A Primer on AI Agents and the Blockchain

Introducing the capacity for anyone (as long as you’re a subscriber) to create GPTs on ChatGPT is a fascinating democratization of the ability to build AI agents. For example, with the Dalle E GPT anyone can create and edit images via text-based prompts. Outside of ChatGPT, there are (in)famous examples like BabyAGI, a digital project manager, and Auto-GPT, which breaks down prompts into sub-tasks and automatically executes them. These GPTs can be built using a conversational interface, which means a creator doesn’t even need coding knowledge to build one. And they can monetize it too.

The introduction of autonomous AI agents to various blockchain-based services and protocols was one of the highlights of 2023 and is also shaping up to be an industry-defining trend in 2024. In this article, we present an accessible introduction to the world of AI agents and how the blockchain industry is shaping up around the promises of autonomous smart agents.

An easy introduction to AI agents

Artificial Intelligence, as a field in computer science, focuses on creating intelligent algorithmic machines capable of mimicking human cognitive abilities such as learning, reasoning, problem-solving, and decision-making. Today, the most common form of AI application is Generative AI, which, given a natural language prompt, enables the creative generation of new content such as text and images as well as video and audio.

However, within the broader realm of AI has emerged a class of AI agents, which are simply intelligent systems that can autonomously perform specific tasks; they are not general-purpose machines. They can range from simple customer service chatbots to complex robots performing surgeries in the hospital. Typically, once prompted, AI agents are expected to autonomously go through the steps required to accomplish a goal.

The most familiar examples of AI agents are GPTs launched by OpenAI last year. ChatGPT is a general-purpose chatbot based on a large language model, but in November 2023, OpenAI added the capability to create custom versions of ChatGPT to accomplish specific purposes. For instance, a research GPT can, given a simple prompt, search millions of academic papers from a third-party database and draft content with accurate citations to real papers. That’s what the Consensus GPT does.

Similarly, the Canva GPT can generate entire designs of logos, social media posts, and presentations from a single prompt autonomously. And it is quite easy for anyone, even without programming knowledge, to create a GPT through the GPT Builder feature, which itself is an AI agent for creating a GPT through conversational prompts. This custom GPT feature has been hailed as the next generation of AI agents with potential applications ranging across various use cases.

AI agents on the blockchain

Increasingly, AI agents are finding a home in the world of blockchain and are rapidly gaining ascendance in the industry. Blockchain analytics platform, Nansen’s primary high conviction for 2024 is that AI agents will become primary users on the blockchain as artificial intelligence plays a more central role in the crypto ecosystem.

AI systems learn through reinforcement and innovators have unlocked how to use token-based incentives to reward AI agents in order to enforce expected behavior and ensure more effective performance.

In a recent blog post exploring the intersections between AI and crypto, Ethereum founder Vitalik Buterin identifies four levels of viability when categorizing blockchain applications of AI algorithms:

• AI as a player in a game: this is the most viable level. For example, AI agents can act as traders, miners, validators, or arbitrageurs in decentralized markets, platforms, or networks, where they can earn rewards or fees for providing services or contributing resources. One benefit of this is that the agents can leverage their superior analytical and computational abilities to optimize strategies and outcomes in complex and dynamic environments.
• AI as the interface to the game: this level has a high potential but comes with risks. AI agents in this category could provide personalized recommendations, alerts, and guidance to users, enhancing their understanding of blockchain technology and enabling them to make more informed decisions. They could also help users verify the authenticity of transactions and smart contracts
• AI as the rules of the game: This level of viability refers to where AI agents act as regulators or enforcers of the blockchain systems, directly influencing or determining the rules, policies, or outcomes of the protocols. This can enhance the quality, consistency, and objectivity of the blockchain systems for more reliable solutions.
• AI as the objective of the game: finally, blockchain technology and decentralized mechanisms could be leveraged to construct, maintain, and incentivize the development of advanced AI systems in a decentralized and open environment. Buterin describes this as a longer-term goal.

The opportunities of autonomous AI agents

Autonomous AI agents hold a lot of promise for blockchain stakeholders that they cannot simply be ignored. Here are some of the opportunities already being exploited and which are sure to be developed more as time passes.

