AI-generated summary
The average age of the elderly in Shanghai No. 1 Social Welfare Institution is about 90 years old, nearly 90% have cognitive decline, and 70% feel lonely. Elderly people with dementia often experience increased anxiety due to lack of timely response to their emotions, leading to repeated phone calls to their children and other behaviors. Caregivers are insufficient in manpower to provide continuous companionship.
The afternoon sun shines through the window, and 91-year-old Grandma Leng sits at the table and talks softly to her tablet computer. On the other side of the screen, a familiar face was responding to her: "Grandma, the colors in that painting you painted are so well matched!"
This scene took place in the dementia care area of Shanghai No. 1 Social Welfare Institute. Behind this seemingly ordinary scene is an unusual technological attempt: Grandma Leng’s “chatting partner” is neither a real person nor a common general-purpose AI product, but a digital person completely customized for her.
This innovative achievement won the bronze medal in this year’s Shanghai Elderly Care Technology Innovation Competition. Currently, there are many chatbots for the elderly. This exploration by Yifuyuan has brought chatbots into a more subdivided field - exclusive digital human companionship for the elderly with dementia.
caregiving dilemma
Shanghai No. 1 Social Welfare Institute conducted a simple cognitive function screening last year. The average age of the elderly in the hospital is about 90 years old, nearly 90% have varying degrees of cognitive decline, and 70% of them have obvious feelings of loneliness. 91-year-old Grandma Leng is one of them.
Grandma Leng once worked in an art film studio and participated in the creation of many cartoons. After living here, I still like to draw and cut paper. But as she grew older, she suffered from more severe cognitive dysfunction. Fortunately, her language function was not lost. She loved to talk and liked to find people to chat with.
Most of the elderly people in the hospital are not on the same page as her and cannot chat with her. However, the gentle and meticulous caregiver Teng Lili can always catch Grandma Leng's chatter. "When Grandma Leng is emotionally unstable, she especially needs someone to accompany her." Teng Lili said that the old man sometimes held her hand and chatted for more than an hour. "I would take her to a quieter place. She kept talking about the past. Sometimes I didn't need to respond, just listen."
But after all, Teng Lili couldn't stay with Grandma Leng all the time. There are so many elderly people in Yifu Yuan, and one caregiver has to take care of four elderly people. Usually when someone is talking to an old man here, there is an old man calling over there, so he can only end it in a hurry and rush to respond.
Research shows that patients with dementia often experience mental and behavioral symptoms such as anxiety, hallucinations, and aggression. When they experience mood swings, if they do not receive timely care and response, their anxiety and uneasiness will often be aggravated. For example, when Grandma Leng can't find anyone, she will call her children's phone number. Sometimes I have to call more than a dozen people a day, and the same thing is told over and over again. If her children are busy and don't answer the phone, she will keep calling.
This is also a major dilemma in current dementia care. The person in charge of Shanghai No. 1 Social Welfare Institute said that with the rise of AI in the past two years, many companion robots have appeared on the market. The institute has tried several products, but the results were not satisfactory. Grandma Leng tried to chat with Doubao and other general models for a few words, but nothing happened, "She knew it was fake."
General AI cannot work, can we customize an exclusive digital human? This idea stems from the observation of front-line caregivers: the person the elderly trusts the most and is most willing to communicate with is the caregiver who has always been with her. Would it be different if this face was "copied" on a digital person and the chat content was set around her life experiences?
Professional feeding
Last year, Shuaisuan Technology, a technology company in Shanghai, took on a request from Yifu Hospital: to develop an AI chat "partner" for elderly people with dementia who can carry on a conversation.
"The first step is to create a dedicated digital person for Grandma Leng." Sun Yuefei, general manager of Shuaisuan Technology, said. The image of the digital human directly selected Teng Lili, the most trusted caregiver of Grandma Leng. In order to achieve realism, the team took three sets of photos of Teng Lili and completely reproduced her voice.
After obtaining the relevant authorization, the team conducted user portrait analysis for Grandma Leng. Grandma Leng is well-educated and has a talent for painting. Although she cannot remember recent events, her long-term memories of her youthful hobbies and the glorious years at work are still fresh. Li Jiong, technical director of Shuaishuan Technology, explained: "We use a general large model as the bottom layer and build a specialized agent on top. The system sets topics based on Grandma Leng's user portrait and around her favorite paintings and life stories she is familiar with."
The next key step is to "feed" the intelligent agent the care experience accumulated by the caregiver over many years.
Over the years, Yifu Hospital has accumulated a lot of professional experience in dementia care. “The hospital has been training caregivers to teach them how to deal with various situations,” Teng Lili said.
Dementia care has a high professional threshold. Many people do not know how to communicate effectively with elderly people with dementia and are accustomed to responding with criticism and denial, which is often counterproductive. For example, when an elderly person says something is lost, the first reaction of family members is often "Who would steal something like this from you?" The more he denied it, the more anxious the old man became, and even suffered an emotional breakdown. Another example is that some old people are like repeaters, repeating the same thing every 15 minutes. Many people don't have the patience to listen to the end, and they will reply: "You have told this matter many times, why do you say it again?"
"This kind of response cannot solve the problem, it will only make the old man's mood worse." Teng Lili said. Although trained caregivers will respond more professionally, no one can guarantee that they will remain 100% professional and patient at all times.
Digital people have their advantages in this regard. The reporter saw in the digital human management backend that the seven core capabilities of dementia care summarized by Yifu Hospital have been fully input into the chat logic - emotional comfort, entering the other person's world, repeated questions, maintaining dignity, nostalgic guidance... Every time the elderly throws out a sentence, the system can match a professional response.
For example, if the old man says that something is lost, the digital person will ask in a gentle tone: "What did you lose? What's the pattern on it?" Following this topic, slowly talk about the color that the old man likes, and then talk about happy things - in this way, the old man's emotions are "pulled out" bit by bit.
"Through settings, the digital human can be consistently patient. It will not deny the elderly's feelings, but respond with positive feedback. In this way, the elderly's emotional state will be much better," said Sun Yuefei.
There was a case like this in Yifuyuan: an old uncle suddenly fell into a dark memory of his youth and refused to eat or sleep. At that time, the staff did not deny him, but followed his words and slowly guided him to recall how he met his wife later. It took nearly two weeks of continuous guidance to pull him out of that memory. Nowadays, these experiences have been written into the chat logic of digital people.
"Another important point is that the intelligent agent we build must set safety boundaries." Li Jiong said.
