Manifesto

alt

The overarching purpose of 'mbd is to expedite the transition to AI algorithmic transparency and choice. In a world increasingly influenced by AI systems, commercial profits from the usage of any AI algorithm should trickle back to the data owners whose data was used to train that algorithm. We believe unlocking user-data-based AI algorithms is the primary, incentive-aligned solution to tackle AI misuse. 

Where we are

The initial 'mbd product is a personalized AI algorithm platform for decentralized social. Our pre-built real-time AI models are powered by rich, on and off-chain datasets. They enable developers to surface more relevant and safer social content for their users. You can launch cutting-edge personalization and content moderation filters in under five minutes. Our 'mbd recommendation system is designed to beat centralized AI equivalents like Twitter or Tiktok algorithms in a head-to-head showdown regarding relevancy, reliability, and responsiveness. We prioritize technical accuracy while adding explainability and higher trust guarantees. All this is done transparently and fairly to account for the value of people’s data in AI systems. 

Why we are here

For now, 'mbd may seem like yet another social media algorithm, but let's take a step back and look at the big picture. 

Every new technology initially has a high unit cost before it can be optimized, and this is no less true for building decentralized AI. However, the 'mbd strategy aims to bypass the usual prohibitive data acquisition, and, research and development costs by

  1. Building incrementally more powerful models that attribute back the value they create from on-chain social data.
  2. Building incrementally easier ways for developers and users to create their own AI-based social feeds. 
  3. Generating real utility by facilitating the integration of AI in Web3 apps and pushing for standarization of training data and processes in the space.

This simultaneously creates new opportunities to accelerate the adoption of existing web3 social applications.
It encourages the creation of more web3 social apps. A flourishing web3 social ecosystem will demonstrate the commercial value of ethical AI, and its benefits will gradually trickle down to other industries. 

We thus wish to sow the seeds of a community-owned AI revolution through web3 social.  

Where we want to be

Our ultimate goal is two fold.

AI for everyone

Give everyone their own unique AI assistant, which they can trust, train, and use to achieve personal objectives. 

  • Trust: AI that lives in people’s personal space and can be trusted by others—local intelligence
  • Train: AI that engages with other people AI’s to share or learn knowledge—community intelligence
  • Use: AI that can recommend, reason, and accomplish tasks—useful intelligence

Our long-term plan is to build an open and decentralized protocol for AI-based recommendations that goes beyond social media to do much more. It will work behind the scenes to put together diet plans, find and recruit candidates, suggest healthcare therapies, and handle investments and portfolios—all without compromising privacy and fairly rewarding everyone participating in the system.  

Our system will adhere to web3 principles of transparency and ownership. Developers will be able to integrate it in their applications in ways that end users understand how AI trains on their on-chain data and rewards them for their contributions to web3. 

Our AI will be accessible to both technical and non-technical users. Everyone can use AI without worrying about centralized forces manipulating and profiteering from their AI usage. 

AI data marketplace for everyone

Creating a marketplace of social algorithms and opening the design space for integrating new approaches to ranking and moderation, like Bridging Algorithms currently used in products like Community Notes. Doing the foundational work to bring powerfult cryptography to build trust-minimized versions of recommendations systems that will enable giving users strong guarantees for no shadow banning as well as unlock learning on offchain social data. You can use it to share knowledge and data privately with others by training your AI further on your web2 history (i.e., browser history, YouTube watching activity, or X liking record).

Engaging on social media by commenting, replying, and posting helpful content will no longer have negative connotations. You can contribute positively to data collections, even when "passively" consuming web2 feeds. This is a step beyond mitigating or even nullifying the adverse effects usually tied to using social media mindlessly. Guided by 'mbd systems, anyone can put more diversity and rich information back into the system!

Why we want to be there

Two questions arise.

