Hello,
In the past two weeks, I’ve sunk into the researcher’s equivalent of an identity crisis.
The questions of ‘who am I?', ‘what do I want to do?’, or in this case, ‘what do I want my research to be?’ typically hum quietly in the background, but once in a while the thoughts get louder and take on a dissonant overtone.
The internal polyphony grows louder and ‘who is your audience?’ becomes ‘who do you want to be?’ On the side, a trebly soprano keeps on insisting “you don’t know what you’re doing.”
———
I’ve realized that at its heart, a personal crisis is a struggle with uncertainty. The first order uncertainty of what next to do, but also second order uncertainty about what values to hold, what communities to be a part of.
Of course, this is all par for the course.
(With great freedom comes great trepidation; the only way out is through etc etc)
Apprentice: Can you tell me what I can do when life gets to be too much? I sometimes feel as if I am on a wild horse that’s going to throw me at any moment.
Ngak’chang Rinpoche: Well—last time I was on a wild horse I rammed my heels down and wrapped both hands round the pommel. I was lucky to have been riding a Crates Endurance saddle and lucky to have remembered Melissa’s constant admonitions vis-à-vis keeping my heels down (Melissa is my riding teacher in Wales). Rosy, the evil Worcester I was riding up near Krinklehorn in Montana, was spooked out by the wind and rattling tarpaulins, so she decided to run me in a circle in an attempt to buck me off. She made about a dozen go-rounds before she stopped, and as I had no desire to hit the ground, panic was not an option.
An amusing story perhaps, but I tell it for a reason. You need to know where your heels and pommel are in terms of practice. Your greatest security is often found in insecurity [italics added]. When we panic—on horseback—we tend to flinch into a fœtal position. That is why we fall.
Chogyam, Ngakpa; Khandro Dechen. E-Mailing the Lamas from Afar . Aro Books, Inc..
Writing
Essays
I’m starting a short series on what I’ve learned from Elinor Ostrom and how her work is important for understanding the digital governance landscape.
The first essay, What is an institution?, contextualizes her A Grammar of Institutions as the synthesis of 20th century debates about what an institution should be. I explain why her framework is powerful and how it might help us learn how to better govern online.
Notes
AI as an institution; monolithic models of corporations as agents fail as the entities get more complex. We might expect AI to do the same; as it develops, modeling it as a single 'agent' with a utility function is less and less helpful (maybe it isn't already).
Moral curiosity over moral ‘clarity’; to paraphrase Emerson, moral certainty is the hobgoblin of little minds, adored by little statesmen and philosophers and divines. If we take seriously the nebulosity of morality, moral certainty is dangerous. Curiosity is one strong antidote to this.
External Links
There is a major replication problem in many fields of machine learning. Machine learning engineers certainly know it and it’s talked about in university labs. For the most part, it isn’t much publicly acknowledged. In the case of reinforcement learning, the top research labs (yes — probably the ones you are thinking of) have a large private cluster that runs experiments on multiple ‘seeds’. They choose and report the best. Here is a technical paper reacting to this and proposing a solution. Anecdotally, a number of engineers and myself have several stories about the failure to replicate a number of influential papers (specifically in deep reinforcement learning). [h/t P]
“Father of the personal computer” Alan Kay on television and democracy, Oppenheimer and the San Francisco Exploratorium, and of course, the catalysts of personal computing. (Highly recommended)
“ALLAN: So one of the things we're really bad about is, because of our eyes, you can't get the visual point of view we want. Our eyes have a visual point of view of like 160 degrees. But what I've got here is about 25, and on a cellphone it's pathetic. So this is completely wrong. 100% wrong. Wrong in a really big way. If you look at the first description that Engelbart ever wrote about what he wanted, it was a display that was three feet on its side, built into a desk, because what is it that you design on? If anybody's ever looked at a drafting table, which they may not have for a long time, you need room to design, because there's all this bullshit that you do wrong, right?
This is why experienced programmers have big multiple screens. They're working on something where the result is going to be fit on one screen, but you have to have all this other stuff, it's like when you make an arch; it's not just piling up the bricks, you have to put this whole scaffolding up. You have to hold everything together until you get the keystone in place.”
Books
Constructing Grounded Theory, Kathy Charmaz. Not actually sure how I feel about this. Grounded Theory is a movement within sociology led by Anselm Strauss. It is in part a reaction against the logical positivist dismissal of qualitative research. ‘Qualitative research can lead to theory building in a systematic way that is just as valid as quantitative research’ is the thrust of the argument. The problem for me is that it still feels mired in the issues around philosophy of science and postmodernism. Ostrom, for instance, remained largely positivist in her orientation.
Planning
Review
Quantitative analysis + essay (not done). Basically got stuck on what the precise hypotheses were and whether the sample I have is meaningfully representative of ML at large
Get in contact with people who know what they’re doing (G, S, T, H,…) (done - waiting to hear back)
Finish notes on Understanding Institutional Diversity. (not done)
Create a timeline (not done)
Check in with IRB (done; Houston, we have liftoff!)
Next Two Weeks
Create presentation for lab meeting
Theoretical sampling of repos. Decide on the 4-6 projects
Make a grouping of similar papers in lit review
Generate a few different potential methodologies
Sit with uncertainty