I spent almost all this week in New York for a meeting of the Council for a Fair Data Future. It was an amazing opportunity to spend two and a half days in detailed conversation with a range of experts, including several good friends and collaborators, people who have been long time heroes of mine, and people whose work I got to know over the course of the week. I’ve come back with a bunch of reflections, but thought I’d focus on two here.
Community and organisation
A lot of our discussions focused on the role of the communities who are affected by data collection and use. I felt that many of those discussions were at cross purposes because they failed to distinguish between communities defined by the collection and use of data and communities who self-identify as a group with some kind of shared identity.
(I’ve been learning about the Moral Foundations of Politics, and it strikes me that these different ways of thinking about community are similar to the distinction Marx made between a class in itself (defined objectively) and a class for itself (defined in terms of what people identify with). But I’m not going to use those terms because they’re really confusing, we’re not talking about class here, and adopting Marxist language may put people off!)
Let me try out some of my thinking about it here…
After battling with a thesaurus, I’m going to call these two types affected communities and express communities (though I’d be really happy if someone wants to provide a better term). A more natural term for the latter might be identity community, but that’s generally used in discourse about “identity politics” to refer to people with shared privilege-related characteristics such as race or sexuality, whereas I’m also talking here about express communities that might be formed in particular locales or workplaces.
Express communities exist outside of the data/AI context. They are the kind of community that you can express your membership of: I’m a woman, a boardgamer, a member of the Council for a Fair Data Future. We are each members of lots of express communities, and we can name them, even if somewhat fuzzily such as “parents of babies”. Some of our express communities are things we choose to be part of, some not. Some we’re part of our whole lives, some we move in and out of as we move home or jobs for example.
Larger communities tend to have smaller sub-communities within them – different teachers might be members of different unions for example – and there are frequently tensions between different sub-communities that those outside the wider community are ignorant of. One example highlighted during our discussions was the distinction between those in the autism community who think autistic people should be “fixed” to fit in better with the world, and those who think the world should be fixed to better accommodate autistic people.
Express communities are extremely important, because that’s where organising happens. You can only get organising within a community that has some kind of shared identity that provides a level of cohesion. That said, just because you have an express community doesn’t mean it’s organised: some are and some aren’t, and there are all kinds of degrees in between. Organisation might look like a shared WhatsApp or Signal group, it might include regular catch ups, it might have an explicit shared purpose, engage in action together, or steward shared resources. Organisation might include formalisation into a recognised legal entity, such as a union or cooperative.
(The Microsolidarity framework is useful here, in particular to recognise the very practical limits in the number of relationships we can handle as humans, that constrains the scale of organised communities.)
A final point about express communities is that they may have longevity outside of the membership of particular individuals. Like the Ship of Theseus, every member of the community may be replaced over time, but the community is still the community. This is significant because certain assets, such as information collected about or by the community belongs to the ongoing community rather than either individual members or the particular membership at one point in time. Depending on the rules and norms of a community, this can limit who can make decisions about sharing knowledge or data, meaning people inside and outside the community have to honour community-level decision making. I’m thinking in particular about what I’ve learned about indigenous data sovereignty here: just because someone is a member of a particular tribe does not mean they have decision making authority to share information about the tribe, or even about themselves if that reveals information about the tribe.
The flipside is that not everyone in a community with a decision-making body will agree with or adhere to the decisions made by that body. For example, I’ve heard women strongly object to the idea that decisions about data about them should be made on their behalf by male elders and to the benefit of patriarchal interests.
Let’s turn to our conception of affected communities. Affected communities are defined by the collection and use of data and AI. They might have data collected about them and their environment, in which case they’re affected by the chilling effects of data collection / surveillance and the way their lived experience gets datafied. They might be individually or collectively affected by decisions made by automated systems or by the deployment of AI.
