Select your language: EN JP

AI's Role in fighting COVID-19: Sifting through the hype

Cole McCollum, Analyst

Many have identified AI and machine learning as potential tools in the fight against COVID-19. However, as in most discussions involving AI, there's significant hype around the topic, making it difficult for organizations to determine where to focus their efforts. In this blog, we examine various AI use cases that promise to help with the health concerns surrounding COVID-19 and analyze their technical feasibility using Lux's AI Framework.

Broadly, AI use cases for COVID-19 can be broken up into the following four categories, as shown in the figure below: disease prevention, early detection of disease (outside of the healthcare system), provision of care (delivered within the healthcare system), and research and development for new diagnostics, treatments, and vaccines. This list of use cases is not meant to be comprehensive but rather shows a sampling of prominent use cases that have garnered attention.

Lux Research Covid-19 ai
Figure 1: The landscape of AI use cases for COVID-19.

To determine the technical maturity and feasibility of each of these use cases, we use the methodology developed in the insight "A roadmap for AI in healthcare." The methodology scores each application on two axes – the amount of reasoning required and the environmental complexity of the application – as detailed in Figure 2.

Lux Research Covid-19 ai figure 2

Figure 2: Amount of reasoning required (vertical axis) and environmental complexity (horizontal axis).

The framework can be divided into four quadrants as follows. Quadrant 1: Applications in this quadrant tend to be more mature and focused on scaling basic human recognition tasks. Quadrant 2: Applications in this quadrant can have a broad impact, particularly in applications that scale the recognition capabilities of an expert. Quadrant 3: Applications in this quadrant show the possibilities of long-term reasoning in narrow environments. Quadrant 4: Applications in this quadrant can have the broadest impact through long-term reasoning in complex environments but can often be limited by the capabilities of today's tools. Below are the results of mapping out the AI for COVID-19 use cases on the framework.

lux research covid-19 ai figure 3

Figure 3: AI framework mapping out COVID-19 healthcare use cases.

Results of the AI framework analysis:

Compared to AI healthcare applications in more traditional settings, which are more evenly distributed in Quadrants 2 and 4 of the framework, AI in COVID-19 use cases more commonly fall into Quadrant 4 of the framework (complex and dynamic settings where expert reasoning is required). This placement is due to the limited historical datasets available, the heterogeneity of the patient population affected, and the novel nature of the virus and associated disease. Without comprehensive datasets available to fall back on, expert reasoning capabilities are often required, and the strength of pattern recognition in today's AI technologies cannot be relied upon too heavily without a significant risk of inaccuracy and bias. For example, although several organizations have claimed to develop machine learning systems for identifying COVID-19 in X-ray/CT scans, the current performance of these systems will likely be overstated outside of controlled environments until larger datasets are collected. That is not to say that some use cases in Quadrant 4 do not offer value to scale or speed up human capabilities; however, adoption should primarily be limited to use cases with little direct patient risk and those that can afford some inaccuracy (e.g., helping to accelerate aspects of drug development with a human-in-the-loop).

New call-to-action

Use cases that fall into Quadrants 1, 2, and 3 of the framework are a strong technology fit for today's AI systems. Such use cases require less personalization or accuracy at the level of an individual result (e.g., providing population-level trends through digital biomarkers or social distancing detection) and can be trained using a more generic dataset or utilize a simple set of rules to segment populations (e.g., identifying at-risk individuals using government guidelines). As economies slowly begin to open up across the world, companies should look for opportunities in these quadrants to apply AI in disease prevention and early detection systems, which will remain a critical need in the months and potentially years to come. Quadrant 3 also contains interesting opportunities to help accelerate R&D efforts. For example, AI is helping researchers wrangle the vast amount of literature published on COVID-19 on a weekly basis. Already, companies like Sinequa and Primer AI are offering free search engines and text mining capabilities on the available scientific literature (Lux members can explore Sinequa's here and Primer's here).

Key takeaways and outlook:

AI is not a silver bullet for health-related COVID-19 use cases and is in some cases currently overhyped, due to the heterogeneity of the patient population affected and the lack of historical and comprehensive datasets for machine learning systems to learn from. That said, there are still significant opportunities to apply AI in scenarios that can afford some inaccuracy or in use cases that provide more generic insights or rely on simple rules to offer personalized results. Additionally, while the most advanced techniques like machine learning and deep learning may gain a disproportionate amount of attention, more foundational data analysis and data visualization techniques still play an important role in use cases like infection forecasting.

For COVID-19, companies should consider focusing on AI applications in disease prevention, early detection, and R&D rather than the direct provision of care given that the requirements are a better fit for today's AI tools and that the use cases will have a longer-term need even past the peak of the crisis. Those interested should also begin investing in AI tools to fight the next pandemic, as funding for global pandemic preparedness will likely increase significantly as a response to COVID-19. As AI continues to progress over the coming years, and researchers improve its ability to reason and operate using smaller datasets, the technology will become even more essential in fighting the next pandemic.

 

 

FURTHER READING:

Additional COVID-19 Resources

- Blog: How to Prepare Innovation Teams for a Coronavirus Driven Downturn

- Blog: Battling COVID-19 With Transformational Tech: Singapore Case Study

- Download the Executive Summary: Artificial Intelligence: A Framework to Identify Challenges and Guide Successful Outcomes 

More Resiliency Resources    Schedule Your Demo