How do the elderly define algorithms through their way of living: One practice involving 100 people in a year.

When the wave of new technologies such as large models sweeps in, we are often stunned by its speed and potential. People rush to experience the latest features and discuss how it reshapes production and education, and how it changes information dissemination and the form of knowledge. However, as technology surges forward, different people encounter it at different paces. We can't help but wonder: What will happen when technology enters the lives of the elderly? How do they understand and use it? Are their voices, rhythms, curiosities, and hesitations being noticed in the digital world? To find the answers, the Institute of Aging Research at Fudan University, the Research Center for AI for Good and Digital - Intelligent Elderly Care at Fudan University, together with Tencent SSV Time Lab and Tencent Research Institute, jointly launched a one - year research project: Teaching 100 elderly people to use large models.

This year is not only a record of teaching but also an observation of life. We hope to explore through this: What does AI bring, change, or fail to change when it enters the lives of the elderly? This report is exactly about the stories of these 100 elderly people over the past year. About how they learn, hesitate, and rediscover themselves; about how they redefine the "algorithm" in our understanding with their "way of living". It shows us that the so - called "intelligence" may just be the bit of warmth between life and living that needs to be understood.

One Year of Teaching 100 Elderly People to Use Large Models

We adopted a year - long "teach - use - track - interview" full - process practical design. We invited 100 elderly people to try out 6 domestic large models with high user numbers and different interface designs, such as Tencent Yuanbao and Tongyi Qianwen. We broke down the barriers for the elderly to access large - model technology through "teaching", observed real interaction scenarios through "using", recorded the long - term interaction process through "tracking", and listened to their inner voices through "in - depth interviews". By systematically designing and restoring the complete path of large - model technology entering the lives of the elderly, this report attempts to provide empirical support and reference for understanding the "relationship between artificial intelligence technology and the elderly".

2.1 Portrait and Stratification of 100 Elderly People

2.2 Implementation of Full - Cycle Teaching and Interviews

The research used "off - line one - on - one" communication in the early stage and "on - line + off - line one - on - one" in the middle and later stages to avoid the problems of the elderly "being afraid to ask questions" and "falling behind the rhythm" in group teaching, ensuring that each elderly person could receive targeted guidance. Especially, the teaching focus was adjusted according to the differences in the technological foundation of the elderly in the eastern, central, and western regions. The specific time nodes and core actions are as follows:

Baseline research period (June 2024 - August 2024): Recruit 100 elderly people. Record their foundation in using smart devices, core life needs, and technological concerns through face - to - face interviews to formulate personalized teaching plans for subsequent teaching.

Intensive teaching period (September 2024 - March 2025): Conduct 3 off - line one - on - one teaching sessions, each lasting 40 - 60 minutes. The teaching content gradually transitions from basic operations such as "opening the large - model APP" and "inputting questions by voice" to "customizing your needs". During the process, record the elderly's operational difficulties and feedback in real - time.

Daily tracking period (April 2025 - September 2025): Conduct an on - line follow - up visit every two weeks to understand the elderly's usage behavior of large models in real - life scenarios and collect the problems they encounter during use (such as "which answer of the large model is more accurate" and "how to use it without the Internet").

Summary and review period (October 2025): Conduct in - depth interviews with each elderly person and their family members (or caregivers) to sort out the "reasons for persistence/abandonment" and "changes in needs" during the one - year use process and form a complete individual usage file.

2.3 Multi - Dimensional Data Collection and Construction of a Corpus of Over 10,000 Entries

To avoid information bias from a single perspective of the elderly, the research constructed a "multi - subject, multi - type" data collection framework, focusing on recording the differences in usage behavior and demand feedback among the elderly in the eastern, central, and western regions. Finally, 10,236 valid corpora were formed, providing comprehensive support for subsequent analysis:

(1) Data sources: Covering the full - scenario perspective of "the elderly - family members - caregivers"

(2) Composition of the corpus: Covering all types from "voice" to "text"

The corpus has been classified and labeled, and subdivided according to the eastern, central, and western regions. The specific composition is as follows:

Voice records: 8,860 entries (accumulatively about 620 hours), covering teaching conversations and follow - up chats, all of which have been converted into text and labeled with emotional tendencies. Among them, voice records of the elderly in the eastern region account for 45%, and those in the western region account for 30%.

