Fighting COVID-19 with AI: efforts and lessons from China

By Zeng Yi and Sun Kang Source:Global Times Published: 2020/3/7 11:11:57

Photo: IC

The COVID-19 outbreak has revealed the necessity of being prepared and acting accordingly as a global human community with a shared future. From government organizations to industrial innovators, China has made tremendous efforts in its Artificial Intelligence (AI) sector by enabling technology to fight COVID-19. We feel it is imperative to share how AI could be beneficial, and what can be achieved, along with what needs to be regulated and governed. 

With a broader definition, AI, including big data analytics, has been used since the beginning to predict the emergence of COVID-19 for a potential outbreak. At China's Peking University, infectious disease models were designed to test prevention and control measures, and provide suggestions and early warning signals. 

Predictions on infected cases in provinces and cities have been provided by DATATOM, and the Chinese Academy of Sciences, etc. Similar efforts are especially essential for cities that have had a massive influx of returning employees after the Chinese Spring Festival. Trajectory information obtained and used for predictions and regulations not only came from the Ministry of Transport (MoT), but also telecommunication companies like China Mobile and China Telecommunication. This is essential, and not only in China, when regulating and controlling human mobility as similar methods could be useful in other countries.

To investigate and understand COVID-19, deep learning models were used to predict potential virus hosts based on gene sequence analysis and from studying different coronaviruses. 

Understanding the structural nature of COVID-19 and its subtypes, variations may be a continuous challenge, and efforts need to be made and shared across the world. Baidu Research released its LinearFold algorithm and services, which can be used for entire genome secondary structure predictions on COVID-19, and is supposedly 120 times faster than other algorithms.

Drug discovery, especially the drug screening phase, enjoys enormous support from AI, especially machine learning, knowledge representation, and reasoning. Potential drug resources related to COVID-19 are open for scientific research through collaboration between Global Health Drug Discovery Institute and Tsinghua University. Huangzhong University of Science and Technology in Wuhan and Huawei Cloud are collaborating to screen over 8,506 drugs available on the market or used during the clinical trial period. Related services could be provided through Huawei Cloud for Biopharmaceutical institutions. Knowledge extraction and inference are also included in the process at BenevolentAI in the UK.

AI has been used to assist the diagnostics of COVID-19, especially with automated CT image recognition and virus detection. By the end of February, Alibaba Damo Academy conducted over 30,000 CT imaging diagnosis as suspected COVID-19 cases with an accuracy of 96 percent, with each case taking only 20 seconds to test. The Institut Pasteur of Shanghai, Chinese Academy of Sciences (CAS) and Huawei Cloud have co-developed the Virus Identification Cloud (VIC), and can be used to detect different viruses including COVID-19. DeepBlue Technology (Shanghai) can automatically perform 2,000 nucleic acid detection tests daily.

Intelligent services have been provided by automated dialogue systems along with robots used to reduce physical interaction between patients and healthcare workers. Alibaba Damo Academy released an intelligent question answering service that reportedly had a 92 percent solution rate. 

Baidu released an intelligent out-call platform which has place over 1 million calls to gather statistics, and make announcements to those who live in local communities or to special groups who need extra care in Beijing, Xi'an, and Shanghai. Hospitals use robots 24-hours a day for drug distribution, food and household goods delivery, treatment to help fight COVID-19.

Automated surveillance has been deployed to find and control potential risks. Automated temperature monitoring and tracking applications have been deployed in subways, train stations, airports, and social service centers to identify and track people with high temperatures, and to assist with necessary actions. Considering the capabilities of automated systems, they could be of great help to assist screening (e.g. the version from Megvii could test 300 persons in a minute, and the version from SenseTimes can identify those who are without masks). 

Tracking isolated people with surveillance tools is efficient and safe. At Qinglongqiao Street in Beijing Haidian District, surveillance systems are used to automatically recognize the outdoor behavior of those who are still in the isolation phase. For evaluations and screenings and starting from Hangzhou, a health QR code has been adopted in over 200 cities. An automated system makes decisions on whether a person associated with the code is healthy and biologically safe enough to be around other people or in specific places. 

Although AI is very efficient as a supporting technology for controlling COVID-19, technical and ethical risks on privacy, bias, safety, and accountability have emerged. Although more personal information might be collected due to biological and social reasons during this period, access control and release of the information should be regulated accordingly. 

In February, personal information on Wuhan residents was posted online and in WeChat groups, and cause biases, isolation, and have negative effects for personal reputations. For the Health QR code in Hangzhou, many feel their colors are questionable since automated decisions were unfairly made by AI, such as receiving the red color only because a person had returned to Hangzhou from other places, while those places had not been marked sensitive. 

In this case, accountability needs to be discussed and designed for responsible use. The techniques and tools required to make predictions are essential to prevent a virus from spreading. Prediction accuracy needs to be improved to avoid unnecessary panic and concerns. 

It was also reported that facial recognition would be used to identify Chinese in some countries. During this period, people may feel worried, while some AI-based automated recommendation services have provided inaccurate or even fake news related to COVID-19 to the general public through mobile phone apps, which has caused anxiety levels to rise and was even coined "information outbreak." 

The use of face masks has also created challenges surrounding the technical safety of facial recognition. Mobile phones cannot identify human faces covered with masks. Hence, many people have chosen to retrain the unlocking function by using face images with masks. This reduces recognition accuracy, making it easier to cheat. Technical safety is not limited to this. Since many AI models are not explainable and come with technical risks regarding safety and robustness, when used for drug discovery, virus detection, health state decision-making, additional consideration are required.

Policies that promote and regulate AI used for fighting COVID-19 are necessary, such as the efforts for promoting the use of AI for COVID-19 from the Ministry of Science and Technology China, and the Ministry of Industry and Information Technology China, as well as the release of a personal information protection policy from the Cyberspace Administration of China, so that the development and regulation of AI against COVID-19 can be supported and implemented at different levels. Not only the collection and use of personal information needs to be regulated, accountability for the use of AI also needs extensive discussions, policy design, and action. Such efforts are not only essential for China, but also provide useful insight on global coordination.

The authors are with Research Center for AI Ethics and Safety, Beijing Academy of Artificial Intelligence.

Posted in: VIEWPOINT

blog comments powered by Disqus