1. Crypto trading bots: these bots compensate for human limitations in seizing market opportunities by automatically executing trading transactions on behalf of their owners. They can work round-the-clock and are connected to exchanges using an API to make trades based on predefined market parameters. Many of these bots can operate through apps such as Telegram or be cloud-hosted. If you don’t trust an AI bot enough to do this yet, services like Flagship Vaults still have human analysts and experts watch traders’ positions and manage their portfolios.
2. Market Prediction and Analysis: AI agents can analyze vast amounts of data, including market trends, news sentiment, and social media chatter, to help users make informed investment decisions.
3. Fraud Detection and Security: many banking and financial institutions have already enabled AI-powered fraud detection systems. Blockchain technology is reputed for its high level of security, enabled via cryptography, but it is not entirely immune to cyber attacks; this is where AI algorithms can fill the gaps since they are excellent at identifying patterns and anomalies by analyzing transaction data in real-time.
4. Algorithmic Trading: Crypto trading bots compensate for human limitations in seizing market opportunities by automatically executing trading transactions on behalf of their owners. They can work round-the-clock and are connected to exchanges using an API to make trades based on predefined market parameters. Many of these bots can operate through social apps such as Telegram or be cloud-hosted
5. AI-powered Agents for Transactions: for instance, MEV arbitrage bots are designed to exploit the price differences between the same asset on different decentralized exchanges (DEX) in order to execute profitable trades. They can also manipulate the order of transactions in a block for the same profitability reasons. Besides arbitrage, other strategies they employ include frontrunning, sandwiching, flash loans, etc.
6. Portfolio Management and Optimisation: AI agents can analyze a trader’s investment goals, risk tolerance, and market conditions to recommend an optimal portfolio allocation and rebalance your holdings as needed. Via market insights from Flagship, you can easily gain reliable information on market trends to inform your trading positions.
7. Trust and Reliability: blockchain technology, through smart cryptography, can help securely verify authentic AI agents and AI models through zero-knowledge machine learning (zkML), digital signatures, and IPFS & Merkle Trees. SingularityNET is a decentralized marketplace that uses smart contracts to authenticate AI agents. The blockchain can also help verify the data sources of AI models for quality and reliability.
8. Web3 Experiences: AI agents can also be useful in building web3 experiences. For instance, the Parallel Colony game was built using AI-powered avatars, which are NFTs with their intelligence generated using a fine-tuned large language model. They aren’t mere NPCs; they operate autonomously, have their own biographies, and can even trade the native token $PRIME in the game.

Projects working on blockchain-based AI agents

• Fetch AI

Fetch.ai describes itself as the first open network for AI agents, enabling businesses to build and monetize AI apps and services through its AI Engine, Agent Services, and other features. Its platform for creating, testing, and deploying customizable AI agents is dubbed Agentverse and it allows businesses to securely host their AI agents on a decentralized platform. And with pre-built templates, businesses can set up their agents in a matter of minutes. Its token (FET) recently surpassed the $1 threshold, signifying the heightened interest of investors and enthusiasts in this project.

• SingularityNET

SingularityNET is on a mission to build a decentralized Artificial General Intelligence (AGI). Presently, it is a decentralized, blockchain-based AI marketplace, allowing users to integrate AI services from the community of providers on its platform. It maintains a partnership with Cardano via its native AGIX token

• Autonolas

Also known as the Olas network, or simply Olas, Autonolas is trying to be a single network for all the building blocks of crypto and it is built on autonomous agent technology. Users also have the opportunity to find exciting agents on its platform and even stake on these agents to earn rewards. Presently, the Olas Protocol is available on Ethereum, Polygon, Solana, and more.

• Balance AI

Balance AI allows AI enthusiasts and users to collaborate on solving problems. Like FetchAI, it is a marketplace for driving innovation through AI models by helping creators share and monetize their efforts. Its current focus is on enhancing the quality and credibility of AI models admitted to its network.

• AIgentX

AIgentX offers various AI-based solutions for trading analysis, business support, community management, and so on, with the aim of helping businesses streamline their processes via AI dApps on its platform. These are AIgentXT, AIgentXS, and AIgentXC. As its suite expands, it has a huge potential to redefine enterprise workspaces in the future.

Not there yet; the challenges and risks of AI agents

Despite the appeal of AI agents and the huge promises they hold, there are obstacles to consider. For one, AI agents are susceptible to adversarial attacks. And worse, adversarial attacks that affect one model can affect another model that was trained on a different sort and set of data provided they were created to accomplish similar purposes. Malicious actors could easily exploit such vulnerabilities in AI algorithms to manipulate their decision-making, leading to unintended transactions and disruptions in the network.

In addition, while many people consider only the positive side, there is also a negative side to the adoption of AI agents on the blockchain: they could be used to create even more sophisticated and convincing scams. This makes it harder for human users to identify and avoid them. Similar to existing blockchain challenges, bugs and vulnerabilities in smart contracts could be exploited by AI agents, potentially leading to hacks and financial losses.

Besides, there are data privacy concerns, algorithmic bias, lack of transparency (black-box algorithms), and an uncertain legal landscape to deal with. Some of these challenges are ongoing concerns and it remains to be seen how adopters of AI agents in the blockchain industry will step up to the challenge. Particularly, an issue to solve is how the black-box nature of sophisticated AI models and agents contradicts the openness and decentralized model of the blockchain. Perhaps both technologies complement each other better in this way.

Closing thoughts

Technically, AI agents are not a new concept when it comes to blockchain and crypto. However, what we are witnessing now is a democratization of access like never before, especially to creators.

On the one hand, it portends the proliferation of low-quality AI agents in decentralized marketplaces. On the other, it holds a lot of promise for mutually beneficial innovations in the confluence of the two fields: AI and blockchain technology. This remains an exciting field that is sure to inform the direction of stakeholders in 2024.