Rob Behnke
May 7th, 2024
Permissionless innovation facilitated by peer-to-peer technologies has become one of the greatest disruptors to modern finance. Adding fuel to the fire is the rapid growth of consumer-facing LLM-based AI applications.
As Artificial Intelligence continues to mature, it's finding fertile ground within the world of cryptocurrency and Web3. This powerful combination of two world-changing technologies addresses deep-seated challenges within the crypto space, offering new scalability, privacy, and intelligent decision-making levels.
Traditionally, blockchain technology, the foundation of cryptocurrency, has been limited in its ability to manage the intricate computational demands of modern AI systems. Additionally, concerns around privacy and the potential for centralized control linger. AI crypto coins can disrupt this status quo.
In this article, we explore the rise of a new narrative — cryptocurrencies in AI-supported blockchain projects (aka AI coins).
One of the core challenges in AI development is the need for substantial computational resources. Training sophisticated AI models and running inferences require high-performance hardware, particularly graphics processing units (GPUs). Rising demand for such hardware has led to shortages, creating bottlenecks for AI research and development.
AI crypto projects address this issue through decentralized computing systems that provide on-demand access to specialized hardware.
These systems make use of blockchain-based marketplaces to connect hardware providers with AI developers, ensuring efficient allocation of resources.
AI crypto projects are driving innovation across multiple domains. One notable area is zero-knowledge machine learning (zkML), where AI models are integrated with smart contracts using zk-proofs. This technology enables off-chain computations to be verified on-chain, enhancing blockchain scalability and allowing smart contracts to benefit from AI without compromising efficiency or security.
Another innovative area is AI agents, autonomous entities that execute tasks on behalf of users. These agents utilize crypto wallets to transact on blockchain networks, facilitating seamless interactions with decentralized systems.
Source: https://outlierventures.io/wp-content/uploads/2024/02/AI_thesis_v9-1.pdf
Cryptocurrency and AI share a deeply intertwined, mutually beneficial relationship. Cryptocurrencies, with their inherent properties of decentralization and immutability, provide a foundation for AI development that addresses critical issues, enhancing trust and unlocking innovative applications.
The decentralized nature of cryptocurrencies offers a trustless environment where AI models and algorithms can interact and transact without relying on intermediaries. This creates the space for secure marketplaces for AI data, models, and computational power.
On-chain verification and information provenance tracking can even help mitigate the risks associated with false/mis-information through means like geopolitical propaganda and deep fakes, which is often the concern with AI-generated content.
Blockchain networks generate immense amounts of data—transaction history, network activity, and token interactions. This wealth of data is a goldmine for training AI models. AI algorithms can analyze this data to detect patterns, predict market trends in crypto trading, optimize blockchain consensus mechanisms, and even enhance cybersecurity within the crypto ecosystem.
Centralized control over AI development raises concerns about censorship and bias, as a few technology giants dominate the field. AI crypto projects offer a censorship-resistant alternative, decentralizing the creation and dissemination of AI models.
Machine learning algorithms can analyze blockchain networks for unusual patterns or behaviors, helping to detect and prevent malicious activities. This integration enhances the overall security of blockchain systems, making them more robust and resilient against attacks.
At the time of writing, the cumulative market capitalization of all AI tokens is over $27 billion. Below are four projects (and their associated AI coins) that are leading the AI-crypto narrative.
Fetch.ai (FET) is a blockchain project that combines artificial intelligence (AI) and machine learning (ML) to create a decentralized digital economy.
The core idea is to use autonomous software agents, also called Autonomous Economic Agents (AEAs), to perform tasks and interact within the network. You can think of Fetch.ai as a giant decentralized AI marketplace for super-smart software helpers.
These AI agents can be programmed to execute various tasks autonomously, from booking you the best flight deal to managing your energy use at home. They can even work together to solve complex problems, like optimizing a supply chain or creating a more efficient financial market.
The creators of the project envision a future where machines can do more than merely follow instructions — learn and act independently on your behalf.
AEAs are digital workhorses powered by AI. They are independent software entities that can act on behalf of users, devices, or services.
What AEAs can do:
Search and Discover: Find relevant resources, services, or data across the Fetch.ai network.
Negotiate: Conduct exchanges or interactions with other AEAs to find the best deals or execute tasks.
Transact: Use Fetch.ai's native token (FET) for payment or value transfer within the network
Adapt and Learn: Improve their decision-making based on past experiences and interactions.
The Open Economic Framework (OEF) provides the environment where AEAs connect, discover each other, and exchange information. It acts as a decentralized directory for agents and the services they offer or represent.
