The Power of AI in a Pandemic

We are in uncertain times. With COVID-19 affecting people all over the world, many are striving to find creative solutions and tools to solve some of the problems this pandemic has presented. 

To combat this global crisis, countries, governments, and hospitals are using a powerful tool: artificial intelligence. Artificial Intelligence (AI) involves developing and programming machines that can simulate human intelligence in the real world. AI is being used to provide critical predictions and insights about the ongoing COVID-19 pandemic. This has guided some of the decisions taken by policymakers, medical specialists, and even people. Let’s take a deep dive into how the incredible capabilities of AI are being harnessed to save lives! 

Figure 1: The power of AI in a pandemic. Reprinted from Ref [1]. 

Power of COVID-19 Predictions 

Did you know the first signs of COVID-19 were detected by an AI powered, early warning system? On December 30th, a Canadian-based startup known as BlueDot became the first organisation to come across a cluster of pneumonia-like cases in Wuhan, China, through AI [2]. This led to the identification of a novel virus that came to be known as COVID-19. 

Employees of BlueDot designed a software that uses AI-driven algorithms to detect areas where an outbreak of any virus may occur. An algorithm is a set of instructions that a machine executes. These AI-driven algorithms were programmed to mine ‘mainstream news, online content, and other information channels in multiple languages to provide early warnings’ [3]. By sifting through and analysing this online content, the software was able to identify and map out the spread of COVID-19, which was just emerging at the time. These epidemiological patterns that the software was able to detect helped people stay away from possible ‘danger zones’ that they were previously unaware of. 

As the pandemic continues to affect countries all over the world, the BlueDot system is still being used to provide valuable information.  Many other companies have also designed similar softwares. These softwares have been provided with even more tools and data sources to use, since we are now several months into the pandemic. For instance, the AI-based systems are not only using mainstream news to make predictions, but also air travel data that has recently been made available. Air travel data can show the most common locations people have been traveling to over the past few months, and this can provide clues about where possible outbreaks may occur in future. This process of analysing historical and current data to make predictions about future outcomes is known as predictive analytics. Predictive analytics is a part of AI and can help extract more value from the data we already have. 

Despite these benefits, there are some concerns about relying on AI to make predictions. It is believed that as the pandemic’s scale is growing, predictions about the pandemic become more generic, and possibly more inaccurate [4]. This is because finding reliable and consistent data is challenging during a crisis like this. Numbers may be exaggerated or downplayed by different media outlets or programs, which can throw off the AI-driven system. Another issue is that the AI system is very limited in its understanding about what is going on during the pandemic. For instance, it does not know whether people are practicing good hygiene at home [4]. Without this background knowledge, the AI system may assume an area has a low risk of a COVID-19 outbreak. This prediction can be inaccurate especially if people in that area are not taking the right precautions, leading to an unforecast spike of COVID-19 cases. From this, we can conclude that predicting the course of a pandemic is definitely a complex process that involves various factors. This means that we have to acknowledge the credibility of any AI-based early warning system before using its output to forecast the future of the pandemic. 

Diagnosing COVID-19

AI also helps with the diagnosis of COVID-19. Many COVID-19 patients have aggressive pneumonia-like infections in the lungs. Radiologists struggle to quickly analyse the numerous CT (Computed Tomography) or X-ray scans of the patients’ lungs and determine whether they show the signs of COVID-19 [5]. To address this, researchers are developing AI-based software. The software can serve as an extra hand for radiologists by combining CT or X-ray scans and flagging the ones that seem to resemble COVID-19 infections [5]. This AI-driven process saves time and ensures that all patients receive the care and attention they need. 

How is this possible? Before the AI software can even be used by radiologists, it needs to be trained. This is achieved by feeding the system thousands of CT and X-ray scans from COVID-19 and lung disease patients around the world. Then, the AI attempts to identify all the scans that belong to COVID-19 patients from the training data. At first, the software does make mistakes because it has not had much experience analysing and sorting the images. For instance, it may inaccurately classify a bacterial infection in the lungs as a COVID-19 infection. As the system is exposed to more training data, it starts detecting common features and patterns in the images that represent COVID-19 cases. Over time, the AI-powered software’s performance improves, and it is able to make more accurate predictions about whether a CT or X-ray scan shows signs of COVID-19. This ability to draw on its previous experience with the training data and evolve is known as machine learning. Machine learning is a branch of AI where systems automatically make observations from the data presented and use this knowledge to achieve better results. 

Despite these benefits, there are some downsides to using machine learning for diagnosing COVID-19. Ultimately, the quality of a system’s output is determined by the quality of its input. Sometimes, the CT and X-ray scans are imperfect. They may be cloudy or contain glitches that can throw off the AI software [5]. This means that images need to be carefully pre-processed before they are fed into the software, but sometimes makes this process unfeasible. Therefore, the AI software needs to be constantly tuned so that it can handle these flaws in the training data better. Another downside is that the AI software cannot explain why it classified the image the way it did [5]. This means that medical specialists and radiologists cannot understand the full story behind the system’s output. Currently, researchers are working to address these issues. Hopefully, in the near future, the AI software will be developed enough to help medical specialists save more lives. 

COVID-19 Conversations With Chatbots

Another application of AI is virtual health care assistants/chatbots. These interactive and reliable systems have helped people all over the world stay informed and take the necessary precautions. Some of the functions of the virtual health care assistants include answering questions about COVID-19, checking if the individual possesses symptoms, and providing personalised advice based on the conversation with the user [6]. The machine’s ability to successfully interact with an individual is enabled by natural language processing (NLP) [6]. NLP is a subset of artificial intelligence that allows the machine to comprehend the user’s words (verbal or written). It does this by converting the user’s response into a form it is familiar with. In this case, NLP allows the machine to determine what the user wants to know about COVID-19 so that it can provide the right information. 

However, NLP does not ensure that the virtual healthcare assistants can successfully answer the user’s questions all the time. Sometimes, it may not be able to understand what the user is asking because of how the question is being asked. Why does this occur? Well, like most AI models, the virtual assistants are trained on a variety of examples so that they can answer the user’s questions. But, these examples may not encompass all speaking and writing styles. Therefore, the virtual assistant is not prepared for all the different ways a piece of information or advice may be requested. This problem is not only restricted to COVID-19 virtual healthcare assistants but is something that other interactive systems face as well. Possible solutions to the issue include bringing more diversity in the training examples or even developing more sophisticated algorithms. But for the most part, the virtual healthcare assistants and chatbots seem to be faring well in bringing more clarity and awareness to the public about COVID-19. 

Searching for Solutions to COVID-19

Finally, scientists are using AI as a way to streamline the search for COVID-19 vaccines and treatments. Scientists review research papers and predict whether they can be implemented as a realistic and successful solution to this crisis. This is important because the government can quickly allocate funds to those studies that are believed to turn out successfully [7]. The problem is that the process of manually going through research papers is time-consuming and meticulous especially when we are in urgent need of a feasible vaccine or treatment plan. This is where AI comes in to help scientists and policymakers tackle this issue with its phenomenal capabilities!

Researchers are using AI to develop an algorithm that can sift through research papers, analyse the information, and predict a study’s ability to be replicated. Replicability is an important aspect of any research because it establishes the validity of a study’s results. AI does this by picking up on any word-choice patterns that can help it rate the replicability of a research paper [7]. The model also takes into account the relevance and quality of the work during the process. This is very beneficial for researchers because sometimes, the human eye can miss the small details. The AI algorithm ensures that most of the patterns and minute details are identified during the process. Even with all this workload, the AI model completes the task quickly with an impressive turnaround time of 5 minutes! But, like any other innovation, there are some possible challenges that the AI model does present to researchers. 

One of the concerns is that the AI model may not assess a paper with the same rigor as scientists do when using traditional methods [8]. This may let a few unnecessary research papers make the cut. To combat this, scientists are constantly reviewing the output of the model and comparing it with the analysis they made on the research papers [8]. Another concern is the possibility of bias being built into the model. An example of bias is where factors like race, age, and gender may impact the AI model’s results and cause it to favor a certain group over another. This may occur if the training data contains bias because the AI model only learns from what you give it. Despite these possible issues, the AI model has an accuracy of around 70% [8]. Overall, this AI algorithm will help us ensure that we invest enough time, energy, and resources in the most promising studies related to tackling COVID-19. 

The Importance of AI

To sum it all up, AI is a powerful amplifier for human talent that is enabling us to uncover new trends and possibilities during the COVID-19 pandemic. This information is proving to be very valuable for the public, doctors, and policymakers all over the world. All of these different applications also show that artificial intelligence is drastically shaping our world and will continue to do so in the future as innovators make great strides in the field. 

The road to a sustainable future is definitely a bumpy one. But, AI can serve as a 'sixth sense' that can help us achieve that by efficiently breaking down vast amounts of data into meaningful, applicable knowledge. This can guide our future response efforts to crises such as the COVID-19 pandemic and ensure humanity will be able to adapt and thrive! 

References

[1] Geralt (2018, May 8). “Artificial Intelligence Brain,” Pixabay, [Online]. Available: https://pixabay.com/illustrations/artificial-intelligence-brain-think-3382507/ [Accessed 19 September 2020]

[2] E. Niiler. (2020, Jan. 25). “An AI Epidemiologist Sent The First Alerts Of The Coronavirus.,” Wired, [Online]. Available: https://www.wired.com/story/ai-epidemiologist-wuhan-public-health-warnings/ [Accessed 31 July 2020].

[3] Anon. (2020, Apr. 23). “Using artificial intelligence to help combat COVID-19,” OECD, [Online]. Available: https://www.oecd.org/coronavirus/policy-responses/using-artificial-intelligence-to-help-combat-covid-19-ae4c5c21/#section-d1e185. [Accessed: 31-Jul-2020].

[4] W. D. Heaven. (2020, Mar. 12). “AI could help with the next pandemic-but not with this one,” MIT Technology Review, [Online]. Available: https://www.technologyreview.com/2020/03/12/905352/ai-could-help-with-the-next-pandemicbut-not-with-this-one/. [Accessed: 31-Jul-2020].

[5] S. Obeidat. (2020, Mar. 30). “How Artificial Intelligence Is Helping Fight The COVID-19 Pandemic,” Entrepreneur, [Online]. Available: https://www.entrepreneur.com/article/348368. [Accessed: 31-Jul-2020].

[6] D. Levin. (2020, May 20). “Using artificial intelligence to diagnose COVID-19,” Medical Xpress, [Online]. Available: https://medicalxpress.com/news/2020-05-artificial-intelligence-covid-.html. [Accessed: 31-Jul-2020].

[7] Northwestern University. (2020, May 4). “AI tool speeds up search for COVID-19 treatments and vaccines,” Medical Xpress, [Online]. Available: https://medicalxpress.com/news/2020-05-ai-tool-covid-treatments-vaccines.html. [Accessed: 31-Jul-2020].

[8] J. Miller. (2020, May 11). “Artificial Intelligence transforms search for COVID-19 vaccines, cure,” Healio, [Online]. Available: https://www.healio.com/news/primary-care/20200511/artificial-intelligence-transforms-search-for-covid19-vaccines-cure. [Accessed: 31-Jul-2020].