Introduction
Commenting on the breakthrough in machine learning displayed by his company’s self-taught AlphaGo Zero program in 2017, Google DeepMind founder Demis Hassabis said his hope was to “harness algorithmic breakthroughs like this to help solve all sorts of pressing real world problems”. Three years later, few would argue there is a more pressing global problem than the coronavirus. So, following a number of AI-related developments in the last few weeks, what are the key takeaways?
1. Detection
As countries battle to contain COVID-19, AI can assist in detection. On 30 December 2019, Toronto-based startup BlueDot alerted clients to a number of unusual pneumonia cases emerging from Wuhan, China, over a week before the World Health Organisation (WHO) issued its own public warning. Processing 100,000 online articles each day in over 65 languages, and with an ability to draw from official medical sources as well as airline data, the system also correctly identified the cities highly connected to Wuhan, where infected individuals might travel to. When lockdown measures are relaxed, as is now starting to be the case in China, AI can offer a timely early warning mechanism to disrupt future outbreaks of the virus.
2. Research
On 16 March in a collaboration between government, industry and academia, the White House announced an open dataset of machine-readable scientific literature, known as CORD-19. Containing 29,000 full-text articles, the dataset was launched to encourage natural language processing researchers to mine the data for insights. Published through Google platform Kaggle, researchers were asked to focus on the WHO’s key questions. The dataset will continue to be updated as connections between the papers are understood.
3. Diagnosis
A number of Chinese companies have sought to deploy AI in diagnosing COVID-19, assisting overburdened medical staff. Technology giant Alibaba’s system, deployed in over 100 healthcare facilities, can detect coronavirus in the CT scans of patients’ chests with 96% accuracy, trained on images from 5,000 confirmed cases, and taking only 20 seconds to make a diagnosis where humans usually need at least 15 minutes. Huawei Cloud recently announced the development of a diagnosis service to assist doctors in assessing lung structure and functionality, which aims to distinguish between early, advanced and severe stages of the disease, also aiding diagnosis efficiency. To prevent the spread of infection, Baidu developed computer vision and infra-red sensors to monitor people’s temperatures. In use at Beijing’s Qinghe Railway Station, the system’s AI flags anyone who has a temperature above 37.3 degrees, providing safe and effective screening in real time.
4. Logistics
With resources stretched, it was announced last week that the NHS will collaborate with a number of tech giants, and London-based company Faculty, to monitor and predict critical demand. Data from the NHS’s 111 telephone service, collated with other sources, will drive predictive AI modelling and, through holistic understanding, provide the optimal allocation of resources. Amazon Web Services will provide cloud infrastructure, with Microsoft building data storage to help the project and Palantir using its analytics software to draw the data sources together and monitor:
•Ventilator use
•A&E capacity
•Patient occupancy across general, specialist & critical beds
•Duration of stays for COVID-19 patients
•Staff sickness
The AI analytics will provide staff with visual dashboards, allowing them to increase resources in geographical surge hotspots and direct patients to the most suitable hospitals, based on demand and resource availability.
5. Treatment
AI is being used to identify potential treatments for the coronavirus by:
•Reviewing existing drugs to assess their suitability
•Developing new antiviral drugs
•Developing a vaccine
Gero, a Singapore-based company specialising in AI-guided drug discovery, uses a system that analyses a database of known drugs and their molecular structure, identifying those that could disrupt the replication of the virus. Six of the potential anti-COVID-19 drugs identified have been approved, three have been withdrawn, and the remaining nine have already been tested in clinical trials for other indications. The regulatory status of these drugs means that, for most of them, it is possible for clinical trials to begin in a matter of weeks.
Separately, in the pursuit of a drug compound that can prevent the virus from infecting host cells, scientists from the US Department of Energy’s Oak Ridge National Laboratory (ORNL) recently deployed the world’s most powerful supercomputer, the IBM Summit. In only 2 days this identified 77 small-molecule compounds with the potential for use in fighting the pandemic, a task that would have taken years in a traditional laboratory setting.
Returning to DeepMind, the company has used its deep learning AlphaFold system to release structural predictions of proteins that might constitute SARS-COV-2, the virus that causes COVID-19, to the scientific community. Understanding the proteins behind the virus could conceivably assist in determining a suitable vaccine, with each day saved in achieving that goal leading to many lives being saved. An open license will allow any researcher to build on, adapt or share DeepMind's findings.
Conclusion
The use of AI to tackle the pandemic forms part of a vital global scientific collaboration that began with Chinese scientists releasing the genetic sequence of the coronavirus. In a climate where big tech faces heightened international scrutiny of its approach to taxation, competition, privacy and the spread of disinformation; the potential success of AI in helping the world overcome the pandemic also offers the prospect of reputational redemption.
Want to hear more from us? 
