You can read the first part in this series here or dive straight in.
When systems get too big and powerful, people have a tendency to group together and demand a stake, a say, in the decision-making of that system. Various attempts to develop ethical guidelines for automated decision-making systems like artificial intelligence (AI) exist, as well as calls for greater accountability through mechanisms such as oversight committees. Participation, however, can take many forms.
What might it be like to be part of a machine — a robot or infrastructure — that makes decisions? ‘Part of’ in the sense that there would be a critical mass of people participating in the machine, and (crucially) that when enough of us ceased to participate it would die? Who gets to participate might depend on whether you have a stake or are impacted by the machine’s decisions. Those who participate would benefit and so might others.
Such a participative automated decision-making machine would be different to AIs that use data extracted from our own actions and networks but give us no power in how they run. It would also be different to the individualistic notion of a singularity, where our bodies are upgraded until we are each something else entirely (yet still operate as separate, competitive agents). It is a version of Haraway’s cyborg that runs on pooled resources and information.
A participative automated machine is a post-human prospect, meaning that we are no longer defined by being separate and superior to the non-human. These machines and infrastructures are constituted of and by us. Cooperation rather than competition defines them. They are not necessarily evil or a threat to our species, nor are they inherently good.
Where would such machines come from?
One place to look for such machines is in distributed systems (the other place, which I will not cover here, is in data cooperatives). Distributed systems exist without a centralised controller. The security, stability and existence of the machine depends upon enough people choosing to direct their own resources and ideas to running nodes. Distributed systems are used right now for transactions and registries, forming what we call ‘the internet of value’. In this extended, imaginary scenario, the distributed machine is also able to make decisions based on what we choose to feed it.
In a recent Moneylab talk, I outlined five phenomena for identifying and knowing these emergent machines:
• Peer-to-contract technologies
• Automated recursive public infrastructures
• Communication and play
I discuss these briefly below.
In an effort to remain decentralised and overcome some of the inefficiencies of peer-to-peer exchanges and trading, blockchain developers created automated financial infrastructures. DeFi, short for decentralised finance, includes decentralised exchanges, prediction markets, asset management tools, and derivatives. DeFi DAPPS (decentralised applications) are mostly built on the Ethereum blockchain. These are what Stani Kulechov of Aave has named “peer-to-contract” technologies.
The term peer-to-contract refers to something more than smart contracts, which are agreements that execute a command when certain conditions are met. The contracts that underlie DeFi contain a collection of assets sent to a contract by willing contributors. Algorithms acting on these assets play a role in determining the outcomes for everyone participating. In crude terms, where a peer-to-peer model is a one-to-one interaction, peer-to-contract is a one-to-sum interaction where the sum is derived from the assets and actions of others.
For instance, liquidity pools are contracts to which people send their token savings, not unlike depositing into a bank account. Each pool, containing two or more asset types, can be constructed for defined users or open to anyone. Algorithms and arbitrage work to balance the assets in each pool, which enables price discovery. If you do not like the dimensions of one liquidity pool you can leave and join or create another.
Liquidity pools are currently being used for a variety of purposes, including: to maintain a balanced portfolio of assets; as volume from which automated market makers determine price; for stability of stable coins; to create synthetic tokens that earn interest; as collateral for loans. Experiments in the peer-to-contract model are looking to data repositories/markets (such as Ocean protocol) or non-fungible token pools.
Participative intelligence and the social life of tokens
Taking data derived or extracted from multiple sources and applying algorithms to it is already the basis of much automated decision-making. How then are these decentralised financial tools and infrastructures different from other machine learning or AI systems?
The common assertion is that distributed technologies are better than centralised technologies as they overcome central control and corruption. In reality, most DeFi DAPPS are only semi-distributed (points of centralisation include some oracles and stablecoin assets). However, they are certainly more participative than other financial infrastructures. Participation is a dimension worth looking at as it changes what tokens are for.
Just as ‘things’ have a social life, so do tokens. Arjun Appadurai (1986) observed that one person’s junk may later be someone else’s heirloom. Viviana Zelizer (2011) argued that money has no single generalisable meaning, as demonstrated through practices such as earmarking. We cannot easily predict economics through individual preferences (or game theory) because preferences are relational, negotiated and socially embedded. Tokens that are deposited into liquidity pools have a different intention and outcome to those kept on an exchange, traded or sent to cold storage.
