Urban Technology at University of Michigan week 82
Interview with Molly Wright Steenson on cybernetics, AI, feedback loops, and data
A phone that offers you suggestions for what it thinks you meant to type... a heating system in your home that turns off when a set temperature is reached... a relationship where one’s behavior changes in response to caring suggestions from their friend… a chef with the salt shaker in one hand and a tasting spoon in the other.
These are all examples of feedback loops, and thus systems of learning and development. One of the central premises of our program is that more computation in the physical world—embedded in the concrete of bridges, the walls of buildings, the streets of neighborhoods—means more feedback loops. Designing feedback loops is a 21st-century imperative. Will these loops simply be used to extract more profit, or can they be a source of equity, improvement, justice, joy, and all the other things that are also important?
It’s not a new question. Early researchers in artificial intelligence, cybernetics, operations research, and information theory were asking similar questions decades ago, at the birth of the Cold War. Molly Wright Steenson wrote a book about this, so we called her to help us make sense of the intellectual pool into which we’ve been cautiously wading.
💬 Hello! This is the newsletter of the Urban Technology program at University of Michigan, in which we explore the ways that technology can be harnessed to nurture and improve urban life. If you’re new here, try this 90 second video introduction to our degree program.
⚡️ ANNOUNCEMENTS ⚡️
🐸 What do we need to know about the early history of cybernetics and AI?
Molly Wright Steenson was working on the web before most people had ever used it. In addition to high-profile publishing projects in the pre-blog days of the World Wide Web, she has worked in various digital consultancies and taught at the legendary design school Interaction Design Institute Ivrea. Then she went back to school, eventually completing her PhD in Architecture at Princeton with a dissertation focused on… you guessed it, the era of cybernetics and early AI. We connected by Zoom to talk about feedback loops while her dog, Emoji, snored away in the background.
Bryan Boyer: As a historian and a researcher you’ve worked on a wide range of information topics, from the role of pneumatic tubes in architecture to the work of Cedric Price. This is very consistent with your early career, in a way, because you started out in the early days of the web. From web to the architecture of information seems like a logical transition, but how did you end up working on the early web anyway?
Molly Wright Steenson: As a kid I was fascinated with the phone book, and I was fascinated with being able to call someone overseas—which I once did in fifth grade—and the possibilities of telecommunication. For me the big “a-ha” with the web was being in college in 1994, working on a project called “online Wisconsin,” which was the first website the University of Wisconsin had. A group of us in the journalism school were working on it, and I had this light bulb go off: “Wait a second, I can use this to contact anyone I care about, anywhere, at any time. This is going to be huge. I need to do this now. And I need to do it before it's too late, before it's commoditized.” So I moved to New York and I got a job at Reuters.
Bryan: Thanks for indulging my biographical curiosity. Today let’s dig into the intellectual milieu that you bring to life so vividly in your book, Architectural Intelligence, including an exploration of the connections between the worlds of cybernetics/AI and architecture.
Before we start to fall back in love with the idea of cybernetics, I think it’s important for us to stop and ask: what are the things that this milieu totally missed, be it questions of social justice, racial equity, gender equity…? And is it just me, or does cybernetics come with a zeal for discovering the invisible structures of the world that verges on conspiratorial?
Molly: Have you read Geof Bowker’s “How to be Universal: Some Cybernetic Strategies, 1943-70”? He talks about cybernetics being a kind of lingua franca. It was a way to talk about systems and all kinds of ways to do exactly what you're talking about. So you're right, and Geof wrote about it 30 years ago.
So two things. In 1948 the term “cybernetics” is coined, but right around the same year communication theory was coined, and with it, notions of systems, information, and feedback. The idea was that if you could describe what a system does in terms of its flows of information, then you could compare systems. So whether you are [talking about biology]; or if you’re talking about anthropology and you’re Margaret Mead or Gregory Bateson; or whether you're talking about psychology; or you're talking about politics like Stafford Beer; or maybe it’s business and operations research (which is still dominant and absolutely central to the curricula of some business schools today), then you’re talking about systems that can be compared with one another.
This was a really powerful idea. I think the interesting thing about it is that folks today in Human-Centered Design, or Transition Design even, don’t realize all of the ways that they’re talking about cybernetics. They’re talking about the flow of information within systems. Cybernetics fell out of favor in 1970 or so because of the Vietnam War.
Bryan: How did war sneak in here?
