Disagree.

If this could be created without using recordings of his voice to train the model, then sure. I agree a "sound-alike" like that is fair game.

But since actual recordings of his voice were used in order to train the model to recreate the sound, inflection, cadence, etc...it should be required to get authorization to use those recordings for this purpose.

> But since actual recordings of his voice were used in order to train

Tell that to all the ppl who contributed to Co-Pilots training set..

I have the same reservations

That said, I think there's a slight difference in that this is targeting a specific individual.

I mean, yeah? That's also something we need to seriously consider. These tools have the ability to copy the labor and property of others in a way that is economically important and socially challenging, and we don't yet have a legal framework to handle it.

Should an actor playing a public figure in a biopic have to get permission to study the subject's voice from copyrighted sources in order to reproduce it for the film (without AI)?

Can actors audio-visually mimic their characters with increasingly convincing realism, and can they be artificially scaled, duplicated, extended, and applied for just the cost of a bit of hardware and electricity?

It is not sensible to hand wave the "human like" learning process and ignore economic reality. This area requires careful thought about how our notions of individual rights apply to new technologies, and what kind of economic system can exist as a result.

One of the popular sentiments here seems to be that training AI should not be treated any different than a human understanding & learning the source material.

Going by that logic, can the AI here be considered a "person" performing mimicry after being "inspired" by Attenborough ?

That attitude sounds... convenient for people (companies) training these models, but completely untenable at a societal level.

Right of publicity.

https://www.inta.org/topics/right-of-publicity/

Does fair use apply here? (I don't know) It's not like the author narrates a full-blown wildlife movie with the voice of David Attenborough. It's just a funny demo which doesn't depend on the specific voice (could be anyone), and also makes it clear the voice is synthetic.

If you use a small clip of Brad Pitt in a 10 sec ad without his permission, he will probably sue you and win. So I don't think fair use is going to protect you.

It is not clear that the voice is synthetic

Hold on, are we looking at the same video and Twitter message? [1] He clearly states what he's doing, and in the video he launches some program that appears to speak in Attenborough's voice. How clear is clear enough?

[1] https://nitter.woodland.cafe/charliebholtz/status/1724815159...

It’s not clear from the voice alone

By this metric I should not ask for permission to train a model on your liking and share with your wife generated pictures of you having sex with other women

Intent matters a lot when it comes to both morality and legality.

Correct except for the "share with your wife" bit. That implies intentional deception, which does not appear to be what the developer is doing.

he probably has verbal authorization XD

> "unauthorized" is a weasel word

Is not a weasel word. There is nothing unambiguous about it.

> it implies that it should be necessary for the developer to ask David Attenborough for authorization

Yes. It does. Now you might disagree with that, but it perfectly unambiguously means that.

It is ambiguous because it makes it sound like we are talking legality here, but there's nothing illegal about making a project like this if you do not profit from it. Therefore, it is a weasel word.

I think you're right, as long as it's clearly labelled as AI-generated.

What? It means nothing of the kind.

Have you never heard about unauthorized biographies?

I wouldn't call it a weasel word because it describes accurately the fact nobody gave them the authorization to use the likeness of Attenborough, and there are laws to protect people's likeness in multiple jurisdictions (https://en.wikipedia.org/wiki/Personality_rights) that might require such authorization (whether they could be applied in the case of AI remains to be tested, but apparently most of them do not question the "means" to impersonate but the act of it).

It raises an interesting discussion about this though, how is this different from someone making an impression of him and doing a comedy show of it for example. They also had to listen to him for hours to reproduce the voice correctly, are they now infringing some copyright when they use this voice?

To me, as long as you clearly state that this is an impersonation/parody, and there is no deception, I do not really see a reason to protect this, even for music. But I'd like to hear arguments for a protection and how it would be enforced, how do you judge of the likeness, how do you know/prove which source material was used for training for example.

How is it different than an impersonator? 1. Let's say there's 1000 people that could, or would, want to truly impersonate him in the world as a job. Now there's 8 billion who don't need rest, who don't need to practice, who won't be indistinguishable from the real thing.

Depends on how it’s represented. If you booked David Attenborough and an impersonator showed instead you’d be annoyed. If TV shows started using an impersonator instead of Attenborough but said it was him both the audience and Attenborough would be annoyed when found out.

I'm not sure what you mean here but let me make my point to the "no big deal" proponents.

Would you be ok if AI was used to impersonate you across the internet? Would you mind if I made a twitter account in your name and phone calls with your voice or upload to Youtube videos of yourself? I wouldn't do anything illegal or to make any money. It would just be entertainment impersonating your likeness and saying whatever I would want you to say.