Is a DAO a Panopticon?
Algorithmic governance as creating and mitigating vulnerabilities in "Decentralised Autonomous Organisations"
Kelsie Nabben
August, 2021
[Note: this version includes off-cuts from the original paper, that did not make it into the final published version, given the word limit. Enjoy].
“The anxiety of being watched by an unseen eye is so acute that the 18th-century philosopher Jeremy Bentham made it the basis of his plan for a humane prison, in which inmates were to be controlled by the knowledge that they might be under observation. Bentham called this instrument of ambiguity the Panopticon. Ever since then, the power of the watcher over the watched has been a focal point of thinking about social control. The philosopher Michel Foucault regarded the panoptic force as an organizing feature of complex societies. Surveillance, Foucault concluded, is the modern way of achieving social coherence--but at a heavy cost to individuality.”1
- Vladimir Z. Nuri, Economics Professor from the University of Munich shares “Spycam City: The surveillance society: part one” by Mark Boal to the Cypherpunks Mailing List in 1998.
Algorithm reputation systems in blockchain systems:
“I’m becoming concerned that DAOs are just labour camps” I remarked in passing at the beginning of a call. “More like social credit systems” replied Rich Brown, former technical community lead at MakerDAO Foundation (aka. chief cat herder – responsible for engagement and incentives in a large-scale DAO). The phrase stuck with me.
“Wait it’s all coordination? Always has been”.
A Tweet by Kevin Owocki (@owocki), Co-founder of Gitcoin, now GitcoinDAO.
“Trust level – Basic”
I first noticed I was being covertly monitored in the panopticon when I was reading a forum post on “Discourse”, a popular blockchain forum for governance discussions. On Discourse, users must create a log in to contribute posts, comments, or “reactions” to posts in the form of heart emojis. I was reading a forum post when the notification popped up *ding*, “You have been awarded Trust Level 1 – Basic”. It was then that I realised, I’m being monitored. By following the Terms & Conditions of Discourse, I learned that I had either entered at least 5 topics, read 30 posts, or spent a total of 10 minutes reading posts. I now had the power to flag posts and mute other users in the system that was designed to “sandbox” new users and grant more experienced users more rights. If I can now just read 100 posts, or spend a total of 60 minutes readings posts, I would be able to “ignore other users” and even edit my own posts for up to 30 days after posting. I understand why these kinds of rules exist in virtual spaces and that establishing reputation and trust in pseudonymous spaces is hard. Perhaps, this is why similar algorithmic reputation mechanisms are being adopted in a number of blockchain communities.
These experiences, along with months of observation and engagement with DAOs led me to investigate the question, is a DAO a Panopticon? This granularity of digital surveillance was the panopticon that the Cypherpunks were seeking to mitigate. The following paper explores the idea of “autonomy” in autonomous systems, and what self-governance means in practice in DAO trust, reputation, and governance with and of algorithms.
Case study: Praise, praise, praise
Another example of the opportunities and risks of algorithmic governance in autonomous organisations is in the calculation of how contributions are recognised and rewarded. Token Engineering Commons (TEC), an off-shoot of the Commons Stack community, is another example of a project of the trajectory to DAO-hood. Based on social experiments at Burning Man, the TEC community had tried to implement a culture of “praise”. At the end of each day when the community convened in the tent at Burning Man, they would chant “Praise! Praise! Praise!” towards those who had done a good deed to provide an overt reward for desirable behaviours. Conversely, “shaaaaame” was echoed over undesirable behaviours, rather than individuals, to provide feedback to the community and shape social norms of behaviour. Now, this social system was encoded in the “praise” mechanism of the Commons Stack community, whereby anyone could list in a Discord chat channel “!praise @name” to a person for a specific action in the DAO “worker economy”. At the end of each month, praise scores are tallied up to create a leader board of praise distribution. The praise system has become both a social, reputational reward as well as a technical “CSTK” reputation token which is distributed at the end of each month to help track the ongoing contributions of the team and community. Now, the scores determined by the praise system were being utilised to determine the initial distribution of tokens in the Token Engineering Commons builder’s pool. A proposal to change the distribution of these tokens from the raw outputs of the praise algorithm created an internal debate known as “PRAISEMAGEDDON”.
In the “PRAISEMAGEDDON” debate to change the outputs of the praise algorithm, the challengers’ analysis found that some types of behaviours were rewarded far more than others, “in particular, highly visible & high frequency tasks like attending meetings, posting on social media or ‘being a great person’”, versus highly specialised, slower volunteer labour contributions, to garner a high percentage of rewards, and proposed that a new, fairer system be decided on for token allocations. What is visible, measurable, quantifiable, and incentivised is what gets rewarded. It was proposed that contributors be sorted into four different tiers to rank them according to types of contributions made, and then ranked in each category according to their praise score to distribute tokens more equitably in each category. Like algorithms in other platforms and contexts, the community found that algorithmic rewards work, but they “give a lot of power to the people setting the incentives”. Two segments of the community now had divided perspectives over whether the praise score analysis that would determine the token allocations in the DAO be altered. One segment of the community held more tokens from the algorithmic praise process and were in favour of respecting the “immutability” and unbiased nature of code. The other thought that it was biased, unfair, and countercultural, and a new system should be adopted that favoured informed constituencies and “algorithmic oversight”.
Blockchain communities are embedded in the ideologies of the early cypherpunks. The idea in Bitcoin was that software code is immutable, or “code is law”, and should be designed so it can’t be co-opted by a monopoly or dictator. This has influenced how blockchain communities’ approach algorithms. There was strong debate over whether the algorithm should be questioned, and the rules of the system could be changed, or whether the algorithm cannot be tampered with.
“We made the system, we can make a new one”.
Github issue by Jeff-Emmett entitled “PRAISEMAGGEDON #5”.
Continue reading at:
Nabben, Kelsie, Is a "Decentralized Autonomous Organization" a Panopticon?: Algorithmic governance as creating and mitigating vulnerabilities in DAOs (August 18, 2021). Kelsie Nabben. 2021. Is a "Decentralized Autonomous Organization" a Panopticon? Algorithmic governance as creating and mitigating vulnerabilities in DAOs. In Proceedings of the Interdisciplinary Workshop on (de) Centralization in the Internet (IWCI'21). Association for Computing Machinery, New York. https://dl.acm.org/doi/10.1145/3488663.3493791