The 3 Forms Of Decentralization Required For A Moderation System
The Decentralized Moderation Paradox — Part 2
When it comes to censorship resistant systems, centralization should always be perceived as a major red flag. Concentration of technical resources and concentration of decision power is sure to create a censorship-prone system. Those two forms of centralization have been clearly identified by all crypto and blockchain projects. We will review them and then introduce a third form of centralization that needs to be considered when dealing with subjective opinions,which is a major part of a moderation system.
A censorship resistant system has to be distributed and autonomous.
1. Autonomous Decision Making
A fully deployed censorship resistant system shouldn’t rely on any kind of leader , not a person, not a team, nor a company. The definition subjective concepts that apply to the entire community shouldn’t be left to asmall group of people.
This aspect is even more critical for global platforms which aim to operate in many different regions, with different cultures, religions and political views.
For example, with a centralized moderation system, the definition and application of concepts like “inappropriate”, “offensive”, “propaganda”, “fake” or “real news”, would just reflect the sensibility and political views of a small subset of the community.
When we talk about “Autonomous decision making”, we don’t refer to some automated software-based actions like when a robot vacuum cleaner moves by itself. “Autonomous decision making” is about the ability of a system to operate and govern itself, potentially involving every single one of it’s participants. The greater the number of participants, the harder it becomes for a single party to influence decisions by threatening people or buying their votes.
An autonomous and distributed system presents a surface of attack as big as itself to an opponent willing the tamper with a decision. Going after some of the components of the system will not impact the integrity of system as a whole.
2. Distributed Infrastructure
An autonomous system, as described above, would still be highly vulnerable if only executed on a handful of physical servers. An effective protection against attacks requires for a system, not only to be autonomous in its decision making process, but also to work on top of a widely spread physical and logical infrastructure.
As it has been proven by Bitcoin, a fully distributed and trustless network is very effective against tampering attempts. It is critical for censorship resistant networks to be distributed in order to render irrelevent the attacks against a few ot its components.
Ideally, all aspects of the infrastructure would need to be distributed: the hardware, the software as well as the providers.
Indeed, we can easily understand how a million nodes are a lot harder to attack than ten.
Similarly, a truly distributed ecosystem should be composed by a variety of applications coming from a variety of providers. If only one company provides the unique piece of software installed on a million nodes, it represents a well defined point of attack (or pressure) for governments or hackers to target. This is quite scary, as it it how most platforms and products are developed today. It is lot easier to convince, influence or threaten a single CEO or a board of directors than to go after a million users, operating a million nodes, which are themselves ran from hundreds of different applications.
3. Globally Decentralized yet Regionalized
Decentralized moderation needs to consider geographical concentration of opinions as the third form of centralization to overcome. Indeed, when it comes to subjective notions, we need to be aware of the cultural and political differences that exist across the globe. We oppose here the “regionalized processing” to the “globalized processing” of the information.
A global system, which processes all opinions in a single pool, regardless of the source location, is sure to introduce centralization.
Most of the early adopters, for example, are likely to come from a handful of locations, but when the system grows, their opinion shouldn’t overweight some potential local consensus arising in other regions; even if the simple math would lead the global system to ignore a smaller group forming a new local consensus.
Additionally, a system which offers an incentive mechanism for labor (i.e. moderate), will lead to a majority of participants (i.e. moderators) coming from low-wage countries since the time/benefit ratio will be more viable there. Therefore, it can be expected that, on a global system, the opinions originating from low-wage regions will come with greater volume and consequently carry a bigger weight.
One last example: the interpretation of “legal” or “illegal” can vary dramatically from one country to the other, especially when (subjectively) applied to a piece of content. Imposing a single legislation over the entire platform would also present a form of centralization.
A regionalized system would take into account the different consensuses emerging from a different locations, even if the numbers don’t measure up to a possible “global consensus”.
The concept of regionalization of the consensus is in direct conflict with the previous sections in which we warned about the risks of decreasing the surface of attack. This is why regionalization shouldn’t mean “hermetic isolation of the regions” but rather “consideration of local sentiment”. This could translate to many different technical applications, and there is a fine balance to be found between “how big a subgroup should be before being taken into account” versus “the critical mass to reduce the risk of attacks”.
Regionalization shouldn’t mean “hermetic isolation of the regions” but rather “consideration of local sentiment”
This also hints at a multi-consensus mechanism where users in different regions could be presented with a different consensus for the same topic or content while using the same platform.
Dream on
When digging into the decentralization requirements described previously, we quickly realize that complete autonomy, distribution and regionalization of a system is going to be extremely hard to achieve. In the short term at least. There are many cases where centralization cannot be avoided. This is true for all so-called “decentralized” projects, and the concept of decentralized moderation is no exception.
We will next look into some of the reasons why centralization is unavoidable for the foreseeable future, despite having clearly identified how, in its many forms, it represents a major threat to a moderation system.
