[ai-control] Re: Considerations for AI Opt-Out

Martin Thomson <mt@lowentropy.net> Mon, 20 May 2024 00:22 UTC

Feedback-ID: ic129442d:Fastmail
User-Agent: Cyrus-JMAP/3.11.0-alpha0-456-gcd147058c-fm-hotfix-20240509.001-g0aad06e4
MIME-Version: 1.0
Message-Id: <ae1ccb43-9ce6-4c17-a9d8-6529023bc24f@betaapp.fastmail.com>
In-Reply-To: <3C9CA076-63D2-4E82-A13B-A02C6921B5F0@mnot.net>
References: <66D05E57-6474-414D-AF4C-409B063E970F@mnot.net> <AM5PR01MB299431A3FBD50BD1F64E64F6A8EE2@AM5PR01MB2994.eurprd01.prod.exchangelabs.com> <3C9CA076-63D2-4E82-A13B-A02C6921B5F0@mnot.net>
Date: Mon, 20 May 2024 10:21:40 +1000
From: Martin Thomson <mt@lowentropy.net>
To: Mark Nottingham <mnot=40mnot.net@dmarc.ietf.org>, Chris Needham <chris.needham=40bbc.co.uk@dmarc.ietf.org>
Content-Type: text/plain
Message-ID-Hash: WROBUOTEEW5HSB4LZZBB5IZGOMFUAM6T
CC: "ai-control@ietf.org" <ai-control@ietf.org>
Precedence: list
Subject: [ai-control] Re: Considerations for AI Opt-Out
Archived-At: <https://mailarchive.ietf.org/arch/msg/ai-control/4GwKmFqJewQ__DGi2w-iUNFDOqA>

On Sat, May 18, 2024, at 13:21, Mark Nottingham wrote:
>> For specs like IPTC and CAWG's that embed the opt-outs in image or video assets, when content is voluntarily uploaded to such sites (rather than being scraped by them), would the social site's T&Cs apply, overriding any assertions in the content itself?
>
> Good question. I suspect precedence needs to be specified, or each site 
> will decide however is most advantageously to it. That might mean the 
> most restrictive policy available always wins.

That's one potential reasoning to follow.  And a fairly reasonable one.

Another might be to have site-level policies apply only in the absence of a content-level policy.  That is, 

1. when the content author did not express a policy, defer to that expressed by the site owner
2. when the content authors does express a policy, that applies because that is the operative policy for that work, which might only be displayed on this site

Consider a forum like Reddit, which might have a general rule against the use of all content (without licenses[1], that is).  Content posted by people on Reddit might have more liberal rules for that specific content that might be more or less permissive than that of the host.  Obviously, this only applies to video/images/etc that can have policies attached through metadata.

However, training that draws on image context to apply labels for training might not be able to function in that sort of environment because it does not solely rely on the image, but how it is placed into a context and the license for that contextual information requires a license.  Training is then less useful without that sort of labeling and is less worth pursuing, despite the image itself being accessible.

Does that fit?

[1] https://openai.com/index/openai-and-reddit-partnership/

[ai-control] Considerations for AI Opt-Out Mark Nottingham
Re: [ai-control] Considerations for AI Opt-Out Michael Prorock
Re: [ai-control] Considerations for AI Opt-Out Mark Nottingham
[ai-control] Re: Considerations for AI Opt-Out Chris Needham
[ai-control] Re: Considerations for AI Opt-Out Mark Nottingham
[ai-control] Re: Considerations for AI Opt-Out Martin Thomson