AsianFin — Human-in-the-loop mechanisms are essential for overcoming AI hallucinations and opaque decision-making, according to Zhan Tingting, assistant general manager of Technology R&D at Guotai Junan Securities.
Her remarks came during the 2025 ITValue Summit, which took place in Sanya and was co-hosted by TMTPost Group and ITValue under the theme “Truth in AI Implementation.”
The summit focused on the top ten challenges enterprises face when deploying AI in real-world scenarios. Zhan outlined a four-pronged approach her team has successfully used to address hallucinations and decision black boxes in long-term AI practice.
First, she emphasized the importance of maintaining controllable atomic AI capabilities through a modular “1+n” large model framework. Second, she stressed ensuring data-level reliability by building both large and small models to guarantee trustworthy outputs. Third, Zhan highlighted comprehensive safeguards across computing power, algorithms, data, platforms, and applications to strengthen AI deployment and management systems. Finally, she advocated real-time monitoring of AI operations, anomaly detection, automated intervention, and human-in-the-loop auditing to ensure effective oversight and control.
Zhan concluded that combining automated AI processes with human supervision is critical for enterprises seeking to deploy intelligent systems responsibly, marking a growing recognition that AI’s transformative potential requires robust governance to be safely realized.
Explore more exclusive insights at nextfin.ai.
