Minor frictions 1: ai’s control layers
— links from the past week
Governance control layers for AI
Elaine Moore in the FT has the the most prescient quote from an AI/ML tech boss you’re likely to read this week:
The implication is this – AI/ML technologies on their own aren’t magic. To work and to be trusted to work by their users they need institutions, platforms and infrastrctures to be carefully configured around them. Media commentators and academics need to do a better job of following how these control layers are shaping who benefits from these technologies and who gets a say in who benefits.
Medical AI is social, not just technical
More on trust. Annamaria Carusi, Peter Winter explain why trustworthiness of data technologies is not a technical feature – and can’t be black boxed. It must be evaluated in context, with the people who are actually using the technology
Linking databases and data services is both a justification and a pre-requisite to drive to interconnect health and care services in the UK right now. The promise is that interconnected services create efficiencies and release hidden system value. But it’s not so simple. Interconnection is infrastuctural, contested, irreversible and is epistemically consequential says David Ribes. In other words, interconnection doesn’t just release information and value to the world – it changes what that information means and how it is valued.
Cory Doctorow was in UCL this week. He delivered convincing stuff on the causes of enshitification – a story of unimpeachable monopoly and asychronous lock-in (customers can’t run from platforms, but platforms can pivot at will). The diagnosis is good then. And the treatment? Doctorow says imposing interoperability and letting users leave at will is our best chance. I’m not so sure that covers everything though. Interoperability brings its own complexities, uncertainties and system risk. And to deal with those we’re back needing more control layers, one of the reasons platforms have prospered over the last thirty years. They’ve been able to carry and arbitrage this risk, before ultimately passing its costs back to folks like gig economy drivers. So interoperability – necessary but so far insufficient. But I’m willing to be convinced.
- Algorithmic Colonization of Africa https://script-ed.org/?p=3888
- Tech Colonialism Today https://points.datasociety.net/tech-colonialism-today-9633a9cb00ad
- Living with Data: Aligning Data Studies and Data Activism Through a Focus on Everyday Experiences of Datafication https://archive.krisis.eu/issue-1-2018-data-activism/
- Uncertainty work as ontological negotiation: adjudicating access to therapy in clinical psychology https://onlinelibrary.wiley.com/doi/abs/10.1111/1467-9566.13029
- Can artificial intelligence be decolonized? https://doi.org/10.1080/03080188.2020.1840225