Intelligent Finance: Highlights from the 2017 China AI Conference

The third China AI Conference (CCAI 2017) was held in Hangzhou on July 22‑23, gathering more than 40 top AI experts for nine keynote talks and four thematic forums. One of the most acclaimed sessions was the "Intelligent Finance" forum, co‑organized by Ant Group and the AI Society of China.

Li Xiaolong: Ant Group’s Intelligent Finance Practice

Li Xiaolong, Technical Director of Ant Group’s AI Department, explained how AI is integrated into financial services, using the intelligent customer assistant as a concrete example. He highlighted three key advantages of AI in finance:

Rich application scenarios: beyond payments, AI supports wealth management, loans, credit, insurance, security, and more.

Abundant, highly digitized data: financial services generate massive, high‑quality data that fuels AI models.

Social impact: AI lowers the barrier to financial services, enabling inclusive, affordable finance for a broader population.

Ant Group’s mission is to provide inclusive financial services to 2 billion people worldwide, built on credit, AI‑driven technology, and a vision of universal financial access.

Ant Group’s AI Business Architecture

The AI platform is organized into three layers:

Algorithm layer : deep learning, reinforcement learning, graph learning, covering text, speech, image, video, etc.

Tool layer : internal platforms for heterogeneous learning, model serving, and other shared services.

Product layer : AI‑powered financial products such as search recommendation, intelligent marketing, smart客服, financial brain, and sentiment analysis.

This architecture powers Ant’s own businesses (payments, local services, e‑commerce) and is intended to be opened to external financial institutions.

Ant Group Intelligent Customer Assistant

The assistant combines two modules: “Guess Your Question” and “Intelligent Q&A”. It handles 4‑5 million daily help requests, with a self‑service rate of 94 % (97 % during Double‑11) and a satisfaction boost of 3 % over human agents.

Incorporating user behavior trajectories improves prediction accuracy beyond manually engineered features.

Deep learning replaces linear models, raising precision.

Automated knowledge‑base construction via semantic clustering extracts standard questions and answers from raw user queries.

Data‑closed‑loop enables online learning and continuous model improvement.

Additional innovations include LSTM + DSSM models that fuse behavior sequences with textual input, and the use of Decomposable Attention and other state‑of‑the‑art techniques.

AI + Finance Topics Worth Attention

Multi‑agent game learning : Professor Deng Xiaotie discussed Bayesian‑optimized auction strategies and the challenges of convergence in dynamic, multi‑player environments.

Reinforcement learning : Professor Wang Mengdi traced AI history from control theory to deep reinforcement learning, illustrating how agents learn complex policies through interaction, e.g., poker bots.

Personalized recommendation and resource allocation : Professor Zhang Shengyu categorized six AI‑finance application areas, emphasizing personalized product recommendation, credit scoring, pricing in sharing‑economy platforms, algorithmic trading, fraud detection, and product enhancement.

Overall, the forum showcased how AI can not only cut costs and boost efficiency in finance but also promote fairness and broader access to high‑quality financial services.