The rapid rise of artificial intelligence (AI) in the web3 ecosystem has made it increasingly challenging to distinguish genuine innovation from narrative-driven hype. During ETHDenver, we had the opportunity to connect with 11 of the most prominent AI projects. This article offers a concise overview of their visions, methodologies, and real-world applications, illustrating how these initiatives are reshaping the technological landscape.
Data
The foundation of any AI system is data. How it is sourced, protected, and verified is critical to building robust and fair models.
Grass
The Challenge: High-quality training data is essential for AI, but access is often restricted by gatekeepers. While vast amounts of data exist on the public web, major websites frequently block access to commercial data centers.
The Solution: Grass is a decentralized data provider protocol designed to make AI training data more accessible and promote a fairer AI infrastructure.
How It Works: Users install a Chrome extension to contribute their unused bandwidth and computational resources. This network of devices scans the internet for AI-relevant data. Grass operates a global network of nearly one million nodes, processing over 1 TB of data daily to create structured, clean datasets.
Application: Grass nodes are operational in over 190 countries, creating a truly decentralized and global data source.
Story Protocol
The Challenge: AI-powered remixing of content is both inevitable and often exists in a legal gray area. A significant hurdle for AI development is the lack of a profitable model that ensures proper attribution and revenue flow for original IP creators.
The Solution: Story Protocol is a composable, on-chain IP layer that allows creators to set the rules for how their work is used, enhancing the discoverability and liquidity of global intellectual property.
How It Works: Creators can purchase license NFTs to transform static IP into programmable IP assets. These assets consist of "Nouns" (data structures and metadata) and "Verbs" (functional modules for licensing, derivative creation, and revenue sharing). Revenue from derivative works automatically flows back to the original creator.
Application: The protocol enables customized licensing for derivatives, including specifications for region, distribution channel, duration, revocability, and attribution.
Space and Time
The Challenge: As Large Language Models (LLMs) advance, there is a risk that the corporations behind them could alter training datasets or model parameters. Providing cryptographic proof that the data used for training remains untampered is crucial for trust and verification.
The Solution: Space and Time serves as a verifiable indexer and zero-knowledge (ZK) prover, delivering proofs for SQL queries and vector searches performed on its indexed data.
How It Works: LLM providers can load their training datasets onto Space and Time. The data is witnessed, threshold-signed with a cryptographic commitment, and later used to generate proofs that the exact dataset was used in training. Their GPU-accelerated prover, "Blitzar," can verify a query on a 2-million-row table in about 14 seconds on a single GPU.
Application: Users can make queries in plain text. The platform retrieves context from a verifiable vector database, writes an accurate SQL query for the prover, and returns a proof within seconds.
Model
Creating open, economically incentivized, and verifiable model ecosystems is key to decentralized AI.
Bittensor
The Challenge: The development of AI is becoming increasingly centralized, with a risk of control being monopolized by a few large entities.
TheSolution: Bittensor is a decentralized, open-source platform aimed at creating a competitive and collaborative ecosystem for AI development.
How It Works: The Bittensor network consists of 32 subnets, which began with AI models but have expanded to include storage, computation, and data scraping. The TAO token incentivizes subnet builders to continuously improve their offerings. Validators rank the outputs of these subnets, and this ranking determines TAO distribution, with the lowest-performing subnets being removed—ensuring a constant competition for quality.
Application: A growing ecosystem of applications is being built on Bittensor, including FileTAO for decentralized storage and Cortex TAO for OpenAI-compatible inference.
Sentient
The Challenge: The development of Artificial General Intelligence (AGI) carries existential risks and is often hampered by the profit-driven constraints of capitalism. The crypto space, meanwhile, is in need of a native, transformative application.
The Solution: Sentient is a sovereign, incentive-driven platform for AGI development.
How It Works: Sentient employs a crowdsourced approach, enabling community coordination to contribute to and train models, thereby reducing costs. It uses open protocols to control inference, allows for model composability, and ensures value flows back to network participants.
Application: By aggregating resources from both web2 and web3 and leveraging token incentives, Sentient aims to motivate developers to contribute to the building of trustless AGI.
Infrastructure
Robust, decentralized infrastructure is required to connect and power all components of the AI stack.
Ritual
The Challenge: Centralized AI infrastructure is subject to permissioning, increasing regulation, and potential censorship and manipulation.