Molly: We should talk about the way the Department of Defense was really a major player, especially for the development of artificial intelligence. For me, the most important thing I realized in the course of writing my dissertation and then my book Architectural Intelligence was to follow the money, and to follow the funding structures and the social structures around the money.
Paul Edwards defines the “closed world” as the military-industrial-academic complex and a group of folks who moved from a very select number of universities and centers for computation after World War II. It’s a brain trust of folks who often were working together on things during World War II coming together and keeping their knowledge in the same social structure [as they moved]. They would go from universities to industry to defense and back again. Take someone like J.C.R. Licklider who was at MIT, then ARPA/DARPA, [and so on]. Folks like Marvin Minsky [an AI pioneer] go back and forth on a similar trajectory.
The point I want to make about this is that moving within that closed world, as Edwards calls it, is different than moving in the structures of the National Science Foundation. The NSF was built to get money out there and to build new research programs that get more people engaged. DARPA and IPTO, which was the Information Processing Techniques Office, were about keeping money and ideas in a closed world and in specific centers. Those centers became the first nodes of the internet and the first areas of major research in computer graphics, computer-aided design, networking technology, and artificial intelligence, so it stays together like that.
There’s also the politics. It was possible to do basic research in technological scientific research with no applied goal. During the Vietnam War, the Mansfield Amendment was passed, and it only allowed university research to be done in the realm of defense if it were applied. It couldn’t be open-ended and generative. The turn toward applied defense funding led to the “AI winter” that diminished research by people like Nicholas Negroponte, Leon Groisser, and the AI Lab at MIT co-founded by Marvin Minsky and John McCarthy.
Then in the UK, in the wake of the something called the Lighthill report, which basically said, “AI was not delivering,” there were several AI winters where the bottom fell out of the promise of AI.
In response, Patrick Winston told folks in the AI lab at MIT to orient their research toward defense goals and command and control research, which they did. Negroponte’s Architecture Machine Group followed suit, because it was either that or not have funding. So, when I’m saying follow the money, follow the policy, follow the politics, that’s what you need to do if you want to understand this stuff.
Bryan: You’ve mentioned mostly men. Are there any protagonists of your research who aren’t middle-aged white dudes?
Molly: I'm glad I wrote the book that I wrote, but it’s a book that came out of a dissertation, and there are other important references. I keep thinking of Maria Göransdotter at Umeå Institute of Design and her dissertation, which is really interesting to me. It says something that’s very straightforward… very simple and yet not simple at all. She has been looking at the idea of futures and foresight in history and pointing out that where you start from changes where you go.
She uses examples coming out of the Swedish feminist history of housework and household management to point to a future of design research. She uses the futures cone idea and points out that where you start from defines the aperture [of possible futures] that opens outward. I’m completely biased because I was her opponent on the dissertation defense, but I keep coming back to that idea.
If you don’t start with white men who were born between 1934 and 1943 in the US, UK, and Germany, then you’re going to get something different. I think of the work that Charlton McIlwain does. I am really struck by his work, including the book Black Software. He points out that during the period of time at MIT I’m writing about, there are a whole bunch of people that you couldn’t write about because they couldn’t get in, because Black students weren’t accepted. But if you start a history at Clemson University in the 1980s and a couple of programs there that brought Black African American students to computer science departments, then you have a history of computation that has Black African Americans at the core, and that changes the futures that are possible.
I think also of a question that a former student of mine asked. She came during office hours and said, “you know, Molly, I have been thinking about what it would look like to have a cosmology of computing that’s completely based on India and developed within an Indian intellectual landscape. What would that look like?” I thought, wow, I hope you do a PhD! That's a really interesting question, right? What is that future like, that doesn’t have a Westernized notion of cybernetics or flows?
Boyer: It strikes me that the basic idea of cybernetics, of stimulus/response in a loop, sounds a lot like how many cultures talk about the ideal relationship humanity should have with nature. I listen to my grandfather talking about farming, for instance, and he would say that you can try something, and maybe it’s not going to work out the way you think it is, but you’ll learn from that, and you’ll do it differently next year. Sometimes it works, sometimes it doesn’t. This is partially the farmer’s way of coping with the unknown, but I think it also gets at this cosmology where humans have far less ability to control things around them, and less illusion that they have control over natural systems in particular.
“What happens if data is an architectural problem and a design problem? What happens if the question of data is human-centered? What happens when designers of all kinds understand that how you collect the data impacts what you even see the data as? What happens when designers always confront the fact that data has biases, because we all have biases? We can point out that someone crossing the street needs a bias to safely get across the street, so those biases are not always negative.”
