Modeling natural communities is a tool that serves natural communities, and as such, one of the first phases of each of our development projects is the phase of mapping data models onto aspects of those natural communities. While we have the same types of data as many other social media platforms – user accounts, conversations, events, notifications, and so on – they arise from a fundamentally different premise, and therefore they get structured differently. In this post I will review how our practice of modeling natural communities manifests in the features of the platform and in our future roadmap.

Who is the customer?

Our customer is the community; as far as we know, we are the only social media platform built around that idea. A community in our model means a group of people in a place who coalesce around a specific interest, such as the American comics scene, or lavender gardeners in Provence. Communities overlap and connect at different geographic levels and broadness of the interests, so really there are communities within communities all the way up to a global scale.

For most of the big social media platforms, the customer is the advertiser and the users are merely the commodity of attention that is traded in an attention market. By contrast, some platforms (like Zoom or Slack) treat the paying user as the customer, and their feature sets are geared towards individual benefits for each user. While Plug is closer to the user-as-customer model, there are differences between that and the community-as-customer model. First of all, users do not pay anything directly to Plug; transactions will be between people and organizations in the community, with Plug earning our keep from those transactions. Our revenue will therefore be proportional to the strength and activity of the communities, which ensures that we continue to treat communities as the customer.

In economics, an impact multiplier refers to the chaining effect of a choice to purchase something: my retail purchase causes the vendor to purchase other things, which causes additional circulation in a ripple effect. When we buy local, the impact multiplier benefits the local economy more than when we buy from a global source, because the ripples from small businesses stay in the local area longer. By the same token, Plug may have the effect of helping specific interest-communities circulate money internally and benefit from that impact multiplier. The money circulation that I am referring to is for products and services specific to the interest – such as dance shoes or classes in a dance community.

In our design meetings, it sometimes comes up that different kinds of users have competing interests. For example event organizers may want the email addresses of people who view their events, but those users do not necessarily want their addresses shared. If our customer was the event organizer, we would share the addresses. But in fact we weigh this kind of request against the need to strengthen the whole community and choose the solution that benefits the whole. In this case we may not reveal the email addresses because privacy outweighs the organizer’s need to do direct marketing.

Why do we model natural communities?

The internet is often said to be a tool that frees humanity from its prior constraints on communication. Platforms that are unconstrained by those natural limits allow people to interact in ways that were never possible before, such as tweeting to the whole world in milliseconds. However, we have noticed that platforms that have such unconstrained models yield behavior that is not beneficial to anyone and tends to suppress the benefits that natural communities have always provided: a sense of place, belonging, purpose, and lasting connections.

It is a mistake to assume that technology will cause people to evolve away from our millennia of tribal ancestry. We can rewire the internet all we want, but it does not change our own wiring. We still need festivals, rituals, and in-person connections. Just because we have the technology to serve people individualized content does not mean we should be immersed in an echo chamber reflecting only our prior beliefs; there also needs to be community-level guardrails for errant beliefs and behaviors to maintain civility.

The unconstrained models tend towards capitalist-style monopolies of attention, where attention is focused on an ever smaller set of mega-influencers, leaving the rest of us out. In the age of free communication, most people ironically have less of a voice than they had before, because we are being heard less amid the noise, even by those close to us.

For these reasons we are attempting to analyze how natural pre-internet communities work and model that.

Modeling mistakes and oversimplifying

First-generation internet tools mimicked the physical world, with electronic mail, bulletin boards, and newsfeeds. People understood those concepts, so that was a realistic starting point. But the bulletin board quickly escaped constraints and became a threaded forum model, which did not match anything in natural cultures. It is easy to model a set of “topics”, within which are “posts”, and where each post has any number of “replies”. We can understand it and use it, but that is not how humans work. The model is a constantly growing tree-shaped data structure that treats old and new the same and aims to classify knowledge in distinct arms of the tree, but our attention is actually linear in time (not a tree), we often change the subject, we classify things messily, and we forget the old. The model mismatch of software like this may be a reason it is on the decline.

Platforms in the real estate sharing economy have the challenge of modeling people’s houses. In their first generations, the model was simply a single rentable unit (a room, casita, or back yard), where the units had no relationship to each other. This can create a lot of friction for users because in fact you might rent out the whole house one week and rent two rooms to separate people another week, but the platform did not know what rental unit was a sub-part of another unit, and could double-book. The architects of those platforms probably had extensive arguments about what the right model is, as we have had at Plug, balancing the need for simplicity with the need to accurately model reality. They have changed their models more recently to be closer to reality but are reluctant to allow the full amount of flexibility that is needed to model the actual variety of properties that could be rented.

With Plug, our first model of an event was drastically simplified compared to real events. We only allowed one span of dates, and only allowed times of the day if it was a one-day event. We added multi-session events later, but the problem of having a more flexible model is that we have to express all the options to users and they easily get overwhelmed if they only need a simple event. All models are simplifications that leave out some “edge cases” (real situations that do not fit in the model). Our challenge is to minimize the edge cases without being too confusing to use.

Why are events so central?

Plug is generic social media, and we could have focused on generic communication with events as a side feature. But instead we went with events as the central feature. This is because in-person events are where human culture is born and lives, and it is what the world needs right now. Events are the binding medium of any real community.

When Plug is mature and is a regular tool in my day to day life, I imagine it will be a bridge between my online and offline life, not a separate thing. Unlike other visions of future social media, we look at it as a tool to help make connections to fulfill life goals, not as a destination that soaks up a lot of your time.

Can we move beyond social status?