The Solution: Ritual is a decentralized network designed to be a convergence point for crypto and AI, comprising an oracle network and a sovereign chain with a custom virtual machine (VM).
How It Works: Its oracle network, "Infernet," allows any smart contract on an EVM chain to connect on-chain workflows to off-chain machine learning (ML) inference. A co-processor enables AI-native operations at the VM level. Nodes on the Ritual network run and service model operations while also participating in consensus.
Application: Use cases include an unpredictable LLM for friend.tech and AI-powered parameterization for DeFi lending protocols.
Agents
Autonomous AI agents represent a frontier where AI can actively perform tasks and manage assets.
Olas
The Challenge: The potential of web2 autonomous agents is limited by an inability to perform KYC, lack of user ownership, platform censorship risks, and poor composability.
The Solution: Olas is a decentralized protocol for collective autonomous agents.
How It Works: In the Olas ecosystem, agents operate off-chain but are registered and managed on-chain. They are organized into "Decentralized Autonomous Services." Each agent runs a finite state machine (FSM) replicated on a temporary blockchain. Agents reach consensus off-chain before executing on-chain actions. The network incentivizes all stakeholders—developers, operators, and capital providers—with its OLAS token.
Application: Olas Predict is an economy of autonomous agents that continuously create and participate in prediction markets for future events.
MyShell
The Challenge: The current creator economy is relatively static. With the rise of LLMs, better tools are needed to help creators easily build dynamic AI applications.
The Solution: MyShell is a decentralized platform for discovering, creating, and staking AI-native applications.
How It Works: Creators can build AI apps on MyShell in minutes, from companion AIs to productivity tools. They can select open-source models, edit prompts, and integrate images and video. The platform's proprietary LLM is trained on extensive private data to enhance role-playing experiences. A native token is used for accessing premium features, rewarding creators, and settling usage fees.
Application: The platform hosts a wide array of AI characters with unique personalities and voices for companionship, learning, and gaming, as well as tools for language learning and content summarization.
Future Primitive
The Challenge: NFTs are native to the digital world but lack inherent programmability, limiting their ability to perform actions on-chain.
The Solution: Future Primitive leverages ERC-6551 to transform NFTs into programmable agents.
How It Works: Using ERC-6551, every NFT is granted the same capabilities as an Ethereum user. It can hold assets, perform actions, and control multiple independent accounts. Each NFT has the same address on any EVM-compatible chain, unlocking cross-chain potential. With Token Bound V4, a smart contract can autonomously execute on-chain actions on behalf of the NFT without requiring the owner's signature for every transaction, though the owner retains the ability to revoke permissions.
Application: This turns static NFTs into active participants in the ecosystem, capable of automating investments, managing DeFi positions, or interacting across various dApps. 👉 Explore more strategies for NFT utility
Frequently Asked Questions
What is the main goal of decentralized AI projects?
Decentralized AI aims to create open, permissionless, and verifiable artificial intelligence systems. The core goals are to prevent monopolistic control, ensure data and model integrity through cryptographic proofs, and create fair economic models that reward contributors and creators.
How do users typically earn rewards in these ecosystems?
Users can earn rewards by contributing resources to the network. For example, they can provide unused bandwidth for data scraping, contribute data for model training, operate nodes that perform computations, or stake tokens to help secure the network and earn incentives.
Why is verifiability important for AI?
Verifiability is crucial for building trust in AI outputs. Techniques like zero-knowledge proofs allow users to cryptographically verify that a specific model was used on a specific dataset without any tampering, ensuring the result is correct and reliable without having to trust a central authority.
What is an AI agent in the context of web3?
An AI agent in web3 is a program that can perceive information from its environment and execute actions on-chain autonomously. These agents can manage assets, interact with smart contracts, and operate based on pre-defined rules or machine learning models, all while being owned and controlled by users rather than a central platform.
How can creators benefit from AI and blockchain integration?
Blockchain and AI integration allows creators to maintain ownership and control over their intellectual property. They can set programmable rules for how their content is used, ensure automatic and transparent royalty payments from derivatives, and leverage new tools to create dynamic, interactive AI applications easily.
What are the biggest challenges facing decentralized AI?
Key challenges include the high computational cost of verifiable computation (like ZK proofs), achieving sufficient scalability to handle large AI models, designing effective token economic models to incentivize all participants, and ensuring the quality and reliability of decentralized services compared to their centralized counterparts. 👉 Get advanced methods for participating in AI networks