Social media platforms are our modern town squares -- they are where ideas are debated, relationships formed, and communities cultivated. But right now, these digital civic spaces are owned entirely by corporations, with all of the predictable consequences: from recommendations algorithms that reward clickbait to moderation practices that benefit advertisers.

We must bring sovereignty, transparency, and shared ownership to internet communities. This is only possible if we re-architect the social network stack from the ground up.

Karma3 Labs is working on enabling an important layer in the social network stack – Open algorithms for discovering people and content on social networks. Infrastructure that enables social graph protocols and developers to surface relevant content for their users, not controlled by a central entity, but by an algorithm of your choice, in an open and verifiable way.

Social Network stack today

Social networks today own every single layer in the stack.

Identity. Usernames and Identity (namespaces) are owned and siloed by the corporation. The moment you step out of the boundary of acceptable behavior they set, you are locked out of the network.

Social graph Data. Who you follow, who you block, and who you are friends with, all your social graph data belongs to the company. The data model for social networks is also limited and not extensible.

Content feed and Recommendations. The discovery and recommendation algorithms dictating what appears in your feed are all opaque. A user is given little to few options to directly change that. It sounds like trivial rights that you conceded, but eventually what happens is that you have no control of the information that forms your worldview.

Moderation, Trust, and Safety. The policies around censorship of people and content are also determined by product safety and operations teams at a single company. The limits of your knowledge and worldview are defined by a handful of people.

Clients and applications. The user experience is controlled by one or limited applications, mostly controlled by the same company operating the social network. This leaves users no alternatives and 3rd party developers are locked out from creating innovative user experiences.

Monetization and business model. The value captured through advertisements, payments, and commerce is controlled and owned by the company managing this entire stack.

The spill-over effect of a bundled social network

1. Everyday efforts to connect with people and engage with content is an effort to wrestle with content recommendation algorithms that aim to hook your attention fixated on the platform using daily-perfecting psychosocial hooks fed on your own data — we’ve collectively failed end-users and developers who try to build a better, open and economically fairer experience on social networks.

2. Eroded monetization for creators - diluted by bots and sybils, dependency on the platform’s recommendation to distribute content.

3. Declining user experiences because of misalignment in the content discovery process between users and these centralized networks and platforms. Advertisement revenues are prioritized over what a user really wants to see or engage with.

New Protocols are enabling the Unbundling

There are a few novel protocol architectures emerging for open social networks. Lens, AT Protocol, Farcaster, Nostr are all pioneers in re-architecting social networks, taking different approaches in their social network stack.

These protocols have some common design principles across the stack:

Identity - Sovereign. With the use of primitives like DID, Public key infrastructure, NFTs, and personal namespaces, these protocols let users control and custody their own identity. (e.g. Lens Profile NFT, Farcaster's ID combined with Eth keypair, AT Protocol's DNS names combined with DIDs)

Social graph Data - Open. The data is not hidden behind centralized servers and APIs. It can be openly indexed by any developer. (e.g. Lens BigQuery, Farcaster Hubs, AT Protocol's PDS)

Content feed and Recommendations - Developer/User choice. Developers can choose their own open algorithms from a variety of sources. They can control the weights or parameters of the algos. (e.g. AT Protocol’s Algorithmic Choice) — Karma3 Labs solves this.

Moderation, Trust, and Safety. Community decides. Social validation and attestation primitives can set policies on what should be censored. (e.g. Mastodon’s Choices of Policies) — Karma3 Labs can rank profiles based on desired or Sybil detection heuristics.

Clients and applications. 3rd party developers. Innovation at the application layer will be largely driven by open developer communities building on top of these social graph protocols (e.g. Nostr’s Clients, Lensverse)

Monetization and Business Model. Protocol participants. Users, data and service providers, 3rd party clients all have the ability to capture value from the protocols. The design space for business models and monetization is still open. (e.g. Lens Tokenization, Mastodon’s Monetization)

Open Algorithm and Verifiable Compute Layer

Twitter recently took the first step in this direction by open-sourcing its feed algorithm. Even though the critical details around the weights of the models and algorithms are private, it has set a precedent of opening up an important part of the social network stack.

The rankings and recommendation layer is a key driver for restoring the values of social networks - people should see what they are interested in and it should be easy to weed out sybil, spam, and unwanted content and people. There shouldn't be manipulation by companies on what content to show to people. There should be innovation on what algorithms to use, all aiming to give a fair, relevant, and transparent user experience.

We at Karma3 Labs have been working to enable an open ranking and recommendation layer. With it, developers can power sybil detection and personalized content discovery.

Explore Our APIs

For any social graph protocol and the developer community, our APIs can enable:

• People discovery: We index and compute the social graph structure – for everyone, with whom they interacted and in what manner – and recommend the people that warrant more interaction (such as the following). Our profile ranking algorithm helps improve the search experience - “I see 100+ profiles with ‘Elon’ in their name, who is the legitimate or trusted profile to follow?”

• Content feed: For a given personal profile, we can tailor a content feed and recommend it, based on not just what’s popular/trending but also who is in the profile’s vicinity in the social graph. This is done through our open layer: With a transparent algorithm and open data, it’s no longer a black box: “Why” each post in the feed was recommended can finally be understood!

We have already built some of these APIs for the Lens Protocol. It is open to anyone who wants to explore!

Open and Modular Architecture

Our architecture employs modular components, such as data sourcing and indexing, transformation, and computing algorithms.

Components define their interfaces, over which they then interconnect. These interfaces are abstract, in that they encapsulate implementation details. Instead, the interface design tends to be driven by the kind of data served. For example, feed recommendation algorithms for the same social network share an identical output interface, such as a sorted list of posts to show to the user, with each recommended post possibly annotated with the basis for the recommendation.

Common, abstract interfaces also enable replacing one implementation with another. This is our key approach to the open compute: Developers – and possibly end users in the future – can easily swap out one compute algorithm for another to offer/consume different kinds of feed, as long as the algorithms are interface-compatible.

Invitation to Contribute to the Open Algorithms

All our current implementations are open-source; we intend to keep it that way. Anyone can fork and enhance our implementation, to produce an incrementally improved version or even a whole new alternative version. In fact, the improvement may as well be a simple parameter tweak, so if one understands the algorithm we are using, one may be able to suggest an improvement without even writing a single line of code.

Developers building new social applications on top of these protocols don't have to bear the burden of indexing and running compute or algorithms, which involves a significant resource and operational burden. Our APIs can enable desired algorithms (weights) to power your client user experience for people and content discovery.

We at Karma3 Labs are fully on board with the new wave of unbundling social networks and are poised to help solve important trust issues as they arise. In particular, we are beginning our social network journey with rating and recommending profiles and feed content, to enable search for/discovery of trustworthy profiles and content. We do this by running transparent and verifiable algorithms on openly available, un-bundled social network data. Our work is public and open source, and we welcome anyone who wants to be on this journey together.