During the pandemic of the new coronavirus pneumonia, caring for the elderly becomes more difficult. Does artificial intelligence play a role in this field?
On April 6, local time, on a live broadcast of the Stanford University's People-oriented Artificial Intelligence School (HAI), Fei-Fei Li, a professor of computer science at Stanford University, introduced the artificial intelligence home system to the outside world. The symptoms of the new coronavirus pneumonia while also ensuring privacy.
The purpose of this AI system is to help the elderly (mostly those living alone) stay in touch with their families or medical caregivers. The best way to protect the elderly is to reduce contact with people, especially those with new coronavirus pneumonia who have not yet shown symptoms.
According to Fei-Fei Li's team, the advantage of this home system is that it allows caregivers to remotely monitor the elderly's existing diseases and basic health conditions, reducing the risk of exposure.
Fei-Fei Li and her team introduced in a live speech that this system was already being developed by an interdisciplinary research team composed of clinicians and computer scientists before the outbreak of a new coronavirus pneumonia.
"Over the past few years, we have been studying an AI system that can help the elderly live independently and manage their chronic diseases. Recently we realized that this technology will also help the elderly under the new coronavirus pneumonia pandemic "Fei-Fei Li said in his speech.
According to Fei-Fei Li, the entire home AI system includes cameras and smart sensors installed at home. In the speech, Fei-Fei Li mentioned four sensors, including camera, depth sensor, thermal sensor and wearable sensor.
The research of the whole team mainly focuses on the first three. Since privacy is very important in this system, research on cameras is more challenging. "The camera can reveal the details of personal activities, but it does not meet the privacy needs of most people." Fei-Fei Li said.
How does the entire system work, and how to ensure privacy? Fei-Fei Li introduced one by one in the speech. When the sensor obtains data, the system sends it to a secure central server for processing.
However, in this process, Fei-Fei Li also admitted that there are still security risks at this stage, such as being threatened by cyber attacks. But she emphasized that researchers will follow privacy and security guidelines throughout the process.
The team equipped the edge devices with encrypted disks, used to delete data related to user privacy, obfuscated faces, encrypted them, and then transmitted them to the cloud.
Once the data reaches the server, a group of clinicians and AI experts will analyze and annotate it to develop a machine learning model. After training, this model can identify some clinically relevant behaviors, including breathing, sleep, diet and other behaviors. Fei-Fei Li said that the team is currently developing a model that involves activities of daily life. The model can calculate whether the user's health has deteriorated.
But this model is not an in-depth and extensive analysis of all the daily activities of users. It is necessary to find a balance between privacy and public safety.
The trained model can be deployed to edge devices and run locally. In this way, the research team built a closed-loop system, and data security can also be guaranteed.
However, this closed-loop system cannot further update and improve the model. To address this, Fei-Fei Li mentioned that the team is envisaging the use of joint learning and unsupervised learning, that is, without manual annotation, the model on each edge device is updated to use the new environment and improve the Lu Greatness.
Through joint learning, the team can limit security attacks to devices to reduce privacy and security threats to the cloud.
Finally, the system also needs a method that can transmit the detection result of the intelligent sensor to the medical staff or family members.
Fei-Fei Li said that the team has not yet found a specific solution, but is considering using a mobile application or Web interface.
"These sensors are not meant to make diagnostic decisions or replace clinicians, but can continue to appear, keep an eye on the elderly at home at any time, and alert clinicians and family members in time."
At the end of the speech, Fei-Fei Li said: "Of course, in every step of this research and the deployment of this technology, we must take a comprehensive consideration of ethics, privacy and security."
The challenges posed by the current new coronavirus pneumonia pandemic include not only ensuring the safety and health of the elderly, but also broader and urgent follow-up of diseases and people who should be quarantined.
When asked whether this system can also solve this problem, Fei-Fei Li said that the team is not willing to get involved in this field. "Our goal is to propose cutting-edge computer vision and machine learning technologies to help solve some of the most important and challenging issues in healthcare, and at the same time propose security guidelines for ethics, privacy, and AI healthcare research." Fei-Fei Li Say.
At present, this project is still in the research stage. The entire team also needs to complete the construction of the data set and model work, and the team did not disclose how much time will still be required to complete.
However, the team has completed a pilot study in an assisted living facility in San Francisco in collaboration with On Lok, a company dedicated to high-quality senior care in the United States, and will enter the next phase of the study.
Special Report: Fighting The New Coronavirus
