Why Decentralized AI Matters Part III: Technologies
For today’s Technology Fridays section we are going to take a different approach. Instead of analyzing a specific product or technology we are going to discuss a group of platforms as part of the series I’ve been writing about decentralized artificial intelligence(AI) platforms. The first and second parts of this essay explored the market dynamics and technology enablers that have made possible the evolution of decentralized AI technologies. Today, I would like to focus some the main platforms in the decentralized AI space as well as some of their value proposition.
As explained in the previous post, there several technology movements such as homomorphic encryption, blockchain technologies and federated learning that combined to enable the first wave of decentralized AI platforms. As a result, this first group of technologies in the space combine traditional AI capabilities with sophisticated cryptographic features. Specifically, the initial evolution of decentralized AI platform has focused on enabling a decentralized and secure runtime to automate the lifecycle of AI applications. While the decentralized AI market is still in a very nascent state, already we can see a number of platforms that are likely to achieve a leadership position in the space.
Arguably the most well-known project in the decentralized AI space, SingularityNET is an open-source protocol and collection of smart contracts for a decentralized market of coordinated AI services. Conceptually, SingularityNET acts as a general-purpose, decentralized marketplace that provides a portfolio of AI agents which can be used in exchange for cryptocurrencies.
The SingularityNet platform extends AI agents with interfaces based on blockchain smart contracts that allow them to join the network and interact with third party applications or other agents. The initial version of SingularityNET smart contracts is based on Ethereum’s Solidity language but other smart contract environments should be supported in the future. To execute operations, the smart contracts exchange AGI tokens as the main economic unit to pay for the services performed by an AI agent.
I recently published an analysis of the SingularityNet platform.
OpenMined likes to brand themselves as a decentralized AI community rather than a specific platform. From that perspective, OpenMined has been implementing a series of tools and frameworks that enable the implementation of decentralized AI applications.
· Sonar — A federated learning server running on the blockchain that handles all campaign requests, holding Bounty in trust.
· Capsule — A third-party PGP server to generate public and private keys in order to ensure that Sonar neural network stays encrypted properly.
- Mine — The individual data repositories of a user. These are constantly checking Sonar for new neural nets to contribute to. The more data that is uploaded to a mine, the more relevant it becomes to Sonar.
- Syft — The library containing Neural Networks that can be trained in an encrypted state (so that Miners can’t steal the neural networks that they download to train).
I recently published an analysis of the OpenMined platform.
Ocean is trying to become the ubiquitous protocol for decentralized AI applications. Conceptually, the Ocean Protocol is an ecosystem for sharing data and associated services. It provides a tokenized service layer that exposes data, storage, compute and algorithms for consumption with a set of deterministic proofs on availability and integrity that serve as verifiable service agreements. Architecturally, the Ocean Protocol includes the following components:
· Providers: These actors have AI data or services that they make available in a cryptographically provable fashion. Services may include: data itself, storage (centralized or decentralized), compute 10 (centralized or decentralized, privacy-preserving or not), and more.
· Marketplaces: Data/service marketplaces are typically how providers and consumers interact with Ocean network, for convenience. Each marketplace is expected to facilitate features such as discovery, transactability or verification
· Data commons interfaces: Side-by-side with data marketplaces that serve priced data are interfaces for data commons, for free or commons data. These interfaces might be webpages, software libraries, and so forth. Keepers. The Ocean network itself is composed of a set of Ocean keeper nodes .
· Keeper: Keppers are responsible for collectively maintaining the network. Anyone can run an Ocean keeper node; it’s permissionless. Participation is open and anonymous.
The Effect.AI platform leverages the NEO blockchain to provide a decentralized runtime to AI applications. At a high level, Effect.AI includes the following components:
· Effect M-Turk: Effect M-Turk is a workforce on demand that allows anyone in the world to request or perform tasks that teach and develop AI algorithms.
· Effect Smart Market: The Effect Smart Market is a decentralized exchange where people can offer and buy AI/ML services and algorithms.
· Effect M-Power: Effect M-Power can distribute computational power to AI models built using deep learning frameworks such as Caffe, MXNet and Tensorflow.
Decentralized Machine Learning(DML) is a recent addition to the decentralized AI space. The new protocol provides a blockchain agnostic runtime to run machine learning models across different devices while also decentralizing other capabilities such as training or data sharing.
Algorithmia recently ventured into the decentralized AI space by launching their Danku, a new blockchain-based protocol for evaluating and purchasing ML models on a public blockchain such as Ethereum. DanKu enables anyone to get access to high quality, objectively measured machine learning models.
You can read my analysis about Algorithmia Danku here.