While generative AI is making rapid progress in office, entertainment, and e-commerce fields, a group of people who most need technological care are often overlooked by algorithms—seniors struggle with smart devices, and left-behind children face psychological difficulties without listeners. The Tencent Research Institute is trying to change this situation. Since 2024, it has launched a special program in collaboration with AI developers, building a sensitive scenario dataset for vulnerable groups, promoting large models from "intelligence" to "empathy."

The research team found that current mainstream AI models have common misunderstandings or emotional deficiencies when responding to complex social scenarios such as elderly care and children's psychological counseling in the local Chinese context. For example, when a left-behind child says, "Mom and Dad haven't come to see me for a long time," the model may only offer general comfort, rather than identifying the underlying loneliness, anxiety, or attachment needs; when an elderly person asks, "How should I take this medicine?" if the AI only repeats the instructions without considering the risk of dosage confusion, it could lead to serious consequences.

To address this, the Tencent Research Institute has partnered with several non-profit organizations to collect real conversations, life situations, and caregiving cases, building the first Chinese aging-friendly AI training dataset. It also plans to incorporate expert knowledge from fields such as psychology and geriatric medicine to create an "expert-level dataset," enhancing the accuracy and humanistic care capabilities of models in key scenarios.

However, this socially valuable work is facing practical challenges. Due to the limited consumption capacity of the target user group, related applications find it difficult to form a clear commercial return path, leading to restricted resource investment. Currently, these AI capabilities remain at the pilot stage and have not been widely integrated into mainstream products. To break through the bottleneck, Tencent is considering **open-sourcing some high-quality datasets**, calling on the academic community, public welfare organizations, and small and medium enterprises to participate together, building a more inclusive Chinese AI ecosystem.