Intelligent contracts: Making smart contracts smart for blockchain intelligence

With the deep integration of blockchain and Artificial Intelligence (AI), more and more blockchain-based AI tasks are accomplished using Smart Contracts (SCs) and create win-win solutions. That is, blockchain provides a trustworthy and decentralized data infrastructure for AI, and AI helps blockchain perform tasks requiring intelligence. Since these special SCs designed for blockchain-based AI tasks have different characteristics from the widely studied SCs designed for business logic, we name them Intelligent Contracts (ICs) for a focused study. In this paper, we systematically analyze ICs and propose a constructive framework for their construction and application. Specifically, we first formulate two construction modes of current ICs, including encoding AI models and scheduling AI collaboration. Then, we compare the characteristics of these two modes theoretically and experimentally as a reference for future mode selection. Finally, to extend the application of ICs and encourage AI-driven blockchain intelligence, we propose a technical route that helps blockchain autonomously respond to AI tasks through the dynamic and optimal configuration of ICs. Using typical AI tasks of classifying IRIS, MNIST, and ImageNet data sets as examples, we implement and thoroughly evaluate two modes of ICs on Ethereum. Based on the constructed ICs, we illustrate their optimal configuration and automatic response process. Experimental results demonstrate the effectiveness and feasibility of the proposed framework.