AI and DeFi Integration: Beyond AlphaGo's On-Chain Gaming （I）

Introduction: When AI Becomes the "New Player" in On-Chain Gaming

In 2025, the Total Value Locked (TVL) in global DeFi protocols exceeded $100 billion, yet issues like market inefficiency and fragmented liquidity remain unresolved. Meanwhile, the explosive growth of AI technologies (such as GPT-4, Llama 2) has introduced a new variable to DeFi—AI agents. These on-chain "robots" can not only execute preset strategies but also dynamically optimize decisions through machine learning and even engage in gaming with other AI systems. However, this technological revolution is accompanied by sharp controversies: Will AI push DeFi toward the abyss of zero-sum gaming? How can fairness be guaranteed in on-chain markets? 

In the first part, we will take a deep look at the iconic case and reveal the technical path of the integration of AI and DeFi, and in the second part, we will focus on the economic impact and potential risks.

AI Reshaping DeFi's Underlying Logic

1.1 DeFi's Evolutionary Needs

Since the integration of blockchain technology with decentralized finance (DeFi), the DeFi ecosystem has experienced rapid innovation and development. In the early stages, DeFi's core principle was "code is law," meaning all protocols and transaction behaviors were controlled by smart contracts, reducing dependency on intermediaries. However, as market scale expanded and transaction volumes surged, DeFi's needs have changed, with more complex financial operations and strategic requirements emerging, pushing DeFi toward more refined and intelligent directions.

High-frequency trading, liquidity management, and arbitrage strategies—very common in traditional financial markets—have gradually penetrated the DeFi realm. With increasing market volatility and complexity, traditional rules and static algorithms can no longer meet demands. At this point, the introduction of AI becomes key to solving this problem. Through deep learning and optimization algorithms, AI can analyze vast amounts of market data in real-time to formulate more precise trading strategies and liquidity management solutions. Additionally, AI can help further lower the barriers to using DeFi projects. Conducting a series of on-chain interactions through AI agents using natural language commands can significantly reduce user learning costs.

1.2 Synergistic Effects of AI and DeFi

With the continuous advancement of AI technology, the DeFi field is gradually shifting from purely code-driven to intelligent, data-driven decision models. AI not only helps improve strategy execution efficiency but also discovers potential opportunities and risks in the market through deep data analysis, thereby enhancing the intelligence level of the entire DeFi system.

1.2.1 Information Advantage: Integration of On-Chain and Off-Chain Data

In traditional DeFi protocols, market data primarily comes from on-chain information, such as blockchain transaction data, asset pool balances, and liquidity pool statuses. However, the introduction of AI enables DeFi to organically combine off-chain data with on-chain data, further enhancing predictive capabilities. For example, by introducing oracles to provide off-chain price data and analyzing this data using machine learning technology, AI can more accurately predict market price trends, thereby helping to formulate efficient trading strategies. This advantage of information integration makes AI an important technological tool in DeFi protocols.

1.2.2 Gaming Upgrade: Automation of MEV

In decentralized financial systems, Miner Extractable Value (MEV) has consistently been a hot topic of concern. MEV refers to the additional revenue that miners can obtain by adjusting the order of transactions in a block. In the past, MEV extraction primarily relied on manual strategies and operations, which were not only inefficient but also easily affected by human factors.

However, with the development of AI technology, MEV extraction methods have evolved from traditional manual sniping to automated intelligent operations. AI can analyze on-chain data in milliseconds and optimize transaction ordering through algorithms to maximize returns. This transformation not only improves efficiency but also drives DeFi protocols toward more efficient and fair directions. With AI automation, the rules of the game in DeFi markets are being redefined.

1.3 Clear Definition of DeFAI

DeFAI (DeFi + AI) is a product of deep integration between DeFi and AI, aimed at enhancing the intelligence, user experience, and capital efficiency of the DeFi ecosystem. Its core is to achieve automated decision-making, complex task execution (such as trading, yield optimization, cross-chain operations), and abstraction of user interactions (like natural language instructions) through AI agents, thereby lowering the barriers to DeFi usage and optimizing financial strategies.

