Add AI policy - #285
Add AI policy#285annevk wants to merge 1 commit into
Conversation
Together with the following changes in other repositories this establishes the initial version of the WHATWG AI Policy: - whatwg/whatwg.org#510 - whatwg/meta#417 - whatwg/participate.whatwg.org#504
jyasskin
left a comment
jyasskin
left a comment
There was a problem hiding this comment.
Apologies if all this was discussed at a previous SG meeting that I missed.
|
|
||
| Artificial intelligence (AI) refers to any software that can be used to evaluate, transform, or generate content that would otherwise be reviewed or analyzed by a human. Examples include machine learning (ML), large language models (LLMs), or generative AI (GenAI). | ||
|
|
||
| The purpose of this policy is to provide guidance on the use of AI when contributing to WHATWG Living Standards, e.g., proposed specification edits, bug reports, pull requests, commit messages, or other contributions. These guidelines may be revised as available AI tools continue to evolve. |
There was a problem hiding this comment.
| The purpose of this policy is to provide guidance on the use of AI when contributing to WHATWG Living Standards, e.g., proposed specification edits, bug reports, pull requests, commit messages, or other contributions. These guidelines may be revised as available AI tools continue to evolve. | |
| The purpose of this policy is to provide guidance on the use of AI when contributing to WHATWG Living Standards, e.g., proposed specification edits, bug reports, pull requests or reviews, commit messages, or other contributions. These guidelines may be revised as available AI tools continue to evolve. |
|
|
||
| ## Review all contributions | ||
|
|
||
| You are responsible for reviewing all code, documentation, issues, and comments you submit, regardless of whether you used AI tools. You are expected to be able to answer any questions reviewers may have about your proposed changes. |
There was a problem hiding this comment.
Perhaps:
| You are responsible for reviewing all code, documentation, issues, and comments you submit, regardless of whether you used AI tools. You are expected to be able to answer any questions reviewers may have about your proposed changes. | |
| You are responsible for the correctness of all code, documentation, issues, and comments you submit, regardless of whether you used AI tools. You are expected to be able to answer any questions reviewers may have about your proposed changes. |
|
|
||
| ## Respect intellectual property rights | ||
|
|
||
| Do not input copyrighted or third-party-owned material into AI tools without authorization. |
There was a problem hiding this comment.
This looks like it forbids prompting models with almost any supporting information, including specifications. Is that the intent? Is the goal to avoid the data in prompts being used as training data? There are some models that guarantee they don't slurp prompts into training data.
The other side of IP rights is in the output of the prompts. I'd like to say "don't copy the output of AI systems into specifications unless you've checked that the output is not a copy of something that someone else has copyrighted or patented", but that's an impossible requirement. Some systems are designed to avoid the copyright issue: I wonder if those are labeled well enough that we could require people to use one? I don't think any of the model developers has tackled the patent question, and we don't actually want engineers doing patent searches.
|
|
||
| The purpose of this policy is to provide guidance on the use of AI when contributing to WHATWG Living Standards, e.g., proposed specification edits, bug reports, pull requests, commit messages, or other contributions. These guidelines may be revised as available AI tools continue to evolve. | ||
|
|
||
| ## Review all contributions |
There was a problem hiding this comment.
I have a related goal that reviewers' comments should result in the contributor's contributions improving over time. With many AI-based flows, the comments become ephemeral prompts that are completely lost after the single patch is finished. In theory, someone using an AI-based development flow could instead use the comments to improve their system prompts, so that the mistake is less likely in the future. Advice to do this doesn't really fit into any of the existing sections, but maybe it's worth adding a section on how to receive reviews productively?
|
|
||
| Do not input personally identifiable information, confidential materials, or other sensitive data into AI tools. | ||
|
|
||
| ## Using AI during meetings |
There was a problem hiding this comment.
This is a different topic than the rest of the document, which seems to be about direct changes to the standards. Maybe automated transcription (with or without LLMs) belongs in a different document, or maybe we should have 2 ##-level headings, with most of the above sections as ###-level?
|
|
||
| You are responsible for reviewing all code, documentation, issues, and comments you submit, regardless of whether you used AI tools. You are expected to be able to answer any questions reviewers may have about your proposed changes. | ||
|
|
||
| Contributions that are generated by AI require meaningful human review before submission. Please review contributions and modify as needed before requesting peer review. You should be able to attest that the work is your own, even if AI produced the first draft. |
There was a problem hiding this comment.
I think we can also suggest some things that make it easier for reviewers, such as:
- In generated text, use as little text as possible.
- Include evidence if applicable (e.g. link to source code or test case).
- Isolate issues or changes to one issue.
3 participants
Together with the following changes in other repositories this establishes the initial version of the WHATWG AI Policy: