Will Artificial Intelligence save us from the Coronavirus?
Every single day I see project posts on Linkedin- someone is trying to predict the number of cases the next day or the next month while someone else is working on the automated machine learning models for the diagnosis of COVID-19 infection in chest x-ray images. There are thousands of dashboards being generated each day claiming to give useful information and ‘insights’ on data. Companies are flooding the market with AI-based fever detection cameras, and gadgets. Many are lauding AI as our omniscient secret weapon, one that spots new outbreaks, diagnoses cases, prioritizes the patients most in need, monitors patients at home, reads the scientific literature, and is on its way to creating a vaccine. Is this another myth-buster article? Or does AI really have a huge role to play?
As the world grapples with COVID-19, every ounce of technological innovation and ingenuity harnessed to fight this pandemic brings us one step closer to overcoming it. AI has indeed helped researchers ingest large volumes of data and provided them with useful inferences. Closedloop, an AI start-up, is using their expertise in healthcare data to identify those at the highest risk of severe complications from COVID-19. Researchers at the Chan Zuckerberg Biohub in California have built a model to estimate the number of COVID-19 infections that go undetected and the consequences for public health, analyzing 12 regions across the globe. AI is also helping organisations adjust to the new conditions by meeting the demands for tools created during the lockdown. [1], [2] and [3] have several other applications of AI that emerged this pandemic.
Another interesting direction of progress has been in the identification of compounds that could be used to create vaccines or even treatments for the virus. BenevolentAI, a UK based AI company that uses ML to fight the world’s toughest diseases, turned its models toward understanding the response to the coronavirus. They used machine learning to help derive contextual relationships between genes, diseases, and drugs, leading to the proposal of a small number of drug compounds that could potentially inhibit the progression of the coronavirus. Apart from this, we have AI-powered chatbots, blockchain platforms, UV disinfection robots, etc.
While there are undoubtedly innumerable ways in which AI has proven to be useful, it is no panacea. Following up from our previous post on rumors during the Coronavirus pandemic, is yet another article on healthy skepticism about AI, and how you shouldn’t believe every average Joe’s post on ‘useful insights’ obtained from some data. Having personally worked on several AI projects in healthcare, I can comment that these models are poorly calibrated. The predictions they make might not be the true representative of the true underlying statistics. As mentioned in [3], the most important rule of thumb is to check whether the results come from subject matter experts. [2] puts this as “To someone with a hammer, every problem looks like a nail”, and this is certainly a trend with several AI enthusiasts working with COVID data. Posts on models that are 98% accurate at some task should not bias you into unreasonable conclusions. Data is always dependent on its context, and it takes expertise to draw meaningful inferences from it. AI works well with predictions on data that clearly follows a trend. These are uncertain times and most of these models have no information to predict what’s going to happen next. Data modelers make several assumptions while creating dashboards and we cannot make decisions or conclusions based on these predictions. Thus, if you are someone planning their next vacation based on an ‘intelligent’ prediction that says travel restrictions will be lifted by October, you now might want to consider giving it a second thought.
To believe or not to believe then? you may ask. Based on all of the current evidence of systematic bias, lack of data, assumptions, and other factors, you should pause and evaluate all the credibility of the model before you install the fancy AI camera that detects fever in your grocery store or books your flight tickets to XYZ land. That being said, the promise of AI is fast heading on its way to fruition. The near-future impact of AI on some of these applications is unfathomable and quite likely brilliant. It has great potential, but its pros need to be hedged in a realistic understanding of its cons.