Core Application Scenarios of AI in DeFi

2.1 AI-Driven Operation Execution

Through the aggregation of various AI models or agents and the integration of APIs from various DeFi applications, users can quickly execute DeFi operations through a ChatGPT-like chat interface. This DeFAI combination enhances the on-chain user experience, providing a one-stop DeFi operation interface. Users no longer need to navigate between different UIs to execute DeFi operations, nor do they need to search and learn how to combine various protocols to achieve their desired results, such as swapping 100 SOL to USDC and then depositing it into a lending protocol. Users only need to convey their intentions through chat in the Chat interface.

Case: Bankr

Bankr is an AI-driven platform that leverages advanced technology and integrates with third-party wallet provider Privy to securely create digital wallets and facilitate blockchain transactions, including trading and transferring digital assets. (Demo video: https://x.com/bankrbot/status/1893788678776070481)

Case: Griffain

A DeFAI (DeFi + AI) project based on the Solana blockchain, its core positioning is as an AI agent engine, aiming to simplify DeFi operations through natural language interaction, achieving seamless connection between user needs and the blockchain ecosystem. Users can interact directly with DeFi protocols through human-like language instructions (such as "buy 200 USDC worth of SOL" or "cross-chain transfer and stake") without having to manually operate smart contracts or understand the underlying technical details.

Griffain deploys multiple specialized Agents internally, each focusing on specific tasks:

Airdrop management agent: Automatically tracks and claims eligible airdrops;

Yield optimization agent: Analyzes on-chain data and dynamically adjusts asset allocation to maximize returns;

Portfolio management agent: Monitors asset risks and provides rebalancing suggestions;

Token research: Generates reports based on on-chain information.

2.2 Credit Assessment and Risk Management

Credit assessment and risk management in DeFi are key to ensuring protocol stability and user fund security. Compared to traditional financial systems, DeFi's credit system is still imperfect, with user credit assessment primarily relying on collateral. However, the introduction of AI technology provides more innovative opportunities for DeFi protocols, especially in credit assessment and risk control.

Case: Spectral Finance's AI Credit Scoring Model

Spectral Finance first introduced the Multi-Asset Credit Risk Oracle (MACRO) score in 2022, combining users' on-chain behavior with credit history data to provide more intelligent credit risk analysis for lending markets.

Core Factors of the MACRO Score Model

Spectral Finance's MACRO score adopts multi-dimensional data, including lending, repayment, liquidation history, and even market volatility factors. These data are quantitatively processed through machine learning models to provide users with precise credit scores. Users' credit scores are based on several major factors:

Lending, repayment, and deposit history: User transaction behavior on DeFi protocols, especially the frequency and reliability of lending and repayment.

Liquidation history: The impact of liquidation records from other wallet addresses in the user's wallet on credit assessment.

Health factors: Such as loan-to-value ratio (LTV) and assessment of other market risk factors.

Wallet history length and transaction trends: Wallet usage time and transaction patterns also affect credit scores.

Market volatility and interaction with lending protocols: The frequency of interaction with different DeFi protocols and market volatility also play important roles in credit assessment.

NFC (Non-Fungible Credit) Tokens and Credit Management

To better reflect and utilize users' credit in the DeFi ecosystem, Spectral Finance introduced Non-Fungible Credit (NFC) tokens. NFC tokens are ERC721 tokens representing users' on-chain transaction history. Each NFC token carries the user's MACRO score and can recalculate the score as new loans are initiated.

Currently, Spectral has shifted towards developing an AI agent framework in their whitepaper, although their AI credit assessment model remains a good exploration.

2.3 Automated Strategies and Yield Optimization

In the DeFi market, automated strategies and yield optimization are key to improving user returns and reducing investment risk. With the continuous advancement of AI technology and machine learning, multiple innovative protocols have begun to explore how to use these technologies to automate investment decisions and asset allocation to achieve more efficient yield optimization. ARMA and Amplifi Lending Agents are two typical examples that help users achieve better investment returns in the DeFi market through automation and intelligence.

Case: Asset Management Protocol ARMA

ARMA is an automated yield management protocol developed by Giza Tech focusing on USDC assets. ARMA continuously monitors and evaluates opportunities across protocols like AAVE, Morpho, Compound, and Moonwell, automatically reallocating funds to the highest-yielding pools. Its core functions include automatic asset reallocation, yield optimization, transaction cost minimization, and maintaining strong security and complete user control.

ARMA's yield optimization strategy operates through several aspects:

Cross-protocol automatic rebalancing: ARMA analyzes real-time lending rates across multiple lending protocols, evaluates APR, protocol rewards, ecosystem rewards, and transaction costs, then automatically transfers funds to the most favorable pools. This dynamic adjustment ensures users can maximize returns.

Smart compounding mechanism: ARMA automatically claims and reinvests accumulated earnings, exchanging all reward tokens back into the original stablecoin, maximizing the compound effect. Compounding frequency is dynamically optimized based on position size, APR, and transaction costs, increasing yields.

Token exchange logic: ARMA uses USDC and USDT to access a wide range of yield opportunities. When the yield difference between different tokens is significant, ARMA executes token swaps to optimize returns. All operations are conducted through integrated DEX protocols, ensuring users can withdraw their original deposit tokens at any time.

Case: Investment Strategy Amplifi Lending Agents

Amplifi Lending Agents is an automated lending strategy optimization tool developed by Amplifi.Fi based on AI algorithms. The protocol optimizes users' lending behavior and yield generation strategies in the DeFi market through deep learning and data analysis. Its core goal is to help users achieve the best investment returns in complex market environments through intelligent asset allocation, risk management, and yield optimization.

source：https://amplifi-2.gitbook.io/documentation 

Amplifi Lending Agents' core functions:

Dynamic asset management (optimization and rebalancing): Amplifi's AI/ML engine continuously optimizes asset allocation based on proprietary institutional-grade strategies. It automatically reallocates assets to the highest-yielding pools and ensures effective management of Bitcoin and stablecoin strategies.

Cross-chain functionality: Through Polyhedras zkBridge, Amplifi provides seamless cross-chain liquidity, with users not having to manually bridge or pay gas fees. zkBridge ensures efficient and secure cross-chain operations, simplifying the user experience while reducing risk.

Account abstraction: Amplifi supports gas-free (sponsor-paid) transactions and Web2 login (such as Facebook, Gmail, etc.), making it easier for users to create and manage wallets. In this way, Amplifi simplifies DeFi operations, allowing users unfamiliar with blockchain to participate easily.

Amplifi's intelligent strategies: Amplifi's AI/ML engine combines multiple optimization algorithms, including evolutionary algorithms, gradient descent, and Markowitz portfolio theory. These strategies balance risk and return through weighted methods and continuously optimize asset allocation to ensure investors achieve the best returns under different market conditions.

Optimization strategy: Amplifi continuously evaluates market changes and dynamically adjusts asset allocation through comprehensive application of machine learning techniques to maintain stable annual returns (APY). When APY fluctuations exceed reasonable thresholds, the system automatically triggers reallocation strategies to ensure the portfolio is always in an optimal state.

Asset reallocation: By monitoring APY changes on DeFi platforms (such as AAVE), Amplifi can automatically reallocate assets when APY fluctuates significantly to maximize user returns.

2.4 Market Analysis and Sentiment Tracking

AI's market analysis and sentiment tracking in DeFi focus on data-driven decision support, with core capabilities in real-time integration of multi-dimensional information and quantification of market risk. AI systems can simultaneously parse on-chain transaction behaviors (such as large transfers, liquidity pool changes), off-chain social media sentiment (such as Twitter sentiment), and macro market indicators (such as BTC price fluctuations), predicting short-term asset volatility trends through machine learning models and identifying potential risk signals (such as contract vulnerability warnings or liquidity depletion risks). At the same time, AI uses natural language processing technology to score news events emotionally, dynamically generating asset allocation suggestions or triggering automatic hedging strategies, helping users reduce information asymmetry and improve decision efficiency in complex market environments.

