The democratic theory of “A Half-Built Garden”

Ruthanna Emrys’ new novel, A Half-Built Garden is out (Indybound, Amazon). If you want to know whether you should buy it, the answer is yes, if you like sociologically and politically sophisticated sf, if you are looking for a realistic but hopeful take on a post-climate change future, if you want a different kind of first contact story, or any combination of the above. If you’re looking for a proper review, go here.

This post does something quite different – it singles out just one of the political threads from the novel. In other words – read the book too or first to get the bigger narrative that gives it proper context. I do try to avoid big spoilers, but I can’t help giving some sense of the book’s background.

Short version: A Half-Built Garden thinks through the relationship between AI/machine learning and democracy from a very different perspective than our current one. It asks a question that very few people are asking, but that is plausibly pretty important. What would online democracy look like if AI/machine learning was used to counter individual bias rather than exacerbating it?

Emrys’ book is set a generation after a major political revolution where autonomous self-organizing networks take effective control of politics across much of the world. The Dandelion revolution isn’t itself described in detail. It’s clear that like real revolutions, it wasn’t and isn’t complete.

Much of the novel’s action takes place in the Chesapeake watershed network. An autonomous network co-exists there with a much shrunken U.S. state, centered on the Capitol buildings in Washington D.C. Many thousands of miles away, the billionaires and their servants retreated to New Zealand to create their own political system of ever-shifting corporate alliances. Much of the plot of the book describes what happens when these mutually wary political systems get shaken up by first contact. A ship lands, with two commensal alien species, with their own assumptions and complicated joint and individual politics. These alien species’ assumptions get shaken up in turn.

I’m not going to go into the plot specifics, nor all of the sociology.There’s lots of really interesting speculation about gender and family, and about the different societies that might arise among people trying to fix a planet, and people (with different numbers of arms, body shapes etc) who have given up on planetary complexity for the predictable simplicities of artificial environments. But what grabbed me – as a political scientist – was Emrys’ approach to democracy. I get thanked in the conclusions for conversation with her about politics (while I don’t remember the specifics of the conversation, I don’t deserve any credit whatseover).

Emrys isn’t the first person to write about how distributed autonomous networks might provide a new kind of democracy. You could look, for example, to the different accounts in Malka Older’s Infomocracy books, or the lightly sketched discussions of Demarchy decision making in Alastair Reynolds’ Revelation Space books.

Those are all great, but what Emrys does fantastically well is to show how her new model of democracy grows out of practical problem solving. A lot of social science fiction – indeed a lot of social science – involves thought experiments along the lines of ‘how do we conjure up fancy new technology and then decide what it could do’ or ‘the state collapses – what happens next!’ What Emrys does, if I’m right, is really different. She starts with the problems – the really quite massive problems – of climate collapse and its aftermath. She then asks how people might try to solve these problems. And she shows how a different, networked form of democracy emerges from the efforts to fix them.

I don’t know whether Emrys has read John Dewey – but there is a lot in common with how she thinks about democracy and how Dewey does. Dewey too begins with problems, and the publics that form around them. One of his big criticisms of early 20th century democracy is that its institutions are stuck around a set of assumptions – to do with nation states – that match poorly to the scale of global problems.

In Emrys’s post-climate change world, it makes sense to work through most problems, not at the level of a nation state, but at the level of a river watershed. One of the characters has a piece of jewelry in the form of a dandelion puffball, symbolizing the revolution. Design captures logic – “under a microscope, each seed stem would expand into the line of a river.” Different riverine networks talk to each other as needed – the shared communication network and its protocols become a significant plot point. But key decisions are taken at the level of the watershed, through online threaded debates.

Again – everything stems from shared problems. I don’t know if she’s a specific influence, but Emrys’s dandelion post-revolutionaries work together in the ways that Lin Ostrom’s common resources people might have worked together, if they were much more tech-ed up. Their online discussions map onto a “network defined by measurable flows of matter and energy and obligation.” Everything that can be measured is measured. The main character talks about how, when the system works, a “functional network could’ve sketched maps across my lens: trade routes and carbon mitigation and labor badges glowed reassuringly green for every ingredient.” When it doesn’t, it’s “like analyzing river chemistry by drinking a glass of water.” Everyone who affects the system is part of its Deweyan public. When the aliens land, and their emissions start hitting the watershed, “that gave them both the right and the obligation to see the data, and share the burden and privilege of mitigating their impact.”

None of these measurements, of course, change the fact that people are going to have different interests. The book’s plot is driven more by disagreement than by overt conflict and violence (which is not to say that there is no coercion at all). The dandelion networks are, in a very fundamental sense, democratic. People talk and argue. In Emrys’ description, “encoded in algorithms and input interfaces, their root was a set of ideas: that everyone brought worthwhile perception and insight to the decisions that shaped society, that our technologies embodied our values, that they should be consciously designed to do so.”

But in Emrys’ future, calm consensus doesn’t automagically emerge from people having it out with each other on online forums. Creating legitimacy, and turning people’s disagreements and value clashes into usable information is a vexing design problem. Her society works in large part because of algorithms, which are designed to make argument as comprehensible and useful as possible, weighting redundancies, and countering the multitude of individual biases that we all are subject to.

She describes a system in which “the dandelion protocols had been designed to help us get past that hardwired resistance to being wrong” that all of us are subject to, our unwillingness to admit that we might actually be mistaken. They balance against “the human tendency to fixate on short-term goals and immediately salient stimuli.” They push for interests that will be under-represented in discussion, so that an algorithm “would advocate for the river, and one that would put in arguments about within-neighborhood interdependence and so on.”

This is a really interesting alternative to our current way of thinking about online communications and how they affect democracy. The last several years have seen a lot of people arguing that online forums are turning democracy into hell. Many people blame engagement-maximizing algorithms for this, arguing that machine learning keeps people clicking on YouTube, Twitter etc by serving them up alarming, divisive and misleading content.

The evidence that machine learning is at the root of our problems is more equivocal than you might think. But Emrys’ question is an important one – and one that not nearly enough attention is being paid to. If we were to think of online discussion, not as true unfiltered democracy (as many Internet enthusiasts did up to a few years ago), nor yet as a necessary threat to democracy (as many do today) but as a democratic design problem, what answers would we get? Which kinds of online forum are likely to be more or less conducive to good democratic decision making? How could we test and improve them?

A Half-Built Garden seems to me to make two sound points about where you ought start. First – that democracy should be linked to practical problem solving. Social scientists and theorists may fall in love with decision making systems for their own sake, but ordinary people very rarely do. If you want people to engage, they need to have practical consequences and benefits.

Second – that you should start from a realistic understanding of how people actually think and argue. Emrys has a long standing interest in cognitive psychology, and it shows. Rather than starting from computer science or economists’ theoretical ideas about how people could make decisions if they were perfectly rational etc, it starts from people as they are. It asks how can a system best counter their individual cognitive flaws, while enhancing their collective ability to sometimes get things right.

Again – in making A Half-Built Garden into something that sounds like a social science text, I’m doing it great violence. It’s a novel. I’m also picking up on the aspects of the novel that I most agree with (this recent article on democracy, with Hugo Mercier and Melissa Schwartzberg starts from similar notions about human cognitive psychology), and surely doing violence to other aspects of the novel that don’t fit neatly. But I would love to see this novel read by political theorists, by social and cognitive scientists, and by people who are trying to think practically about design questions.

I have no idea how readily you could turn Emrys’ ideas into networks that both keep track of practical problems and counter for people’s biases in thinking about how to solve these problems. Still, I would love to see people try, to figure out what might work, and what might not.