Data Together is a community of people building a better future for data. We engage in a monthly Reading Group on themes relevant to information and ethics. Participants’ backgrounds range decentralized web protocols, data archiving, ethical frameworks, and citizen science.
This reading group is something your own collective can do too! We encourage you to draw on our notes document template for this month's topic. The template lists readings, calls out themes, and suggests discussion questions.
This blog post is derived from our conversation, but is not a replica of it; we rearrange and paraphrase throughout. See the recorded call for the full discussion!
BRENDAN: Welcome to the final topic of this semester of Data Together: decentralization. This is a topic that has pulled a number of us together over the years, months, minutes.
Tonight we're going to look at decentralization in three major contexts. I thought it would be nice to bucket these as technology, groups of people, and the state.
Sarah Friend's ten-minute introduction to decentralization was a great framing for a lot of this conversation. She had three questions right out the gates:
- What is decentralization?
- Do we want decentralization?
- Are there limits to decentralization?
As a starting point: What does decentralization mean to you?
MAUVE: A decentralized group is one where power is bottom- up. Individuals ultimately have autonomy.
KELSEY: One thing I want to highlight about the word decentralization is that it's defined in opposition. I don't think that we have a non-antagonistic word for what we're talking about.
What it suggests is that there's a tendency to centralize. In power structures, it's really easy to have there be one source of truth, and much harder to have there be several sources of input– and also decision-making.
Rich Bartlett has written about decentralized power structures and patterns as it relates to the Enspiral cooperative. He talks about how they have to actively work to continuously decentralize. If you're in a decentralized organization, and what that means is that you just don't name someone as a leader, then you end up with people secretly in charge, rather than transparently in charge.
STEFAN: It seems like with centralization you have the need for representation. With decentralization, there's more need for facilitation, as opposed to representation.
A lot of these examples that are upheld as examples of decentralization are of people governing themselves. What that involves and entails is a lot of conversation, facilitation, communication.
When you see centralization, there isn't a two-way street between communicators. What's communicated out is broadcast.
BRENDAN: It's very interesting to hear that almost all the initial reactions seem to originate from this political side of things. I think it's worth it to try and pull the conversation towards the technical pinning.
There's a tension that often exists when we introduce the technology framing to the politics framing, where the technologists really want us to interpret the phrase “decentralization” quite specifically: a specific suite of characteristics of a of a network and and the way that it behaves.
It was really interesting to see Vitalik Buterin say decentralization is compute power, distributed relatively evenly:
- Does the whole network compute faster?
- Is the whole network able to process more transactions than the sum of its parts effectively?
- And can we make sure that control of the network isn't concentrated in 51% of the people on the network?
That's a very cut and dried definition of decentralization. But the thing that draws a lot of us to this, in my mind, is this political framing: we're here because we want to have an effect on a method of organizing.
KELSEY: I like the specificity of that technical definition because it's also clearly about power. It's just that the power is really specifically defined in this network. Because it's code, it's made of rules.
But this tension between political and technical uses of the same words is really fascinating. There's so much cognitive dissonance in using the same language and thinking about both things in the same conversation.
When we talk about decentralization in terms of social structures, it is almost entirely about increasing individual trust. And when we talk about it in a technical context, it is almost entirely about removing the need for trust through technology.
MATT: That's confusing to me, too. From what qualities or associations does decentralization acquire a kind of moral authority for people like us?
In doing some of the readings, I felt guilty of a kind of slippage between the technical and moral/political senses of the term: attributing to the technical features, the moral characteristics of shared power.
As Deconstructing Decentralization points out, that there's no need for those technical structures to deconcentrate power. You can have technically decentralized infrastructure that still concentrates power in a small number of hands.
MAUVE: I think Bitcoin and the fintech side of decentralization is an excellent example of decentralization that's super centralized, just like projects end up having centralization hidden underneath.
In the fintech case, it's actually super obvious where the centralization is: it's inside money. If you have it, you get more money, and it's a barrier to entry to everyone else.