  1. Do users care about such an AI system? 
  2. Are data marketplaces commercially viable?

Transferring power back to end users

Despite big tech's attempt at Pavlovian conditioning everyone to passively consume information over the last decade, we believe end users are intelligent enough to know what they want. Users haven't cared to fine-tune recommendations so far because 

  1. The UX to give feedback to the apps is unintuitive and frankly discouraging.
  2. Users don’t get the immediate feedback needed to actually understand how the underlying AI recommendation systems work.
  3. Users don't trust the app to work or act upon any feedback they give.

Web2 social media companies have been trying to blend ads into their feeds as seamlessly as possible for decades. Users intuitively understand that these companies are not invested in collecting explicit feedback about what end users want to see because it is against their commercial interests. It is too much power to concede.

'mbd plans to rebalance the power dynamic by giving end users the most comprehensive editing tools to customize their feeds. But we want to go a step beyond that. We will enable users to create multiple feeds using only natural language. Users will be able to experience the internet through the “lens” of others or create different feeds for different interests. Instead of reducing humanity to marketing segments based on past interactions, our AI systems will help users find the right experiences that align with their changing interests, curiosities, and objectives. 

Unlocking the data supply

Initial proponents of web3 saw the value of using a distributed ledger to remove friction and increase transparency for data marketplaces. Others saw the value of using blockchain data to solve the AI cold start problem. Despite the creation of the Web3 protocol ecosystem to support the concept, no meaningful web3 data marketplace has emerged yet.

We believe the primary reason for this is the absence of an appropriate data pricing mechanism that makes the market inefficient for smaller players. 

On the demand side, only big companies (think sophisticated hedge funds and entertainment giants) that already own large datasets can afford to buy new data. They already know what they are lacking, and can reduce the investment risk in a data purchase. On the supply side, data is abundant but impossible to monetize efficiently. For example, apps offering all user browsing history for $10! This has resulted in a fragmented data market with many layers of data brokers—a significant challenge for privacy and digital sovereignty. 

'mbd aims to solve this problem by enabling data owners to estimate their data's commercial value based on its potential to improve specific open AI/ML models and objectives. We will provide clear incentives to anyone wishing to participate in collaborative training. We hope to achieve this by providing a permissionless federated AI training environment that enforces an incentive layer in a smart contract. 

Our system will incentivize users to maximize the efficiency of pre-training and contribute to enhance fairness in post-training. We believe our approach has the potential to unlock data supply and help surface useful information trapped within the vast data collections of today.

How will we get there

We plan to use a web3-based approach of progressive decentralization to achieve our goals. We will decentralize incrementally on every aspect at every model iteration to achieve broader network effects. Progressive decentralization will finally bring the data transparency and end-user choice that all web3 developers are fighting for.

We hope to gain credible neutrality through the on-chain orchestration of:

  • Business: Evolve towards a fair, low take, rate & share revenue with all participants.
  • Governance: Evolve into a decentralized autonomous organization (DAO) that lives on-chain.
  • Technology: Move from closed source and centralized AI to open source distributed training and inference.

Our immediate goal is that second-generation 'mbd models recommend every media type (images/audio/videos) from on-chain and off-chain data sources at approximately half the price point of current models and be trained in a distributed manner between different social media apps. As a fast-growing technology company, all our free cash flow is invested back into research and development to drive down costs and bring new decentralized products to market as soon as possible. 

It is essential to note the sequence of events here. Value and utility are created first by providing the broader crypto-ecosystem with powerful AI that brings Web3 UX to parity with Web2 alternatives. This is followed by active decentralization to align incentives for all participants by retrospective fair distribution. When someone uses 'mbd-enabled web3 social apps today, they are helping pay for the development of their personal AI down the line.

Our technology pathway is summarized below.

  1. Build the best AI recommendation engine using on-chain data that bests centralized social media technology.
  2. Create the easiest tools for developers and users to create new feeds.
  3. Build the best communities by rewarding users based on the on-chain data trained on their unique & relative contribution.
  4. Decentralize technology by distributing the AI training between all / any users.
  5. Reward users based on all their data used for training as long as they can prove it's theirs.

As these steps unfold, we hope to revolutionize how people interact online and ensure the largest economy of all, the AI economy, remains open to everyone.


.