Thoughts
In the home stretch.
I'm wrapping up my semester-long research project on machine learning classification systems. I've been investigating how to operationalize the many suggestions for governance and regulation of this space.
Here's the current draft of the intro (minus the citations). The works referenced list can be found here. If you have feedback or relevant recent research on these topics, drop me a line at rnw13@georgetown.edu. I'm eager to learn more!
Where's the ethics?
Alongside courses and research at Georgetown, I've been working through some Coursera courses this semester—Machine Learning and Structuring Machine Learning Projects. Both are taught by Andrew Ng, a very fancy guy in this space (and the co-founder of Coursera).
They're great. Ng is a terrific teacher and makes the content relatively accessible. Plus, you get to put things you learn into practice.
Still, I'm torn about them. It's a bit worrisome that all of this computational power is being put into the hands of people around the world for free without adequate context about broader research principles or responsible implementation.
The Machine Learning course dives right into the math and its implementation, which is fine to an extent. I remembered linear algebra and my stats.
But this course misses an opportunity to incorporate an ethical and research framework into the flow. These topics don't need to be introduced during the first lecture, but certainly in early lectures and carried throughout.
Some of this might be covered in the week on advice for system design and applications. I haven't gotten to that yet and will update when I do.
But this framing is certainly not the tenor of the Structuring Machine Learning Projects course. That is also framed more in terms of technical skill—how to set up projects, find errors and fix them, and not waste time down rabbit holes, for instance.
These courses are popular: 70,834 people took the time to rate the Machine Learning course, and it currently has 4.9 out of 5 stars.
You might lose some of that audience if you incorporate broader research principles into these technical courses. And doing so may not shape all minds. But it's still worth it, not least to send a signal.
Getting all nostalgic.
As my last semester comes to a close, I've found myself in a reflective mood in the brief spaces between the job search and the making money and the research wrap-up.
In particular, I've been thinking about my maternal grandmother a lot. She was pretty awesome. She lived longer than her siblings, the father of her children, and her children (my mother, my aunt, and both of their husbands). An extremely kind, hard-working, gentle person, she was at the same time tough and stubborn and resilient. She was, and remains, an inspiration.
I wrote a bit about her in a little reflection piece for one of my classes last year. You can read it here or below. It doesn't do her justice.
But it was a great experience, trying to sort out what I was learning in that class through the lens of family history. I couldn't help but think of those who raised me as I made my way through the history of computing and semiotics, from Shannon's information theory to extended cognition. Yes, I am a nerd.
My grandmother Lillian was born in 1918, a handful of years after C. S. Peirce’s death and Alan Turing’s birth. She lived to see the internet, and taught herself HTML code so she could embed midis of her favorite old songs in the body of emails she’d send me. Because she couldn’t see very well, she worked from a WebTV attached to the large screen of her television set. Needless to say, she was amazing.
I wrote some more things about AI in the gnovis journal.
I've been trying to sort out my thoughts and arguments about human agency and what has broadly come to be known as artificial intelligence. Every time I come across something related to the idea, I make notes about it in a file. And when thought clusters reach a critical mass, I write something about it.
That's what I did in my latest blog post for the gnovis journal. In this short, relatively academic piece given the outlet, I raise some questions about the narratives and language used to talk about "AI." (Am saving the topic of what "AI" actually means for another day.) Take a look and let me know what you think—I'm eager to get outside input to help develop my thoughts further.
There's a lot more to sort out on this. But the thought cluster that is closest to surfacing has to do with the idea of why for-profit companies developing AI-type technology might care, or not care, about restoring human agency to the discussion. Hopefully more on that soon.
Machine learning in the abstract.
I'm coming up for a bit of air after spending the first half of this semester talking to everyone who would talk to me and reading everything I could get my hands on about fairness and accountability in automated decisionmaking systems. Luckily, the transition period between that research and the paper-drafting phase coincided with two great conferences—the Conference on Fairness, Accountability, and Transparency (FAT*) and Princeton CITP's AI and Ethics conference. This was my first foray into academic conferences in this space, and both were energizing.
Overall, I was struck by how many strong, smart young people are leading the field or are interested in it. I watched the FAT* livestream and didn't attend in person, but heard that a large portion of the audience consisted of students, in addition to the youthful presenters. At CITP, too, a number of both panelists and audience members were young. That's inspiring.
I was also generally impressed by the openness of people to dialogue and criticism, with the goal of moving the ball forward for the benefit of society. Lofty goals, but there were also a lot of people clearly getting s*** done.
The presentations I heard and the conversations I had gave me a sense that I'm on a good track with my current line of research. In my final independent study at Georgetown (I did two of them with two different products instead of one thesis), I'm mapping out the life cycle of machine learning classification systems, potential intervention points for regulation/oversight, and the stakeholders involved. This grew out of a desire to better understand the state of research in this space and what can actually be done in practice. I hope the final product will be useful both to those just entering the automated decisionmaking/AI/machine learning/big data governance/accountability/fairness/ethics area of focus and to those who have been around for a while. (More on sorting out that mess of terminology later on.)
The process is a bit frustrating, as I know that this is a rapidly evolving area, and I'm only biting off a chunk. That's why I'm focusing my work on a high, systems level. What I produce is going to end up raising more questions than it answers, and I regret that I don't have more time as a student to pursue some of those lines of research further. But hopefully, with this roadmap, I'll be able to shed some light on necessary areas of work for others and conduct some of it on my own in a different capacity.
For now, it's back to outlining!
I have a short article out in gnovis, my MA program's interdisciplinary academic journal! It's a very quick pass through research I did for an independent study in fall 2017. Head on over to gnovis's site to read about how language can stump computers, and what that tells us about the limits of our knowledge about the world and ourselves.
UPDATE: I presented about the research on which that snippet was based at the 18th annual STGlobal Conference.
Take it down a notch.
A lot of the material out there about AI that says it's for the general public is actually just regurgitating terms that appear in more technical papers. It's not working to "democratize" the topic, despite claims to the contrary.
To get information to the public and empower people to critically assess developments, it's important to speak in terms that laypeople can understand (something I've done in the international affairs space for all of my career).
This is a list-in-progress of best practices for talking about AI to the general public—a style guide of sorts.