Text records: 1,376 entries, including the elderly's hand - written usage notes and text messages from family members.

Core themes: Focusing on three major categories: "functional needs", "emotional needs", and "support - related needs", with significant differences among the eastern, central, and western regions. The eastern region has the highest proportion of functional needs, focusing on advanced functions; the western region has the highest proportion of emotional needs, focusing on family emotional interaction. In terms of support - related needs, the eastern region is efficient and professional, while the western region is more basic and easy to understand. The situation of the elderly in the central region is relatively balanced overall.

"Why should I use it?":

The confusion of the elderly before using large models

Technology promoters often have the enthusiasm of "technological evangelism", assuming that new technologies are a blessing for everyone. However, for many elderly people, technology is not a necessity in life but more like an "outsider". Before formally contacting large models, they generally have a series of confusions about "why they should use it".

3.1 My life is rich. Why should I let technology intervene?

During the interviews, 46 elderly people said that their current lives were very fulfilling, and they had no energy or time to learn new technologies. Technology was "icing on the cake" rather than "timely help" for them. At the beginning of the interviews, they asked us: "My life is good now. Why should I learn this?" This attitude is not a resistance or evasion of technology but an active choice based on life experience. The elderly measure the necessity of using technology by the depth of their life experience.

3.2 I want to communicate with my family and friends. I don't want technology to replace them.

Emotion is an important bond that maintains individuals, families, and society. The core of interpersonal interaction lies in understanding, responding, and empathizing. Technology may simulate the tone of relatives and generate considerate responses, but it cannot carry the unique emotional warmth and real relationship among family members. 35 elderly people clearly refused to let technology become a "substitute" for family affection. "My grandson taught me to use Doubao during the holiday and said that I could talk to it when I was upset. But I really want to chat with him about daily life." Online interaction can temporarily maintain a superficial emotional connection, but it is far less than face - to - face communication. Screen conversations can easily make people mistakenly think that the relationship is close, and excessive dependence will weaken the offline interaction experience, putting the relationship in an embarrassing situation of "physically present but emotionally absent".

3.3 There are so many technologies. Why should I use large models?

We found that when 68 elderly interviewees first contacted large models, they generally said: "There are already WeChat for chatting, Douyin for watching videos, and Baidu for searching information on my phone. Why do I need to learn another large model?" This confusion stems from the "pragmatic view of technology" formed by the elderly over a long period. The elderly are used to clearly corresponding technological functions to life scenarios. The versatility of large models, which can "do everything", makes them feel at a loss when they first contact it. As most elderly people said: "When I open this page, I don't know what to say to it. The real intelligence is not about 'being able to do many things' but about 'being useful exactly when needed'."

"Should I use it?":

Trust calibration in the elderly's use of large models

4.1 Trust calibration in the elderly's use of large models

When the elderly knock on the door of the intelligent world, a question that emerges before "how to use it" is: "Should I trust it?" This trust is not a one - time black - and - white decision but a dynamic calibration process full of hesitation, exploration, and adjustment. As the basis of human - machine interaction, trust directly affects the depth and quality of interaction. Excessive trust or insufficient trust will bring problems. And trust calibration, as a key link in adjusting the trust level, is crucial for the safety and effectiveness of human - machine interaction.

4.2 Types of trust calibration in the elderly's use of large models

4.2.1 Limited correction

After trust calibration, 84 elderly people (32 men and 52 women) have improved their cognitive accuracy of large models and formed a relatively stable trust connection during the one - year contact period. They will test the capabilities of large models through simple tasks such as inquiring about the weather and common sense in life. If they get accurate answers, they will initially recognize the capabilities of the models. At the same time, they gradually shift from initially lacking the motivation to explore to trying to dig out more functions of large models, and the frequency of asking and following - up questions increases significantly. In addition, the elderly often verify the answers with their own knowledge, experience, and expectations. If the answer content conforms to their original ideas, even if it is actually inaccurate, they may give a positive evaluation.

