Online speech is the object of many regulation attempts, AI-based detection of problematic social media content being one of the most important but also of the most controversial. Apart from the technical difficulties to classify content automatically – well known in the data science community – there are also tensions between different conceptions of online speech.

We propose to the SBD++ community a hands-on workshop on content moderation that consists in forming a group of moderators to discuss a set of problematic tweets, and come up with a common "moderating style". Instead of starting with professionally annotated datasets, let us use the negotiations in a small human collective as a way to reflect and discuss participatory practices in data science.

The workshop is addressed to people :

• whose research concerns automation of content moderation,

• and/or who are interested in participatory formats in data science.

Theoretical background: research in STS and design has put forward the idea that a variety of "publics" – i.e. actors affected by or concerned with technological changes c emerge and self self-organize around technoscientific controversies. The workshop addresses whether and how the disagreements that characterize those "adversarial publics" can leverage participatory practices in data science. We believe – and are interested in your thoughts about it – that this could have bearings to rethink popular participatory formats such as dathatons and data challenges.