Closing remarks

Niall Chithelen

By way of closing remarks, I’ll just string through major connections and questions that have emerged these last few days. I guess these are all micro/macro. It has been too rich for me to really say anything cohesive

One theme is how people come to work together. The conference theme is ‘community,’ but we’ve mapped a divergence within working together: between consensual choices to enter into community as collaborators and encountering unexpected structures, to sign contractual obligations, or to sign up for a social media platform. Each of these comes with its own set of power relations. In some of these cases, people do not realize they’re working together, or that they’re in community.

We talked less explicitly about exclusion—maybe another form of boundary work. Are the people in Southeast Asia, South Asia, Subsaharan Africa who ‘train’ AI—are these people tech workers? Similarly, it is interesting to consider the escape from technology as a process of going to a place where other people already live.

This brings me to another of the main conference themes: territory. Yesterday I wondered, can we say where this research and work takes place? Today, though, we also got an immediate sense that we were learning about China, and specifically Shenzhen, Panxu, Yunnan. So beyond the community scale, I think this prompts a question about the nation: where is the U.S., silicon valley, Austin, etc. in the other presentations? Where do people go to escape (though maybe it’s also Thailand)?

The sense of place is so essential when we consider the work of early-1800s Japanese naturalists. But less obviously so when we consider open-source coding, online liberalism, or enterprise AI. Many types of places came up, though: the university, the conference––and places like the data-driven farms that mix multiple types of places that previously seemed distinct. One place that appeared repeatedly was ‘the lab.’ What do we refer to exactly, or what do we picture when we speak of the lab in terms of AI?

And then there’s a sort of locational creep: you sign up for something with the sense that it brings you closer to a few people. Then your community scales up, and the territory it comprises does too––but this scaling can also come with deterritorialization and reterritorialization, or spatial fixing: outsourcing, supply chains, divisions of labor, and the process of exclusion scales up as well.

Relatedly, something we discussed and brainstormed when framing the conference is geopolitics, the relationship between states, but it’s interesting how little this came up until Silvia’s and Moira’s talks today.

This may intertwine with another vernacular theme of the conference, that of harm. We talked about the semi-inadvertent harm of AI and people’s physical and mental health decisions. Geopolitics brings in the deliberate use of AI as a tool of state violence. And this connects too to the political economy of AI.

It’s an odd connection but Michal Kalecki wrote in the interwar period about military Keynesianism being a growth model premised on imminent war. Fascism is too, of course. In this model, growth is necessary and natural and seems to require conflict.

Is AI part of a similar growth model? Its weaponization seems key, now, to its sources of funding, especially to the extent that AI remains unprofitable, and that its supply chains are built out in ways that are economically inefficient but seem geopolitically necessary, or the material being supplied—data—becomes new fodder, a product, for mass surveillance.

One also imagines what happens when data-based companies definitively break the law or simply lose the support of the state. Does the state seize all of their assets (which would be data)…?

Another point would be what Lily mentioned during the discussion yesterday, the targeting of AI infrastructure in wartime. The cloud is intangible, but the data center is not. Data centers, semiconductor fabs—these are potential vulnerabilities.

The data centers and fabs bring us also to a theme of materiality—also less explicitly discussed but appearing repeatedly. The domestic politics of the data center seem increasingly significant, since the data center seems to have gained a menacing and literally toxic, polluting aura, to the extent of it being a cause of major political consternation and direct violence.

There’s also an interesting shift that happens with AI, where the way we interact with our devices changes, and we expect our computers and phones to do more things for us—this can be manifested in very direct and physical ways: talking to devices, typing less, reading less. Our social or affective relationship can change radically as well, as we interact with our agent, or servant.

And there are also overlaps with existing material practices: just as we have the paper on which the Japanese naturalists drew their puffins, we could consider the train tunnel AI advertisements, the industrial park, the collection or burning of e-waste, things that remind us that physical infrastructure, digital worlds, and human bodies, remain intertwined.

These are a lot of different points and thoughts for what’s supposed to be a neat wrapping- up. And I haven’t talked about scale. At a different conference I did some years ago with Holly, also on scale, the idea of an “anti-scale” came up; something like Borges’ famous stories of the map that is one-to-one with the territory, or the man who remembers everything in his life—so it’s one-to-one but it’s just not the same, or it annihilates one’s sense of what same was supposed to be. I wonder if AI is not somehow anti-scalar in this sense; as a technology, it exists within particular political, economic, and social power relations. But as a form of intelligence, it does something very strange to our ability to assess and understand what’s going on. And as a technology, it’s an endlessly moving target.

LLMs operate at such an inconceivably large scale that decisions it makes seem, as Solon explained, de-formalized, untraceable. The scale of AI also prevents us from understanding other things—anything we’re relying on AI to figure out or assess for us has now emerged from a process of reasoning that we cannot trace.

At the same time, AI has decentralizing effects: it can empower people to make their own health-care decisions. It’s also atomizing, isolating: if you have a relationship w/ AI, do you have a relationship at all? If something goes wrong in your use of AI—its medical recommendation, or malicious act—will people really know how it happened? Perhaps through linguistics, you can analyze the prompt but you can never know what the AI did. You can use AI to make other things ‘legible,’ but what can you do to make AI legible?

To regulate, or adjudicate, the outcomes of this process, does it matter that you don’t understand the process, or the intentionality it may or may not have?

In that sense, AI is the ultimate black box.

There is an essay by Langdon Winner, which I’m slightly altering on person, called “on opening the black box and finding nothing there.” With AI, one is “opening up the black box and finding everything there.” And then perhaps you have to close the black box again, because it’s unapproachable. But also, you can ask the black box, “hey what’s going on in there?” And it will tell you something, something fabricated and probably satisfying.

This easily falls into AI’s unknowability as the central problem, or AI as inevitable. Have to revert to pure input-output logic. But, like Winner’s essay about the black box, the real problem isn’t just the technology, it’s the denial, or refusal, of agency in this process; that people could choose how to use it, choose not to use it, choose not to develop it.

And I think that is a final theme of this conference—action. What can we do, what are people doing already, and what does this action mean? At what scale does action have to take place to intervene?