The OEF uses smart contracts to facilitate agreements and interactions among AEAs, ensuring secure and reliable transactions.
This innovative blockchain-like structure combines a distributed ledger with a directed acyclic graph (DAG). This hybrid approach allows for both efficient scaling and the ability to handle complex interactions between a large number of AEAs.
Fetch.ai uses a modified Proof of Stake consensus mechanism where nodes stake FET tokens to participate in validating transactions and maintaining the network.
Bittensor (TAO) is a decentralized protocol designed to create a market for machine learning models. It facilitates the sharing, training, and evaluation of these models in a peer-to-peer network, incentivizing collaboration and the development of more effective artificial intelligence (AI) systems.
Unlike traditional AI development, which can be siloed and resource-intensive, Bittensor promotes collaboration. With Bittensor, developers can:
Contribute AI Models: Share their pre-trained AI models on the network.
Access Shared Models: Leverage existing models from the network as a foundation for their own projects.
Train Models Collectively: Multiple participants can collaboratively train and improve AI models within a subnet.
The network consists of two key participant types:
Validators: They assess the performance and quality of machine learning models.
Servers: They provide the computational resources necessary for training and running the models.
Bittensor operates on its own blockchain, separate from mainstream blockchains like Ethereum. This blockchain facilitates the creation and interaction of specialized sub-networks called subnets.
Think of subnets as niche communities or specialized marketplaces within the larger Bittensor network. Each subnet is dedicated to a particular domain of machine learning or AI. Examples could include:
Text-based AI models specializing in language translation or text summarization.
Image processing subnets focused on tasks like object recognition or image generation.
Subnets dedicated to predictive analytics or financial modeling.
The community can propose the creation of new subnets dedicated to specific AI needs. Each subnet features distinct incentive mechanisms. This ensures that rewards align with the value of contributions specific to that domain.
Instead of using power-hungry Proof of Work (like Bitcoin) or relying solely on staking (like Ethereum), Proof of Intelligence (PoI) prioritizes nodes that actively contribute to improving the AI models within the network.
Incentivization through PoI works as follows:
Nodes submit their work (e.g., translations, image classifications) to be evaluated by other nodes within the subnet.
Nodes that provide the highest quality and most valuable contributions receive higher reputation scores.
Reputation scores directly impact the amount of TAO tokens a node receives as rewards.
Such a system encourages continuous improvement of Artificial Intelligence models within the Bittensor network, rewarding those who focus on the quality of their contributions.
Public blockchains, like Ethereum, are herculean — they contain an ocean of data spread across several nodes.
Consequently, retrieving specific information from these scattered data sets is time-consuming and resource-intensive. This inefficiency poses a significant problem for developers building decentralized applications (dApps) that require real-time access to blockchain data.
The Graph (GRT) addresses these challenges by providing a decentralized protocol for indexing and querying blockchain data. This protocol offers an efficient, scalable, and flexible solution for extracting the necessary information from blockchain networks.
The protocol is held up by 6 critical pillars:
GraphQL
Subgraphs
Indexers
Curators
Delegators
Consumers
Using GraphQL, a query language developed by Meta, The Graph allows developers to specify the exact data they need, enhancing the efficiency of data retrieval. This capability is particularly important for dApps, NFTs, and other blockchain-based services that rely on real-time data for their operations.
The data in The Graph is organized into "subgraphs." A subgraph defines the specific data to be indexed from a blockchain and how that data should be stored.
Developers create subgraphs by defining the structure and relationships of data needed for their applications.
Indexers are node operators that index the data from subgraphs and serve the data to the network. They earn rewards and fees in GRT (The Graph's native token) for their services.
Curators signal which subgraphs are valuable or worth indexing by staking GRT on them. They earn a portion of the query fees from the subgraphs they curate, creating an incentive to identify useful data.
Delegators are individuals who do not want to run indexer nodes but wish to participate in the network. They can delegate their GRT to indexers and earn a portion of the rewards.
Consumers are end-users or applications that query subgraphs to retrieve data. They pay query fees, which are distributed among indexers, curators, and delegators.
Akash Network (AKT) takes aim at the cloud computing giants like Amazon Web Services (AWS) and Microsoft Azure, but with a twist: it's decentralized. It is a decentralized cloud computing marketplace that aims to provide a more efficient, transparent, and cost-effective alternative to traditional cloud services.
“Providers” and “tenants” are the two integral pieces of the Akash Network:
Providers: Anyone with spare computing power on their servers or computers can list their unused resources on the Akash Network marketplace. This could be anything from a home computer to a massive data center.