Moreover, DeFi works only when enough people are willing to back it or farm it. While we talk of code as reflecting the intentions and biases of its creators, DeFi infrastructures are imbued with the desires and tangible opportunities of their participants generally. These automated machines therefore need to be considered in relation to capabilities and not just incentives; capabilities that are both socially constituted and machine automated. They may produce new tangible capabilities, such as the ability to pay someone a token that earns interest.
There is nothing to say that the outcomes are good just because they are collectively governed. The yield farming frenzy of Q3 2020 exposed some heinous greed, a bit like the ICO craze of 2017 but (thankfully) less accessible to the general public. DeFi has already experienced a number of exploits, glitches and scams. We have also seen communities rally around these events and attempt to rescue protocols in trouble. Self-interest and common good motives are both observable in DeFi.
Participation in DeFi DAPPs, such as providing liquidity, often gives the liquidity provider governance rights conferred through governance tokens. Governance can include the ability to vote on interest rates, how a DAPPs treasury should be spent, or upgrades to the protocol. An ‘ungovernance’ (Soleimani 2020) trend is also emerging, whereby some groups are experimenting with seeking to automate as much of the system as possible. Human governance minimalism raises interesting questions around collectively constituted infrastructures that jettison the usual systems for decision-making. Predator bots are already deployed to act on behalf of their human creators, including to act on new opportunities as they find them. Therefore, while participants have agency and influence in these systems, so do non-human actors, and these can outpace our ability to defect unscarred (voting with ones feet).
Composing new capabilities
There are other ways to understand the rise of DeFi, such as its composability (meaning the ability to use existing apps, oracles and liquidity pools as building blocks for new applications). Anthropologist Christopher M. Kelty, writing on open source communities, came up with the concept of recursive publics. By that he means that code is a form of speech and open source communities effectively write the terms of their own speech:
A recursive public is a public that is vitally concerned with the material and practical maintenance and modification of the technical, legal, practical, and conceptual means of its own existence as a public; it is a collective independent of other forms of constituted power and is capable of speaking to existing forms of power through the production of actually existing alternatives (Kelty 2008, p.3).
DeFi applications could be conceived as recursive public infrastructures, which create the conditions of their own use and standards.
All financial infrastructures are communication infrastructures, performing information-matching to settle payments and accounts at the end of the day. Lana Swartz in her recent book also looks at the cultural and social network dynamics of money, from payments specific to social media platforms to credit cards imbued with status, aesthetics and desire.
The communication of DeFi is memes and games. DeFi tokens have been branded by emojis (yam, spaghetti, sushi etc). DeFi’s developers also use memes to capture attention and drive users to discussion boards that function as knowledge clubs. In order to capitalise on these applications, you need skills in crypto platforms and finance concepts, but also a gamer mentality.
When DeFi grows up?
At the start of this article I posed the question, “what might it be like to be part of a machine that makes decisions?” Firstly, we’d have to confront the fact that these robot overlords are not just made by us, but they are constituted of us — or at least require our active participation. They would be built from our own personal resources and put to work for the collective interest (or harm).
We would find that our ways of managing and working with these machines is likely to be more like play than interacting with services or corporations. Play, in this sense, is a structured and purposeful activity. The things we play, such as games and musical instruments, have parameters and rules — as will these machines.
The communication that characterises that engagement may be viral and memetic. We may encounter and learn about them in ways that feel social and entertaining, rather than formal and compliance-like.
Finally, our ideas of what governance is might need to change. Rather than seeing governance in terms of voting, rules and committees, it might look more like code auditing; data analysis rather than deliberation. We would do this in part to ensure that those who are impacted are actively involved and incentivised to stay involved. Instead of lamenting the loss of deliberation in automated decision-making, perhaps we need to see these activities and actions as citizenship too.
Postscript: Discussing post-human worlds through the current workings of blockchain technology is a loser’s game. They involve speculation and can fail spectacularly. However, these applications and communities are also valiant attempts at future-building. If we accept that the future only exists in the here and now, in the things we create and strive to change, then there is value in imagining and interrogating these futures.
About me: I am a Professor at RMIT, working across the RMIT Blockchain Innovation Hub and the Digital Ethnography Research Centre. I am also an Associate Investigator (AI 🤖) of the ARC Centre of Excellence for Automated Decision-Making and Society. I acknowledge the support of the Australian Research Council, FT19010372.
I use Medium for work-in-progress communication and may change my views and analysis as the work progresses. I welcome constructive and respectful feedback from blockchain practitioners. DM me on Twitter @elinorrennie or email firstname.lastname@example.org