And then even if it's distributed, the power dynamics aren't necessarily decentralized. Decentralized is one thing, but then there's also, is it a distributed system? Is it peer to peer? And then, is there a single implementation? These are other technical qualifiers that drastically change the way it works.
ERIC: As a geographer, I come to a lot of questions through the lens of scale, but especially questions around decentralization. The question for me would be, at what scale is a network decentralized? It can depend on how far you're looking into the network or what extent of the network you're looking at.
This came up for me in Elinor Ostrom's article that we read: great, cities are doing all this work to combat climate change. And it's a sort of decentralized approach to climate change. But cities are just one level of analysis.
Yes, it's more decentralized than nation-states or international agreements. But cities also have their own structures that are very uneven and unequal, that we may not call distributed or decentralized.
HEATHER: We think of politics as being a space that's value-laden, whereas science has meant to be value-free. It hasn't always been that way, but that's how it is now.
Stefan wrote a piece where he suggested the thought experiment of flipping that on its head and searching for value-free politics, and value-laden science.
There's plenty of work out there by Bruno Latour and others about value-laden science. But then thinking of politics as something that could somehow be value-free– I thought it played with this idea of trust.
When we try to remove trust, in the case of decentralizing knowledge networks, it's a way of us looking for a value-free tech.
What happens if we stop trying to do away with the responsibility of trust in our tech, and then stopped worrying about it quite so much in our decision-making networks?
DAWN: If we want to figure out how to speak about science as value-laden, what does it mean to think about politics as value-free?
It felt a bit like Fukuyama's argument about the “End of History”: this idea, written in the 90s, that we tend towards democracy: liberal Western democracies are this ascendant form. We did it, we've solved politics! But we haven't.
So I wonder if trying to think about how something gets rendered value-free helps you understand the assumptions, and the things that are currently treated as value-free, that aren't, around you.
STEFAN: This is about seven or eight years ago now that I was writing about this. And to be clear, I'm not advocating a values-free politics, just using it as a lens. I think I was turning to Manuel De Landa's work quite a bit at the time. He wrote a lot of stuff in his book, “A Thousand Years of Nonlinear History”, about how social and political forms are just materialistic creations and you could study them in that very neutral way.
HEATHER: In relation to what Mauve said, one way I think about the money problem is in relation to hoarding. Our problem with money right now is that there are people hoarding money. Hoarded knowledge and information is again a hoarding of capital.
Stefan and I used to organize skill and knowledge sharing workshops. But we would continually butt into this problem of knowledge hoarders. They like to be the one that knows how to brew something from scratch, or they don't want other people to know where you can gather nettle.
DAWN: I think sometimes there's a failure in those spaces to actually have an accurate analysis of power. Maybe that's why they don't see what that does as a form of holding on to power.
STEFAN: Some of the aspects of that hoarding problem were in the Seeing Like a State reading.
HEATHER: I think part of the thing we're all circling around is, what people go into decentralization wanting, and then what it's capable of doing. And then what's gumming it all up.
STEFAN: I would say that the distinction between algorithmic and heuristic thinking is really helpful on this front.
If you're looking at it algorithmically, you could see the code beneath the decentralized networks themselves as a form of centralization. They had to be drafted and composed and written and agreed upon by a group of experts. Once they exist, they produce effects without too much human input, but the power is already baked into it.
The heuristic element of decentralization is more around human feelings, the ideas and the values that they have, and how they come together and reach agreement or work through disagreement.
BRENDAN: I think that framing echoes what Kelsey was saying a about this cognitive dissonance around the word “trust” carried between political and technical contexts. I think that there's a bit of a misnomer around the phrase “trustless environments”, used in a technical context.
I think the base truth of trustless environments is cryptography– in the way that cogito ergo sum can be the base truth for an entire philosophy. When you talk about cryptography, it is oriented in math, which feels like a universal law. When you encrypt something, its foundations are in mathematical proofs. And those mathematical proofs form a kind of rock upon which you can layer other techniques towards something that resembles trust.