Case: AIXBT

A specialized AI agent tool focusing on the cryptocurrency field, integrating real-time data analysis and natural language processing technology to provide users with market trend insights and trading decision support. It can track and parse hot topics, market sentiment, and key indicators on Crypto Twitter in real-time, helping investors quickly capture industry dynamics and identify potential investment opportunities. AIXBT's core functions include multi-platform data aggregation, social sentiment analysis, and automated strategy generation, particularly excelling at analyzing tweets from over 400 KOLs, distilling high-value information, and generating executable trading suggestions.

Additionally, AIXBT has interactive capabilities, allowing users to communicate directly with it through social media to obtain customized analytical reports. In the future, as AI agent technology matures, AIXBT is expected to further lower the barriers to participation in the crypto market, driving DeFi toward more intelligent and automated directions.

Technical Architecture and Competitive Landscape

3.1 Pathways for On-Chain AI Models

Currently, the most mainstream potential development direction is on-chain lightweight implementation. This involves reducing on-chain resource consumption by computing off-chain and pushing results and computational proofs. One approach is zkML.

zkML verifies the inference process of machine learning models through zero-knowledge proofs (such as zkSNARKs), ensuring the correctness of computational results without disclosing model parameters or input data. Its core process is:

Off-chain computation: Run the ML model off-chain to generate inference results.

Generate proof: Use zk technology to generate a concise mathematical proof indicating that the inference process conforms to the predefined model (such as model weights, architecture) and has not been tampered with.

On-chain verification: Submit the proof to the blockchain, where smart contracts verify the legitimacy of the results by verifying the proof, without needing to re-execute the computation.

This mechanism solves the high-cost problem of running ML models directly on-chain (such as EVM's computational limitations) while ensuring transparency and censorship resistance.

Related use cases already exist, such as Modulus Labs' ZKML framework.

Modulus Labs' core technology is off-chain computation + on-chain verification, with framework features including:

Verifying AI execution integrity: Ensuring AI model inference results are not tampered with through zero-knowledge proofs (such as ensuring the authenticity of NFT price evaluation model outputs). Case: Collaborating with Upshot to put AI-generated NFT price evaluation results (100 million per hour) on-chain through ZK proofs, achieving the combination of "AI capability + on-chain security."

Anti-cheating mechanism: In gaming scenarios (such as AI Arena), verifying the consistency between player-trained AI models and the models actually deployed in competitions, preventing cheating behaviors driven by economic interests.

Open API: Providing "tamper-proof AI" APIs allowing dApp developers to integrate verifiable AI functionality without sacrificing decentralized security.

3.2 Comparison of Leading Projects

Beyond the infrastructure layer, competition in the DeFAI application layer is intense. Overall, product barriers aren't very high, and no application protocol has been found to have product stability and accuracy far superior to other competitors. Meanwhile, the product-market fit (PMF) of existing applications is being tested by users. In today's rapidly developing AI infrastructure, more effective DeFAI applications may emerge at any time, replacing existing leading projects.

Currently, the leading project is Griffain. Their product allows users to interact with the blockchain through natural language and then perform on-chain DeFi operations. Griffain's product update speed is impressive. They currently have products such as Agent Sniper, Agentic Commerce, multi-agent API aggregation (Energy system), etc., already launched, with the overall building direction being an integrated natural language operation platform, allowing users to browse and operate various daily on-chain information through the bot chat window.

As AI continues to reshape the DeFi landscape through advanced technical architectures and innovative applications, it is crucial to recognize the potential risks and ethical challenges that accompany this transformation. In the next part of our analysis, we will delve into the darker side of AI-driven DeFi, exploring the systemic risks, regulatory hurdles, and ethical dilemmas that could define the future of this emerging ecosystem.