Bryan: What do you think feedback loops meant to Cedric Price?
Molly: I’d like to zoom out a little bit from feedback loops and talk about information and the way he framed architecture, because I still find him really compelling and I've been researching his work for 17 years now. I think it’s the fact that he viewed architecture as something that wasn’t fixed, that architecture was about learning, and that learning was a way of interfacing with architecture… it’s a way of changing society and opening up society. If you have architecture that is unfixed or that challenges our fixed positions, that’s a win for Cedric Price.
He asked questions in ways that are almost riddles. You keep coming back to them. He was also really interested in the tools of project management and ways to show and diagram flows of information.
Bryan: I hear you saying that his native medium was information and that architecture is a way of working with or dealing with that information. Perhaps the way that architects often think of themselves as working with light, and you just happen to need walls and ceilings and things to manipulate that light, is a parallel metaphor here.
If we live in a world with more and more information, more data, and more computation—how might somebody today or in the next decade take advantage of this proliferation of information?
Molly: Remember that data are always from the past: the root of the word data means “given.” What happens if data is given? What happens if data is an architectural problem and a design problem? What happens if the question of data is human-centered? What happens when designers of all kinds understand that how you collect the data impacts what you even see the data as? What happens when designers always confront the fact that data has biases, because we all have biases? We can point out that someone crossing the street needs a bias to safely get across the street, so those biases are not always negative.
It would be a lot more interesting to see the questions of data capture, data development, and the correction of biases as questions for designers and architects as well as data scientists. This would mean shifting the locus of where design begins.
Bryan: This is still one of the profound differences between digital design and physical design. In digital design, data exists by virtue of server logs and so many other sources, just for “free,” and are so proliferate that it’s hard to ignore (but also hard to use well). In the design of physical products or spaces, however, it’s costly to amass a statistically significant understanding of use. The learning loop that’s there for design teams is still profoundly different on the digital side than the physical, architectural or urban side.
At a dinner years ago organized to celebrate a lecture by Adam Greenfield, who was visiting Cambridge, I happened to sit next to someone whose work I came to know and now respect very deeply, Fabien Girardin. He asked me, “Why don’t architects prototype their buildings?”—implying that they could design better buildings by learning from the prototypes. It’s a fair question, which I answered with another question: do you want to pay for your house twice? Part of the difficulty in translating the learning capacity of systems, and digital systems in particular, to the physical world is that the feedback loops are just so much more expensive to create and therefore to learn from. This is a long-winded way of saying I agree with your prompts for the ways in which designers can reconfigure their relationship to data.
OK, last question. What is your favorite city, and why?
Molly: Berlin and Minneapolis. Berlin, because it is a complicated city with a sense of humor. This was the city that, when I was in fifth grade and going to German camp in the summers, hoping to go to Berlin someday, that I was thinking about. I made it there as an exchange student in early 1990, the year the wall fell. The buildings have these pet names like the “pregnant clam” or “lipstick” or “powder box,” and there was also this history of East Germany that, some years later when I visited again, was disappearing in front of us. It’s a palimpsest, but it’s more. Not all cities have a sense of humor, but Berlin does.
Minneapolis and St Paul are my hometown. It’s also really complicated to say that Minneapolis is my favorite city today, in an age when George Floyd and Daunte Wright were both murdered, but I still love that this place taught me to love cities.
Links
🌎 P-L-A-C-E-S. Worldle is Wordle but with countries. h/t Matt Haughey
🧱 Good dissection of the externalities inherent to crypto. Excellent.
🖥 Someone makes 1:100 scale vintage computers that boot up. h/t Stephen Cohen
⚠️ The construction material pyramid stacks up the various carbon footprints so you can “think about the amount.” h/t Rebecca Williams
🛻 “A 🧵 of flat-pack cars, trucks and things” h/t Reilly Brennan
These weeks: First-ever program lunch. Nice to see faces and talk about hometowns. Planning our Cities Intensive continues apace. New tabs are spreading across the spreadsheet. Ann Arbor history deep cuts, courtesy of Reilly Brennan. Then to Boston and Cambridge: Catching up with Sarah Williams, Michelle Ha Tucker, Elizabeth Christoforetti, Justin W Cook, and meeting The Sara Hendren. Generous minds under gentle skies. 🏃