Hi {name},

In the introduction of Connected Places this week I wrote about how I think power is an important and underused method of understanding the fediverse. The billionaire oligarchs that own the centralised Big Tech platforms have so much power because there is intrinsically a lot of power and influence in having control over the communications platforms of billions of people. Decentralised social networks take that control and power away from the central leader, but that does not mean that it has disappeared altogether. Instead, power and influence has spread out over many different nodes, and often has become less visible as well. 

Spreading people out over many different places in a decentralised network, with each place being its independently owned community, would seem to create a network without ownership. But a network without formal ownership leads to new questions about power, and who gets to shape what the network looks like. 

Mastodon announced new Terms of Service for the mastodon.social and mastodon.online servers that the organisation runs. The ToS got retracted again after criticism from the community around certain clauses regarding content licensing and forced arbitration. But the interesting part is to whom the ToS actually applies. 

Mastodon announced the new ToS with a statement that "We explicitly prohibit the scraping of user data for unauthorized purposes, e.g. archival or large language model (LLM) training. We want to make it clear that training LLMs on the data of Mastodon users on our instances, is not permitted." This creates the assumption that the ToS applies to everyone. And more crucially, that the ToS also applies to a few large AI companies that do not necessarily have an account on Mastodon.social. 

Prohibiting AI companies from slurping up all data they can find without permission seems like a good thing to strive towards, and it fits with the culture of mastodon.social to strive towards that. But doing so via a ToS in a federated environment leads to interesting side effects and unanswered questions:

- An uncharitable reading of the section that prohibits LLM training on the mastodon.social data also seemed to prohibit the use of RSS readers. This seemed to not have been the intention of the Mastodon organisation. But if another instance did decide to intentionally prohibit RSS in their ToS, how should other instances react?

- What knowledge is expected from people in a federated environment? If someone on another server decided they wanted to scrape their federated timeline to use for LLM training, are they expected to check the ToS of every server that they find a post to?

- If server A has a ToS that explicitly prohibits something, and server B has a ToS that explicitly allows the same thing, who is expected to act in which manner? Fediverse servers federate by default with any server they encounter, are admins expected to check the ToS of every new server they see for compatibility?

- Are ToS from server A binding to users on server B? How does this work when the servers are located in different jurisdictions?

I don't have clear answers to any of these questions. What it does show though, is that in a federated network, setting rules for a server can quickly expand to impact many more people beyond the server itself. And with no central actor in control, it is unclear how differences between the different places and their policies on the network can and should be resolved.

Thanks for reading, and until next week!