I intentionally want to invoke this Cartesian framing (in the sense of Descartes), because it fundamentally assumes that it is possible to begin at zero and build a completely cohesive system from a single line of thought. What I think is scary about this framing is that along the way, you cannot do that without at some point layering on a technology that intertwines with heuristics, values.
As someone who has struggled with building this type of technology, I think there is a place where the heuristic enters, and it's very difficult to know where that is; you feel so confident because of the math.
MATT: This brings us to Walch and her piece on Deconstructing Decentralization. It seems to me that that what is at issue in that piece is where the ground is to be found laying the foundation of the argument. Now we can start solving problems because certainty is here, right? And in the same way, the certainty of the cryptographic solution becomes an ideological condition, which wins arguments that maybe it shouldn't win.
Walch's argument is that you can find concentrations of power, and watch them being applied to these systems that you believe are immune to the application of power. It shows that the systems are not purely technical, but are also social. And when we pretend that they're purely technical, we misrepresent them to ourselves.
I'd like to hear where you felt like she went wrong in that argument, because I was so friendly to it.
BRENDAN: I just felt like her argument was too weak. I think she attacked the wrong things.
The proof in her piece is a story about when a bunch of folks stepped in and patched Bitcoin to prevent a centralization, but they behaved in a centralized way. She uses this as evidence that this is a an impurity in the system.
To Mauve's point, you can always write another implementation. There are 25 bajillion cryptocurrencies out there, and that is itself a kind of decentralization. Bitcoin isn't the only game in town.
I think that the design of the network itself invites much more interesting questions about about centralization, decentralization, and its central tenets or claims.
We were sucked into “decentralization” for political reasons. And then the cryptography keeps us connected, because it seems to make good on some of those promises. But as soon as we hit the implementation layer, this stops being the kind of decentralization that we showed up for.
MAUVE: I think we should put trust on the people. If I want to trust someone to have my data, I should trust them in real life.
Capturing every meaning of trust in a smart contract just doesn't make sense to me. Real contracts and promises you make to people have much more ambiguity. In a machine, it's much harder to express that ambiguity and the fluid nature of trust.
I love crypto for making sure that data can't be intercepted or changed as it's being transported between people. But I'm not a fan of crypto automating decisions and stuff that humans could actually be doing in a way that makes more sense for their specific context.
KELSEY: We've talked before about the problem of scale in trust: if it's 20 people, you can look around the room and agree to be good to one another; that's how social cohesion works, in a way that it doesn't work on, say, a country scale, a city scale, an Internet scale.
What if we use cryptography in networks to create not trustlessness, but smaller communities of trust?
In the decentralized web, there are a lot of elements that you don't have to have to think about, such as data verification, because that's built into the protocol. But ultimately, you have to trust somebody to provide the root hash that corresponds to the data you want. There is no version of a decentralized web framework where you don't have to trust anyone at all.
I feel like maybe what “trustless” systems are trying to do is not to get rid of trust, but to reduce the number of entities you actually have to trust to a size that is manageable on a social level.
ERIC: There's scale in the sense of number of people, but also geography, but then also temporal scale.
MAUVE: I like the idea of reducing the number of people you need to trust; it makes trust a little bit more manageable.
The problem I have is that the specific smaller amount of people you would currently have to trust are tech bros, investors of tech companies, business people, or people with money. That's not who I want to trust as my building block. I'd rather trust a person I'm interacting with.
I think trusting fewer people is nice when those are people you can trust rather than those are people that you have to trust and can't really opt out of.
DAWN: I'm concerned about only using the concept of trust, because it's a form of methodological individualism. It expresses, at the scale of one person and their motivation, a way to think about phenomena that might be expressed differently in collectives.
If I'm in a circle of 20 where I know five people really closely, I would feel more comfortable saying something in that group than I would in another group. I was trying to think about where that trust breaks down, or if there's an implicit individualism that is being surfaced there.