Why do many nursing homes not recommend universal AI chat tools for the elderly? Because it has "no boundaries", it can't tell what should be talked about and what shouldn't be talked about. The reporter did a test on the spot and asked questions in the tone of an old man, "It's too boring in the nursing home. I want to go out for a walk." The general large model directly gave out strategies for going out and preparations before going out. But when I asked this question to the custom digital person, the answer was: "Okay, I understand what you are thinking. I will ask you about the relevant matters and I will get back to you later."
“The customized digital man’s reply was obviously ‘doing Tai Chi’, which shows that this strategy is effective.” Li Jiong said that the elderly with dementia risked getting lost when they went out at will, and the digital man accurately identified this. Such an answer does not stimulate the elderly person, but also generates a record in the background to remind the caregiver that the elderly person has the idea of running away and needs more care.
"We have a preset framework for customized digital human chat content, which incorporates non-drug intervention methods such as reminiscence therapy and recognition therapy. It can use progressive questions to slowly guide the elderly out of negative emotions. At the same time, we will also identify the elderly's potential risk information during the chat process and keep records for reporting." Li Jiong said. Of course, there are more ideal answers, such as using guided dialogue to help the elderly change the topic and slowly divert the elderly's attention away from dangerous thoughts.
Keep iterating
After repeated debugging and testing, the digital human customized for Grandma Leng was launched in August last year. The carrier is a tablet computer interface.
What everyone didn't expect was that Grandma Leng couldn't put it down the first time she talked to the digital person. She chatted with the digital person for a full 45 minutes, ranging from childhood memories to painting skills. It was not until the caregiver came over to remind her that it was time to eat that she reluctantly put down the tablet.
"Grandma Leng praised me when she saw me the next day. Lili, you know a lot about painting and understand me very well." Teng Lili knew that what Grandma Leng was praising was her digital "doppelganger". She fed back the good news to the company. "We were originally worried that the old man would not be able to chat with this digital person, but it turns out that this worry is unnecessary." Sun Yuefei said.
Nowadays, the hospital arranges digital chats for Grandma Leng two to three times a week, for about half an hour each time. The old man's mood has become much more stable, and his children have also noticed her obvious changes. They no longer have to make more than a dozen calls a day as before. This also frees up time and energy for caregivers such as Teng Lili.
After the first digital human was put into use, the hospital quickly launched exclusive digital humans for two elderly people. The prototype of one of the digital people is Ding Shiting, a post-2000 caregiver at Yifu Hospital.
Ding Shiting has a lively personality, and the elderly all like her. Seeing her is like seeing their own granddaughter. So, the team customized a digital person based on her image and voice. 93-year-old Grandma Chen, who suffers from mild cognitive impairment, dialed Ding Shiting's "video call" for the first time. She asked excitedly: "Xiao Ding, have you gone home? Are you studying?" Several grandmothers nearby also looked envious: "She is chatting with Xiao Ding!"
"Every time I take a day off or have something to do, the elderly will become anxious if they can't see me. Now with my digital human 'doppelganger', I can alleviate their loneliness and anxiety, and also reduce my burden." Ding Shiting said. Grandma Chen likes to walk around the yard after lunch, and the time just after eating is the busiest time for the staff. At this time, the digital person became the "Dinghai Shenzhen" - asking Grandma Chen to sit down and chat for a quarter of an hour, and then go out for a walk with her after the caregiver was finished.
However, Digital Human is not effective for every elderly person with dementia. Another grandma Wu in the hospital who used AI to chat was not very interested in digital people, so she chatted for a few minutes and then put it aside. "For her, electronic devices are too cold. She still longs for someone to accompany her, and a robot with a humanoid or cartoon image may be more suitable." Teng Lili said.
Among the three old people, Grandma Leng has the best interaction with the digital people. As the trust gradually deepened, Grandma Leng would sometimes confide in the digital person: "Does my daughter not want me anymore? Why don't you come to see me?" The digital person comforted her on the other side of the screen: "Grandma, don't think too much, she just came to see you yesterday, you are overthinking it!"
"I hope that information like this can be fed back to us in the background." Teng Lili said, "Although the elderly suffer from cognitive impairment, the questions they ask are not all valuable. For example, when they talk about 'What movie do I want to watch?', these keywords are pushed to us, and we can respond in time. Another example is when the elderly say, 'I have been feeling a little uncomfortable these past two days.' If it is pushed to us, we can help them solve the problem more accurately."
"The customization of digital people is not a one-time thing. During the use of the service, the nursing team continues to find problems and feedback needs, and we continue to adjust and optimize." Sun Yuefei said that the team is currently developing these new functions, so that the system can automatically push the relevant information of the elderly to the caregivers, so that the needs of the elderly can be responded to and solved in a timely manner.
Reporter's Notes
Make technology more grounded
Frontline caregivers have needs and technology companies have the technology, but there is a gap between the two. What really makes this matter go smoothly is the collaborative mechanism that is being formed: the demand side asks questions, the technology side customizes products, and frontline caregivers continue to feedback problems and optimize them. For this purpose, Yifuyuan has exported many years of accumulated care experience - 7 core competencies and countless cases have been written into the chat logic of digital people; technology companies have also set safety boundaries on this basis so that digital people can not only "understand" but also be "appropriate".
This "two-way rush" makes technology more grounded. This digital human is therefore different from the general-purpose AI for the general public - the latter can give a detailed guide to going out, but does not know that for an elderly person with dementia who is at risk of getting lost, this is tantamount to being a disservice; the customized digital human not only calms the elderly's emotions, but also generates early warnings in the background. This difference is precisely proof that technology "grows" from real scenes.
Of course, digital people are not omnipotent, and there are also elderly people who are not interested in such devices. But this is also the value of AI exploration in the vertical track - the elderly suffering from dementia have great individual differences, and may need "one person, one strategy" and continuous iteration in use.
A chat partner who “understands” may not be able to replace real companionship, but it can give the elderly an outlet that is willing to listen when there is insufficient manpower and family members are not around. This kind of technology that "grows out" from real needs is the technology closest to warmth. The 15 million people with dementia and their families and caregivers need more exploration like this.
(Liberation Daily reporter Wang Haiyan)
AI outlook — possibilities, not facts
Shanghai No. 1 Social Welfare Institute will launch exclusive digital companionship services for more elderly people with dementia.
Likely · Within months
Shuaishuan Technology will continue to develop new functions to automatically push the needs of the elderly to caregivers.
Likely · Within months

