This flow chart is also canonical

(sincere apologies but I do not think I can alt-text this)

I'm concerned they won't get the data they need fast enough. Will there be enough adopters? It needs to be better than a maid. The problem of bootstrapping human-generated data is hard and interesting to me. Tesla was able to sell great electric cars with alpha-level self-driving added on.

In applied organizational contexts that innovation might be in the system the model is a part of. In social science it might be a change in conceptualization or study design. Not as sure in ML research, but you see this kinda counter point between learning tasks / benchmarks and models.

But a lesson from History and Philosophy of science is that when we get stuck by limitations of our technical apparatus, just grinding on the technique isn't going to work forever. An innovation in the bigger "theory" or framework is needed to reveal why technical approach failed.

I approach ML as an applied user and as a thinker about its role in organizations and societies. The engineering mindset tends to foreground the technical work of model building, and breakdowns in the model are taken as inadequate technical work like limitations of data, cost functions, etc.

I replied to Eugene.

So I'm trying to do work and contribute to methods that can build measurement <-> understanding dynamos. Also, that's not the only philosophy of science, it's just one mechanism by which progress in science can happen.

This clarified my confusion about the "postitivist" and "interpretivist" approaches in social science. The cartoon interprevist seems to insist it's impossible to measure what can't be directly observed, while positivists often reify the instrument and lose sight of the concept.

Very broadly, my approach to "computational social science" has a close analogy to temperature. It's about positive feedback between measurement and understanding as we extend (theoretical) conceptions that are initially intuitive or embodied to data we can't experience first hand.

I would say the same. I always think about how the first thermometer was human sense perception.

Inventing Temperature is where I started.

No writer was more clarifying to my vision of my broader research agenda.

"Once the problem is mechanical, it can be solved by a machine. However, if machines can’t function, they have no role in decision making. We can only compare human to machine decisions on the problems where we level the playing field for the machine."

This SLAPS!

WOOSH!

non-ironically this will only change when you go back to scientists, engineers, and economists being utopian socialists

A similar analysis of open source repositories would be so interesting. Thanks Zach!

The work was support by NSF grants IIS-1908850 and IIS-1910202 and GRFP #2016220885. It was facilitated through the use of the advanced computational infrastructure provided by the Hyak supercomputer system at the University of Washington and TACC at the University of Texas.

I encourage you to read the blog post or the paper (here's another link) if you're interested in learning more.

This work started as a chapter of my dissertation. Thanks to the committee — Professors Benjamin Mako Hill, Kirsten Foot, Aaron Shaw, David McDonald and Emma Spiro.