Addressing AI risks as well as AI benefits
As AI tools become more powerful and widely accessible, Fang is increasingly examining their unintended consequences. When ChatGPT was released in late 2022, he immediately recognized both the opportunity and the risk.
“We quickly realized two big problems: the susceptibility to misinformation and the potential for social, gender, and racial bias,” Huang said.
Fang and his collaborators conducted research demonstrating that AI-generated content can reflect and amplify existing biases embedded in training data. These findings have important implications for organizations leveraging generative AI for communications, recruiting, marketing, and decision support. As policymakers seek to ensure fairness, transparency, and accountability in AI systems, research that identifies biases and mitigation strategies is becoming increasingly important.
For Fang, maximizing the benefits of AI requires equal attention to minimizing the risks. Responsible design needs to be built in from the beginning, rather than being retrofitted after problems arise, he argues.
From theory to practice
Many of Fang’s projects are designed with direct business applications in mind. One example is his work on industry classification systems that group companies into sectors for use by governments, investors, and financial analysts.
“Traditionally, the industry has been classified manually,” Huang said. “The process is time-consuming, costly, and subjective.”
Existing systems, such as the U.S. government’s Standard Industrial Classification (SIC) and the Global Industry Classification System (GICS) used by financial companies, rely on human judgment to assign companies to categories. Fang’s AI-based system automatically groups companies together by analyzing annual report filings and identifying similarities in business activities and language patterns.
The result is a more adaptive, scalable, and objective classification approach. You can also automatically assign new companies to industries based on their disclosures. Applications range from portfolio construction and risk assessment to executive compensation benchmarking and academic research.
“This is an example of how AI can make existing processes more accurate, efficient and transparent,” said Huang.
Let the AI explain itself
Transparency is especially important when AI systems are used in high-stakes situations. In a study published in business administrationFang and his co-authors have developed an interpretable AI model to help diagnose depression associated with chronic illness.
“For mission-critical tasks such as medical diagnosis, it is not enough for AI systems to be accurate,” says Huang. “You also need to explain why you made certain predictions.”
The model reflects aspects of human reasoning by learning representative “prototypes” from the data. When evaluating a patient, the system identifies which learned prototype best matches the patient’s symptoms and uses those comparisons to explain the diagnosis. Rather than acting as a black box, this model provides inferences that clinicians can evaluate and question.
Such interpretability aligns closely with emerging regulatory priorities that emphasize accountability and accountability in AI systems. For Fang, transparency is not just a compliance requirement, but a design principle that strengthens trust and long-term effectiveness.
Rethinking regulation and innovation
As new AI regulations take shape in the U.S. and abroad, Huang encourages organizations to rethink how they structure compliance.
“Many people see regulation as a constraint,” he says. “I think that’s the goal.”
He likens business strategy to an optimization problem. Traditionally, companies have sought to maximize profits or minimize costs. Huang argues that social goals such as fairness, accountability, and transparency should be incorporated directly into that optimization framework.
“In the long run, companies will benefit from aligning their economic and social goals,” Huang said. “Responsible AI builds trust. Trust is essential for sustained success.”
Rather than slowing down innovation, well-designed guardrails can facilitate more thoughtful and resilient AI adoption. Organizations that proactively incorporate responsible design principles may be better positioned to adapt to evolving regulatory expectations.
nurturing the next generation
In addition to contributing to research, Mr. Huang is committed to mentoring doctoral students and developing future scholars.
“We need to train students so they can join us,” Huang said. “I really enjoy watching them grow into independent researchers.”
Many of his former students have gone on to their own academic careers, extending the impact of his approach to responsible, application-driven AI research.
As we enter an era in which artificial intelligence is more regulated and has consequences, Fang provides a stable and informed voice with decades of focus on use-focused and responsible AI. His research emphasizes the core principle that innovation and accountability are complementary rather than competing goals. Designing AI systems that are consistent with transparency, fairness, and social values may ultimately determine how the technology can reshape business and society in the years to come.