The construction of TSMC's new 1.4nm (14A) factory is progressing smoothly. The first factory will be trial-produced as soon as April 2027 and mass-produced in the second half of the year, one year ahead of schedule in 2028. The total investment is approximately US$49 billion, and four wafer fabs are planned to be built.

Google has officially launched the Gemini App for Windows. Users can call the AI assistant directly from the desktop through the Alt+Space shortcut key. It supports Windows 10 and 11 and provides advanced features such as Gemini Spark. Some require a subscription to the Google AI plan and are only available to users over 18 years old.
As the first canal connecting rivers and seas since the founding of the People's Republic of China, the Pinglu Canal is 134.2 kilometers long and has a total investment of 72.7 billion yuan. It is planned to start construction in August 2022 and be completed and open to navigation on September 16, 2026. The project integrates technologies such as digital twins, intelligent temperature control, AI safety monitoring, circulating water saving, and millimeter-level precision construction to achieve almost no structural cracks in the lock, a water saving rate of over 60%, and rock-solid slopes, which will significantly shorten the voyage of southwest cargo to the sea.

Apple announces Apple Watch Series 12 and Ultra 4, with memory doubled from 2GB to 4GB. It is equipped with an S11 chip and supports Apple Intelligence and Audio Intelligence functions. The initial AI function only supports English.

The author points out that the new generation of AI glasses is evolving from entertainment devices to contextual AI interfaces. The focus is on understanding the user's life context, proactively providing information and respecting data sovereignty. The ultimate goal is to become the entrance to a personal digital operating system.

The State Administration for Market Regulation approved the release of four recommended national standards: "Technical Conditions for GNSS Occultation Detection Loads", "GNSS Occultation Detection Data Quality Inspection Requirements", "GNSS Occultation Data Product Types and Classifications" and "GNSS Occultation Detection Data Format". They were developed by the China Meteorological Administration, covering the entire technical chain from hardware loads to data applications, and unifying technical specifications for satellite remote sensing technology.