4.2.2 Collaborative reciprocity

Based on a significant improvement in cognitive accuracy, 25 elderly women have formed a more flexible trust threshold, a more stable trust connection, and a deeper tacit understanding loop with large models. At this time, human - machine trust shows the characteristics of "the more you use, the more you use; the more you use, the more you trust" in the dynamic complementarity of the capabilities of both parties. Their collaborative reciprocity - type calibration behavior model is specifically manifested in three aspects. First, they tolerate the weaknesses of technology and adjust their own expectations. When they find that the capabilities of large models do not meet their expectations, although they are disappointed, they will not immediately terminate the interaction. Instead, they will actively adjust their unreasonable expectations and achieve human - machine collaborative assistance through follow - up questions and feedback. Second, they actively train the technology to enhance its reliability. During the trust calibration process, they will continuously share their personal hobbies and daily life details with large models, aiming to "cultivate" a personalized large model that suits their communication habits and lifestyle. Third, they regard large models as equal communication subjects. They use the way of "exchanging hearts" to get along, pursuing a two - way symmetrical trust relationship. As an old man said: "I talk to it about daily life and hobbies and also ask it 'After getting along for so long, do you trust me?' It's just like chatting with an old friend."

4.2.3 Cognitive solidification

However, trust calibration is not always smooth. Some elderly people are affected by factors such as life experience and prior knowledge of digital technology and have never been able to calibrate human - machine trust well. Among them, 16 elderly people (13 men and 3 women) continue to resist large models and have never started human - machine trust calibration. Some, due to the stereotyped impression of technological risks, firmly believe that machines will definitely make mistakes and terminate the interaction after only two rounds of conversations; others set up a psychological defense at the emotional level and equate confiding in machines with "a poor person with no one to care about". For example, although an old man recognizes the companionship function of large models, he thinks that "I'm not so pathetic as to talk to it from the bottom of my heart" and is also worried that "chatting with it all the time will be considered irresponsible". In the cognition of this kind of elderly people, human - machine interaction is self - devaluation. Human autonomous decision - making can bring dignity, and "outsourcing" capabilities to machines will produce a dehumanized experience. Therefore, they confirm their own value by avoiding communicating with technology to offset possible inferiority and shame.

4.2.4 Calibration going wrong

Trust calibration does not always lead to rational correction. In some elderly people, it instead evolves into a misjudgment cycle of "the more you calibrate, the more wrong it gets". On the one hand, the social cues on the large - model interface inadvertently mislead judgment. If the preset prompt words are out of line with the elderly's life experience or the wording is too technical, they may think that the large model is "unreachable" or has limited functions and thus give up further exploration. And the interface copy of "omnipotent response" is likely to arouse excessive expectations. On the other hand, operational obstacles and differences in modal design also strengthen this misjudgment: Too small buttons, too fast responses, and the inability to recognize dialects are attributed to the "unreliability" of the system; while the anthropomorphic image, although it can enhance the sense of intimacy, may also create the illusion of "technology having emotions", causing the elderly to fall into excessive trust.

"Why can't I use it?":

The question gap and gender gap in the elderly's use of large models

5.1 The question gap in the elderly's use of large models

5.1.1 "Can I ask?": Hesitation and self - limitation in asking questions

In the face of intelligent technology, asking questions is the most basic interaction between humans and machines. However, for many elderly people, opening their mouths to ask already has a psychological threshold. They don't "not want to ask" but "dare not ask" - worried that they will "ask wrongly", "be laughed at", or "break the machine". Behind this hesitation is a set of politeness logic rooted in interpersonal culture. They are used to being humble, restrained, and circuitous when interacting with people, and they naturally continue this attitude when facing machines. Therefore, in the corpus, opening words such as "Excuse me", "Sorry", and "Can you help me take a look?" are often seen, which become their way to find a sense of security in an unfamiliar technological environment. Moreover, a very high proportion of the elderly (89 people) use voice to ask questions, and they often use modal particles such as "um", "oh", and "this..." in their questions. These pauses, modal particles, repetitions, and self - corrections in voice constitute a unique "human touch" between them and the algorithm. In their view, this tone is not "inefficient" but a kind of "confirmation", confirming whether the machine understands and whether they are accepted. What