Tenants: Businesses or individuals needing computing power can rent these resources on-demand, similar to how you might rent a car on a ride-sharing app.
Akash operates on a proof-of-stake blockchain, which provides a secure and scalable infrastructure for the network. This blockchain underpins the marketplace and facilitates secure transactions and smart contracts.
The core service offered by the network is Akash DeCloud, which provides decentralized and secure cloud computing. This service enables developers to deploy and manage containerized applications across various cloud platforms, similar to Kubernetes.
Deployment and lease management are crucial aspects that facilitate the renting of cloud computing resources between providers and tenants.
A deployment in the Akash Network represents a tenant's request for cloud computing resources. It contains the specifications for the application or workload the tenant wants to deploy.
Here's how the deployment process works:
Manifest: The tenant prepares a manifest (a YAML file) to define the application’s requirements. This includes details like the container images, resource requirements (CPU, memory, storage), and other configuration parameters.
Order: The tenant submits the deployment manifest to the Akash marketplace. An order is created for the deployment, which is essentially an open call for bids from providers who can meet the specified requirements.
Bid: Providers review open orders and place bids to fulfill the deployment. Each bid outlines the provider’s terms, including the cost and the specific resources they offer.
Lease: The tenant reviews the bids and selects one that best meets their needs. This selection finalizes the lease, which is a contractual agreement between the tenant and the provider. The lease binds the provider to supply the specified resources at the agreed-upon terms.
A lease in the Akash Network is the contractual agreement between a tenant and a provider of cloud computing resources. Lease management involves overseeing and maintaining these leases throughout their lifecycle.
Here’s how lease management works:
Deployment: Once a lease is established, the provider deploys the tenant’s application using the details provided in the manifest. Tenants can control and manage deployment via Akash CLI or other interfaces.
Payment: The tenant pays for the resources according to the lease terms, using the network's native AKT token. Payments are typically made periodically, based on the duration and usage specified in the lease.
Monitoring: Both the tenant and the provider can monitor the deployment to ensure it is functioning as expected. The Akash Network provides monitoring tools to track performance, uptime, and resource utilization.
Lease Termination: Either party can terminate the lease when it is no longer needed or if issues arise. The lease can be terminated by mutual agreement or if one party breaches the terms. Upon termination, the provider deallocates the resources, and any remaining prepaid funds are refunded to the tenant.
Dispute Resolution: The Akash Network has mechanisms for resolving disputes between tenants and providers. These mechanisms include arbitration processes facilitated by the network's governance structure.
From a security standpoint, the intervention of Artificial Intelligence (AI) in Web3 is a double-edged sword. Of course, the security flaws in AI-supported crypto ecosystems can be addressed and eventually patched — until then, concerns loom.
Phishing scams are rampant in crypto. According to a report by Chainalysis, nearly $400 million was lost to phishing scams in 2023 alone. With AI, there are even more ways for scammers to target victims.
AI-powered bots and systems can craft highly personalized and targeted phishing attempts, making it even harder for users to spot scams.
Malicious actors can even use AI to rapidly scan smart contracts and blockchain code for vulnerabilities – finding flaws faster than manual processes. For instance, GPT-4 can successfully exploit 87% of vulnerabilities when provided with specific CVE (Common Vulnerabilities and Exposures) information, showcasing its potential to autonomously identify and leverage these security flaws.
At the same time, AI can be an incredible boon for ensuring the air-gapped security of DLT systems and smart contracts.
In a 2023 data-driven Halborn report titled Can ChatGPT Detect Smart Contract Vulnerabilities, we highlighted how ChatGPT-4 showed a high detection rate for well-known vulnerabilities, with a success rate of up to 86.6% when prompted specifically. This indicates that AI can significantly enhance Web3 security by quickly identifying and mitigating common vulnerabilities.
While AI has shown great promise in enhancing Web3 security, it also has notable limitations. The same report highlights that AI struggles with complex or rare vulnerabilities, often requiring multiple attempts to detect such issues correctly. AI’s effectiveness also diminishes when dealing with complex Capture the Flag (CTF) challenges, where multiple vulnerabilities interact in intricate ways.
Therefore, the optimal path forward for Web3 security lies in a hybrid approach that combines AI’s capabilities with human expertise.
For more insights on our ChatGPT and smart contract vulnerabilities report, watch this video interview with cybersecurity researcher John Hammond and Ferran Celades, the Principal Security Architect at Halborn.