Rank and reputation are real in society, always have been, and always will be. Democratizing the world is a good thing, giving people equal opportunity and equal votes, but that cannot make people equally esteemed. People are not equal; some are more valuable to the community, some are bystanders, and some are destructive.

At Plug we want to discern reputation as a kind of objective group consensus. We need to find out who is central and who is peripheral to a community in order to present the options in a way that serves the community as our customer. Some say there is nothing “objective” about human relations, as we can all have different ideas about who is or should be a leader. But the main point here is to get rough objectivity (or inter-subjectivity) from data gleaned from users, rather than focus on subjectivity as is done with the leading social media. When I log into facebook or YouTube, I see only content that is based on my connections and likes – fully subjective. By contrast, everyone logging into Plug will see the same content presented the same way in a given space. Natural communities create some level of objectivity, so we aim to calculate and show that. (We also have some subjective suggestions, but it not our main focus.)

The signals of rank come from the same sources as other social media – likes, follows, and some passive signals – but we use those data points differently. Instead of using it to recommend extreme and polarizing content targeted for user engagement, we use it to suppress extreme and spam content while highlighting the most objectively legitimate content.

We even use the word “elitism” in the code base. Although it sounds snobby, it reflects the reality of social organization. People are generally open to being contacted by people who have legitimate needs, or those with equal or higher social status. We are advised to block our contact information to keep away scams, marketers, and anything else disingenuous. Plug’s “elitism calculator” code determines whether contact information should be shared with another user. As that gets refined with a greater user base, it will make it possible to trust messages from strangers in the same community. We model spaces (communities) in a way that allows a person to be in any number of spaces, but the reputation of a person in one space is calculated only in that space. Therefore an expert in lavender in Provence might not have much reputation in some other space, and would not be able to automatically have a large reach outside of where their expertise has been proven.

What is the moderation model?

Moderation, in the sense of flagging inappropriate content, has become a social media power game and source of conflict in a way that never existed before the internet age. On many platforms there is a system of allowing anyone to say anything to an unlimited number of others, assigning certain people authority to judge whether input is acceptable, then punishing them with threats or expulsion if they fail to meet the moderators’ demands. Toxicity reigns if moderators are either not doing their jobs or if they are over-zealously moderating.

This whole situation is a side effect of poorly modeling natural communities, where much more effective and less divisive methods have always existed to reduce the effect of toxic participants. Consider a gathering with mingling and a speaker or panel. A toxic person will find that people walk away and they end up with no audience; they will never get to the microphone on stage. One has to prove oneself over time to gain access to the panel. In Plug, we are modeling toxicity with low rank and suppressing low-ranking content, so it matches natural community. We do not even need moderators, thus avoiding the power games. When viewing conversations or other content, it may say “4 other responses”, making you click to reveal them; in this way, no one is silenced, but the fringe (possibly toxic) contributors are not given the virtual microphone.

The cultural concept of purposefully trolling online gained popularity as the models of social media became increasingly divorced from natural community. It remains to be seen if our plan will work at large scale, but we think so because we believe that just as poor models gave rise to trolling behavior, better models will bring out the better side of people.

Modeling communication and relationships

In this final section, I will leave you with some more insights about how we are working on aligning our models with natural social structures.

  • Organizations are first-class entities in Plug. Users will be able to create a personal account and a connected organization simultaneously when signing up, avoiding the situation where some users are using a fictitious company name as their account name. This clear distinction allows people to share responsibility for an organization with others. We will also add organization-level communications so you can easily ask a question of an organization and then converse with its representative.
  • The “brand” is a concept that allows organizations to divide out their offerings into different categories, each with a page. We came to this aspect of the model after studying how organizations in our target communities host repeating events and courses, and found that they often have a few distinct kinds of events that they want to show separately. The idea of a brand is fairly generic and can be used for different scenarios, so it is a model that maps to more than one structure in society.”
  • “Roles” are a generic model of any kind of named connection between our main entities. People can be part of events and organizations via roles, which can be named based on the needs of the event, like adjudicators or VIP attenders or caterers. We use the same model of roles for attendance, event hosting, leadership, and services. We are expanding this to allow organizers to assign tasks and permissions, to customize how their event page displays, and handle subscriptions and other payments. Using this same system of roles, organizations can be part of events, such as a band (an organization) playing at a concert.
  • We intend to add features that let users organize how the system works for them, so it is more like a tool or utility rather than something that draws your attention down a road of endless engagement. One set of features will be bookmarking and shortlisting content. For example, you will be able to check off a number of “maybe” events that you might want to go to, view a calendar of your commitments and attendance, and thus quickly go back to where you were. A subscription service will allow you to get highly filtered notifications so you only get what you really want. In these ways Plug will feel like a trusted assistant, staying quiet a lot of the time and only telling you the important things that you explicitly requested.
  • We chose to make conversations unthreaded because that matches how in-person conversations work. Standing back and observing a gathering of people, we see that conversations are either private with 2-3 people, or public with 3 to about 8 people, and above that size they tend to become more one-way situations where most people form a non-participating audience. Conversations often stop or divide or merge to maintain limited size and duration. So on Plug we will be modeling limits on conversation size and duration, as well as adding in a feature of side conversations.
  • We are warned that “the internet never forgets” and to be careful what you commit to writing. But that was not natural before the internet: idle talk in person about current events quickly becomes irrelevant and forgotten. So at Plug we are discussing modeling a forgetting timeframe, along with specific ways to scrapbook memories for long term. Accidental insults or other slip-ups can fade and be forgotten, keeping social relationships intact, while important memories can be retained.