Thread by @henryfarrell

1. I'm really pleased about this webcast conversation- stanford.zoom.us/webinar/regist… with @zephoria @billjaneway @cmcilwain @zeynep, Marion Fourcade and me. It's part of @CASBSStanford project on moral political economy, and a forthcoming @DaedalusJournal special issue.

2. What Marion and I are contributing is a piece on what we call "High Tech Modernism." It will be part of the special issue (which @margaretlevi and I are editing), and we're excited to see it going out into the world (draft is available at dropbox.com/s/3wy36804jhlc… ).

3. The idea behind it is pretty simple, but, we think, useful. James Scott has famously written about the "High Modernism" of the 19th and 20th century - the process of bureaucratic categorization and information collection that reshaped the world and made it "legible."

4. We argue that High Modernist bureaucracy and 21st century machine learning have something specific and fundamental in common. Both are engines for classifying things. Hence, we can think about machine learning and related forms of algorithmic decision making as closely related

5. Machine learning is a kind of "High Tech Modernism," which has crucial features in common with its paper-and-filing-cabinets predecessor. It relentlessly classifies and reclassifies people, things and situations.

6. Equally, there are very important differences. High Modernism imposed pretty crude categories (filing cabinets had their limits) and had to radically remake society (imposing surnames; replacing rookeries with the clean straight lines of boulevards) to make people visible.

7. High Tech Modernism, can apply much more individualized classifications, and change them on the fly (although as @kjhealy has pointed out to us, the old patterns of class and stratification keep on re-emerging in the new more finally tuned images of the world).

8. And it doesn't need wide boulevards to make people visible. Instead, it can capture their behavior from sites like this one, and the ever improving sensors that we all carry about in our pockets, and use to check the weather, call our loved ones, arrange meetings and so on.

9. Perhaps most politically important - the classifications imposed on us and our behavior by machine learning are far less _visible_ than their High Modernist predecessors. We are invisibly shunted into categories that affect our credit, the choices that are offered to us etc.

10. In ways that are woven into our everyday lives, and far harder to identify and mobilize against (when they discriminate against us) than their cruder 19th and 20th century equivalents. There _are_ politics around and against High Tech Modernism - but they are harder.

11. Thinking about machine learning in this way helps perhaps to change some of the technical debates about bias. Bias often can't be readily fixed through better training data, since it is woven into the nearly invisible feedback loop between the category and the categorized.

12. As politics becomes sidelined, people don't have the opportunity to challenge the ways in which they are categorized. That is a problem that avoids simple technical solutions and introduces new and more difficult questions - how do we bring politics back in?

13. Equally, the distinction between High Modernism (which was horrible in its own special ways) and High Tech Modernism potentially allows us to identify what is happening, without romanticizing the recent past.

14. For example, a lot of the fights over Silicon Valley can be understood as a bitter dispute between two rival elites - one, the professional classes of High Modernism, whose power is waning, and the other, a new priesthood, whose might is waxing.

15. There's plenty more that Marion and I could say (and likely will in the webcast conversation) - but that should be enough to give a flavor of our ideas and whether they're of interest. Finis.