This policy clarifies our position on the use of AI services, particularly chatbots built on top of large-language models, within the Metagov Slack. We anticipate that such services will be deployed for research purposes by members of the Metagov Slack, and we wish to clarify and encourage respectful use of these services that balances research priorities with sensitivity to the rights of the members of Metagov Slack and the norms of the Metagov Slack as a public space.

In particular, we wish to clarify that part of the function of such an AI service or chatbot falls under our data processing policy and that part of the function of such an AI service or chatbot falls under our data publishing policy.

Data processing

An AI service or chatbot introduced as a bot, automation, or Slack application in the Slack or other digital service operated by Metagov is covered under our current data processing policy, and may process historical data in the Slack, insofar as:

1. it processes that data for the primary purpose of providing a service to the Metagovernance Project and/or the Metagov community, pursuant to Slack’s terms of service or any other relevant terms of service, and

2. any modification or improvement of the chatbot or AI service through training, fine-tuning, or other processing of the Slack’s historical data is not used or replicated in any other context or community beyond Metagov.

Note that modifications and improvements of the chatbot or AI service arising through other data and mechanisms, e.g. analysis of how users interact with the service, are covered under Slack’s terms of service.

Data publishing

A chatbot or AI service is covered under our data publishing policy insofar as it is trained on explicitly published data or data curated following the policies specified in the data publishing policy, in particular the opt-out and opt-in criteria.

This means that a chatbot or AI service operating under our data publishing policy may use the published Slack data to modify and improve itself for research purposes and for other purposes allowed under fair use laws. Practically speaking, it means that a chatbot or AI service that seeks to benefit from our data publishing policy should not interact with or process any data by an “opted-out” individual or channel, as this data is not intended to be publicly available.

Note that it is currently unclear whether the act of training a commercial AI model on publicly available data constitutes fair use; this policy is not intended to be a commentary on such legal discussions.

Transitional provisions

This policy should be regarded as provisional. After passing this policy, the board will revisit it after additional testing, in particular after the first implementation of an AI-based application or research project that uses data pursuant to this policy.