Exploring the Use of Artificial Intelligence in Tackling the Pandemic
Artificial Intelligence (AI) has become a subject of increasing interest for academics, practitioners, policy makers, businesses leaders, and members of the public alike.
Exploring the Use of Artificial Intelligence in Tackling the Pandemic
Artificial Intelligence (AI) has become a subject of increasing interest for academics, practitioners, policy makers, businesses leaders, and members of the public alike. Therefore, it comes as little surprise that questions have been raised in regard to the role AI may play in tackling the global pandemic. Can we trust AI to deliver meaningful insights into and treatments for Covid-19?
In order to answer this question, it is important to first clarify what we mean by ‘Artificial Intelligence’. AI is a broad field with many different approaches and, as a result, the concept has been defined in a multitude of different ways. Nilsson (1998), for example, suggests that AI is concerned with intelligent behaviour in artifacts. This ‘intelligent behaviour’ frequently takes the form of decision-making, problem solving, learning, and so forth. In other words, activities usually considered to be defining characteristics of human beings. The term ‘homo sapiens’, after all, means ‘human being’ (homo) + the ‘wise’ (sapiens). At present, AI-supported technologies, such as those found in your smart phone or virtual assistant, can be used to support human intelligence. However, the hope (or fear, depending on your perspective) is that AI will advance to a state where it will match or even exceed our intellectual capabilities. When, or indeed if, this will be achieved remains a subject of significant debate.
A review of how AI is being utilised to address the pandemic suggests four primary applications, specifically:
- Early Detection
- Diagnosis of Covid-19
- Identification of Treatment Options
- Rapid Reviewing of the Research Literature
Let’s consider each in turn…
One of the most widely reported stories related to AI and the novel coronavirus involved the Toronto start-up BlueDot. BlueDot was founded by ‘accidental entrepreneur’ Kamran Khan, an epidemiologist and doctor who treated patients during the SARS outbreak in 2003. Deeply affected by what occurred, Khan was determined to avoid getting ‘flatfooted’ by another outbreak and wanted to ‘anticipate rather than react’. BlueDot, which employs a mix of physicians and programmers, was among the first to identify the risk from Covid-19. BlueDot’s AI system scans through over 100,000 official and mass media sources in 65 languages each day in order to detect outbreaks in real-time.
On 31 December 2019, the system alerted one of the company’s employees to a potential pneumonia-like outbreak in China’s Hubei province. The employee recognised parallels to the previous SARS outbreak and pursued further modelling of the disease, which led BlueDot to publish the first scientific paper on Covid-19, accurately predicting its global spread. Further, they used global airline ticketing data to determine how the disease would specifically migrate between countries during its early stages. It has since been reported that an automated service called HealthMap at Boston Children’s Hospital and a model run by Metabiota based in San Francisco also detected early signs of the pandemic. But once a potential pandemic has been identified, what role can AI play next?
Diagnosis of Covid-19
The technology is also being used to help to diagnose specific cases of Covid-19. One of the first projects was COVID-Net, a convolutional neural network (a form of AI that is adept at recognising images) developed by Linda Wang and Alexander Wong of the University of Waterloo and the AI company DarwinAI in Canada. COVID-Net was trained to identify signs of Covid-19 in chest x-rays using images taken from patients with various lung conditions including bacterial infections, non-Covid viral infections, and Covid-19. The European Commission also announced that it was investing in an AI tool that would allow the diagnosis of Covid-19 in less than one minute. The tool collates CT images collected from patients, the AI then searches for lesions of ground-glass opacity and tissue density (key characteristics of Covid-19 related pneumonia). If such lesions are found, an alert is sounded allowing radiologists to prioritise the reviewing of suspicious cases. In addition, the AI can compare images of the individual patient’s lungs over time and thereby track the progression of the disease. Another project involving researchers at the health science company Zoe in collaboration with Massachusetts General Hospital, King’s College London, and the University of Nottingham has developed an AI diagnostic tool that can predict if someone if likely to have Covid-19 based on their symptoms. But, if such a diagnosis is made, can AI assist in terms of identifying potential treatment options?
Identification of Treatment Options
Given the speed at which the coronavirus can spread, a major focus has been placed on repurposing existing medications to treat Covid-19. This is faster and more cost effective than developing new drugs. Here, AI has the potential to play a significant role. One of several relevant projects was undertaken by BenevolentAI, in conjunction with Dr Justin Stebbing of Imperial College Medical School. The project involved a two-stage process: first, using Machine Learning (a subset of AI) to identify two human protein targets to focus on and, second, using another algorithm to find existing medications that could hit these protein targets. This was completed in a matter of days. They identified a clear option worthy of further study: baricitinib. It is still early, but several trials have been conducted which indicate that baricitinib can be effective. According to Justin Stebbing, AI ‘makes higher-order correlations that a human wouldn’t be capable of making, even with all the time in the world. It links datasets that a human wouldn’t be able to link’.
Rapid Reviewing of the Research Literature
The final use of AI cuts across the previous three. Literature on Covid-19 has grown exponentially since the novel coronavirus was first identified. For example, the US National Institutes for Health’s Covid-19 portfolio alone listed over 28,000 articles in June 2020, more than any researcher or clinician could reasonably read. Here, AI tools can provide a (partial) solution. Largely thanks to advances in Natural Language Processing (NLP) technology, AI can sift through the research to find studies most relevant to the user and, in some cases, extract specific findings. However, they are still in the development stage, and their value is largely unproven. While the technology can’t be used to make clinical or research decisions, it still has significant value in terms of helping researchers filter huge volumes of data and identify patterns within it (Nature 2020).
Final Points of Reflection…
So, will AI make a meaningful difference in terms of tackling the pandemic? It is a difficult question to answer definitively. The projects highlighted above are still in their early stages and we will have to wait to see if they bear fruit. There are still concerns that the ‘hype outstrips the reality’ when it comes to AI. Therefore, it has been contended that the technology will help with the next pandemic, if not this one. For this to happen, however, researchers need access to more data and the means for achieving this can be contentious, particularly when they involve collecting sensitive personal information. The World Economic Forum argues that it is not AI itself that will make a difference, but rather the knowledge and creativity of the people who use it. The main advantages of AI over humans at present are speed and pattern detection. Humans, however, are much better at dealing with novelty and nuance. As a result, if AI is to be of benefit in addressing the pandemic, human-machine collaboration is most certainly key.
Dr Laura Steele