Some imperfect representation of that exists in public key cryptography. We have all these chains of proofs and ways that we say we know each other to each other. Or on Scuttlebutt, peers end up transmitting in a way that also doesn't rely on individual relationship.
BRENDAN: It is a very common trick, when we're developing technologies, to co-opt language. As the the second movers, we've pulled these meanings (“trust”, “decentralization”, towards technical centers.
Mauve, you're talking about trust as a relationship between people, not as a provable thing, but we have this word trust that has been moved into this “provable thing” territory.
KELSEY: Whenever we make qualitative things quantitative, we're usually wrong and trust things too much.
I worry about anything that that turns something not math-y into something that looks math-y enough that people who don't understand how it came to be will just look at it and say, “well, that's probably right, it came out of a computer”.
MAUVE: That was the whole point of the Seeing Like a State reading to me: it was all people coming in and saying, we need to quantify. And as part of the quantification, they introduced biases and erased a whole bunch of stuff that was actually super relevant to the context.
BRENDAN: We started the conversation with this notion that decentralization is defined in opposition.
The theme of “The Most Dangerous Notion in Reinventing Organizations” is that this stuff has been around for a while. The article cites some very subtle forms of operation that different indigenous Californian tribes used before we came in and sort of data colonized them.
Maybe that represents a version of the world in which decentralization is a default state, not a counter-hegemonic force.
STEFAN: In Seeing Like a State, James C. Scott talks about metis (practical) knowledge, as opposed to officially defined practices and processes. This is demonstrated by “work to rule”, where workers within industrialized factories would strike by working exactly according to the rulebook as set out by management. It made everything totally inefficient and ineffective. The humans developed their own techniques to operate their machines the best way.
I think the same can be said about the blockchain space, which then to further on Brendan's critique of the Walch article. You could say that article does miss the point because it's holding that entire ecosystem to an impossible standard, because there will be a need to jump in and tweak the technicalities to help it work better. That's just how we relate with technology.
MATT: The the idea of an algorithmic solution is related to the idea of a rigid, legal structure which guarantees equality before the law. This is the kind of ideology that promotes those modernist systems that James Scott is deeply opposed to and has very important critiques of. But it was developed to address important kinds of injustice.
What is it about some systems that seem to promote or to dampen injustice without the rigid formality and weaknesses of algorithmic thinking?
One thing about systems that are based in the land is that there's a shared commitment to preserving the land. You can't get away with with abusing that commitment because the land is not forgiving of your destruction of it.
The question is, where is that commitment in these other endeavors, or these other technologies? How do you guarantee that commitment? Or how do you how do you promote or create that commitment?
- Sarah Friend. Decentralization and its Discontents (Slides)
- Angela Walch. (2019) Deconstructing ‘Decentralization’: Exploring the Core Claim of Crypto Systems: pp. 11-24
- Kleppmann, Martin & Wiggins, Adam & Hardenberg, Peter van & McGranaghan, Mark. Local-first Software: The seven ideals for local-first software
- Brancati, D. (2006). Decentralization: Fueling the Fire or Dampening the Flames of Ethnic Conflict and Secessionism?: pp. 651-660, 681
- Elinor Ostrom. Green from the Grassroots
- Rachel-Rose O’Leary. This North Syrian School Is a Baby Step Toward a Blockchain Society
- James C. Scott. (1998) Seeing Like a State
- pp. 309-311 (beginning of chapter 9),
- pp. 323-328 “Practical Knowledge Versus Scientific Explanation”
- pp. 333-339 “The Social Context of Metis and Its Destruction”
- Jessica J. Prentice. The Most Dangerous Notion in Reinventing Organizations
- Adi Robertson. How the Biggest Decentralized Social Network is Dealing With its Nazi Problem
- Darius Kazemi. Run Your Own Social
- Longer talk of Sarah Friend. Decentralization and its Discontents