To start, I introduce Goal-Oriented Algorithms in Large Space (GOALS) involving transformer-based reinforcement learning or panel trees, which are particularly suited for answering questions related to optimal decision-making and modelling grouped heterogeneity in social science research. In several specific financial applications, I show how GOALS can effectively and flexibly manage investment portfolios, generate test portfolios or latent factors for evaluating extant pricing models or more accurate pricing, and separate assets of higher and lower return predictabilities under different macroeconomic regimes. Among them, I highlight how GenAI based on GOALS can assist corporate decision-making that entails complex, high-dimensional, and non-linear stochastic control during which managers possessing various business objectives learn and adapt via dynamic interactions with the market environment.

With pre-trained foundational models and GOALS effectively capturing human agents’ optimizing behavior in a given economy or market, or supplying the machine equivalent of that, I further introduce the concept of data-driven generative equilibrium for counterfactual analysis. Specifically, I show how one can take a data-driven approach to examine the counterfactual equilibrium in the online lending market when borrowers endogenously adopt LLMs to complete loan applications. Extending this further, I will discuss AI-Agent-Based modeling and how that can be combined with economic world models.

I conclude the talk with remarks on several caveats or challenges when applying AI in social science research: (i) When using AI agents for experimentation or to generate counterfactual data, : we need to understand AI agents as a new species that potentially differ from humans, necessitating the new field of behavioral economics of AI; (ii) Large models and computations may not be the optimal path, and we need to effectively bridge theory with data; (iii) Before we can train AI social scientists, we need to build the ImageNet equivalent for empirical studies in the social sciences.

Lin William Cong is currently the President’s Chair Professor in Finance, Computing and Data Science at Nanyang Technological University and the Director of the Global Institute of Finance, Technology, and Society (GIFTS). A leading scholar at the intersection of financial economics, AI, and digital technologies, he has played a foundational role in building the fields of digital economics, (blockchain) platforms, tokenomics, generative modeling, and AI for social science.