Affected communities usually overlap with multiple express communities. The rollout of a new diagnostic AI system in a given hospital will affect patients, carers, physicians, technicians, other healthcare workers, and (if you really want to be expansive) the entirety of the hospital community, the wider local community it sits within, and providers of competitor and complementary products. Think of the impact like a raindrop falling on the surface of a pond.
In some of the work that’s been done on how to give a formal role in data governance processes to affected communities, there is a requirement for there to be a legally recognised community representative that an organisation wanting to use data has to negotiate with. This is the case in the Indian Expert Committee Report on Non-Personal Data Governance Framework, for example, where the representative is termed a data trustee (and may be an organisation rather than a person).
My argument is that I don’t think it’s likely that a legally recognised organisation (or even a combination of them) is going to be able to represent the diversity of the affected communities of a particular data or AI service. If you think about the diagnostic AI system I used as an example, there’s not going to be one organisation that can represent the interests of all the different stakeholders who will be affected by it; indeed it’s likely that different stakeholders have different interests, which need to be balanced. It’s also likely that there are significant subgroups of stakeholders with different interests: I mentioned different parts of the autism community earlier – if the diagnostic tool were one for autism there could be significant differences in opinions between these sub-communities.
This isn’t to say that data institutions (or trusts, or cooperatives etc) aren’t important; I think communities stewarding and using data for their own purposes is important. But I don’t think they’re going to solve the problem of how to exercise collective, democratic control over data and AI systems where data isn’t collected or used by the communities themselves.
Anyway, plainly there’s more work to be done in this area, not least to explore what I’m sure is a significant existing literature on the nature of community, but more importantly good practices on identifying and engaging with a complex ecosystem with conflicting interests and varying levels of capacity and engagement. The other thing on my mind here is incrementalism: in dealing with this level of complexity, it’s impossible to believe that a perfect solution is going to be settled on first time – governing data well is an ongoing process, not a one-off decision point.
Workshop design
The second thing I was reflecting on coming out of the Council for a Fair Data Future meeting was how difficult it is to run good workshops. I learned a lot about this during my time at ODI, in particular from designing offsite days for the team. Good workshops do three things:
- Generate something, such as ideas, prioritisations, analysis, or consensus.
- Create and deepen connections and relationships between participants.
- Inspire and motivate individuals, prompting new thoughts and areas for investigation.
It is hard to do all of these things, and people often focus on the first without making room for the latter two. Workshops can provide opportunities for these through just a couple of small things:
- in-workshop 1:1 interactions, such as being asked to turn to your neighbour to discuss something, or specific exercises like impromptu networking, as well as good long breaks (at ODI we also did 30 minute buddy walks)
- giving time for personal reflection, at least at the end of the workshop and preferably several times throughout it, encouraging people to write a take-away on a post-it to keep (this is something I’ve learned from Tim – a final round of people sharing their reflections is a lovely and valuable way to close out a workshop)
Some other thoughts on running good workshops:
- Get familiar with a range of workshopping techniques and exercises, and what they’re useful for. For example, take a look at:
- Think about the mix of personalities and neurotypes that you have in the room, and ensure that you have a range of activities that can include everyone; you will have to specifically design space for non-extroverts to contribute, and give them opportunities for quiet time to help them recover from doing so!
- Think about the cadence of the day and how it will fit with people’s energy levels. People are often raring to go (but lacking in focus) at the start, so that’s a good time for wide-ranging generative exercises. There will be an energy slump after lunch, but you can counteract that by forcing people to get up out of their chairs and move around.
- Think about the energy the facilitator(s) are bringing to the room, because that will be reflected by the participants. Use different facilitators for different exercises both to spread the load and to provide different energy – do need to lift the energy levels, or provide focus? Ensure that facilitators who know something about the topic area you’re discussing can hold themselves back being a participant.
Balancing these things and getting the content you want to get out of a day is not a trivial task, and workshops are never perfect. I’m sure I’ve missed things too. Materials like Gamestorming and Liberating Structures provide a lot more depth than I’ve covered here.