Curiosity’s Wager 2: Situating: the arrangement—and our place in it
Part 2 of 10 in Curiosity’s Wager.
To make Curiosity’s Wager practical, I draw on my information science training, my dissertation work, and science and technology studies. I want to look beyond the AI model to the arrangement it participates in. That includes people, tools, information, habits, relationships, permissions, responsibilities, obligations, and duties. These all shape what happens. We need to situate these tools in time and space and across people. What brought someone to use a tool? How were they trained?
What made this seem like an appropriate action? What happens afterwards? It includes outcomes we see and don't see, direct and indirect. It includes the histories of the material used to train the models, the companies and laws involved, and the practices people develop around them. I draw heavily on the handoff analytic developed by Deirdre Mulligan, one of my advisors at UC Berkeley, and Helen Nissenbaum. It helps us examine what happens when activity moves between arrangements, for example, from people doing something themselves to using a computer or AI to help. More can change than simply the functional performance of the task. Values, responsibilities, and what becomes visible can change too. What arrangements become possible? What gets encouraged or rewarded? What might disappear from view? Importantly, I want to include ourselves in the picture. Our choices, our learning, and the understanding we develop or fail to develop.
The arrangement and our orientation to it are malleable, even if they aren't infinitely flexible. People with immense power make decisions about these tools. Still, that doesn't predetermine everything the tools will become or how they will be used. We can reshape our own practices and work with others to change the wider arrangements. There are substantial questions about training data, labor, ownership, and environmental costs. Their particulars differ across systems and places. They deserve investigation and action. For the rest of this talk, I want to focus on some enduring hazards I can examine in my own conduct as someone designing, building, using, and talking about these tools. I think they may offer useful questions for you too. I'm calling them seven vices in AI design and use. That framing has moral bite deliberately. These are things we have responsibilities to confront.
For each, I'll describe what concerns me, how it appears in my work, and how I'm trying to respond through shared curiosity, including where I haven't found yet an adequate response.
Series contents · Previous: Can we change what we believe? · Next: Credulity: what grounds this belief?