A conversation with Deepak Jha, Deputy General Manager- AIPF (Artificial Intelligence Platform), NEC Technologies India, raises many questions on the true potential of AI tools in this pandemic situation
Raelene Kambli
In the past few weeks, we all have learnt how data analytics and artificial intelligence (AI) have been applied to tackle the COVID-19 crisis. Be it identifying and monitoring the pandemic, its pattern, its epidemiology and its social-economic impact or its role in research and development of vaccines, supply chain management, and healthcare service provision. While the potential of these technologies has been tried, tested and proven in many areas of predictive analytics, process automation, genetics and more, the value it created for governments, epidemiologists, researchers, medtech developers and healthcare providers has been splendid.
In a recent conversation with Deepak Jha, Deputy General Manager- AIPF (Artificial Intelligence Platform), NEC Technologies, India, we learnt about the potential of AI-based solutions like drug-screening techniques and its potential to understand essential part of the virus (like protein structure) to determine how the virus functions which will play an integral role in designing drugs and accelerating the developments of vaccines to combat COVID-19. The conversation also delved into understanding the key loopholes and problems associated with AI application.
AI tools and its application in tackling COVID-19
Jha seems to be optimistic about these AI-based solutions. “AI is playing an unparalleled role in helping companies and researchers address the challenges of COVID-19 pandemic. We have witnessed different cases where some are focusing around detecting the outbreak by analysing the data from social media, air travel and health reports, while others are using it to find the structure of the virus to work on the vacancies and treatment. A couple of countries are using AI to diagnose the disease using CT scans and X-rays data whereas others are using drones to deliver medical supplies ensuring the zero-contact process. Healthcare bodies across the world have developed AI-based bots to spread awareness about COVID-19 and answer the queries coming from public.
Adding to the drug discovery effort answered in the previous question, it is important to highlight the high processing power of supercomputers which can help research firms with faster processing and calculation which is critical for developing a cure at the earliest.
Sharing some interesting global examples of AI-based solutions and their utilisation in COVID-19 crisis management, he cites:
Early warnings and alerts: The best example is BlueDot’s AI-based model which is among the first in the world to identify the emerging risk from COVID-19 in Hubei province and notify the customers via their insights platform with early warnings of infectious disease. BlueDot published the first scientific paper on COVID-19, accurately predicting its global spread using proprietary models and delivers regular reports including which countries reported local cases, how severely cities outside of China were affected, and which cities risked transmitting COVID-19 despite having no official cases.
Tracking and prediction: After the Zika-virus outbreak in 2015, AI-based neural network model was developed to predict its spread. These models can be further re-trained using variables coming from COVID-19 to predict the spread. Similarly, models developed to predict other flu are being re-trained now with the new data to identify the pattern. These tracking and predictions are important inputs for the government to plan and prepare for the pandemic. In China, Megvii Technology with the help of more than 100 R&D team devised and implemented an innovative AI-enabled temperature detection solution that integrates body detection, face detection and dual-sensing via infrared cameras and visible light. Similarly, in India, DRDO’s AI subsidiary has launched SAMPARC app to track COVID-19 patients by using geo-fencing and AI-based automated face recognition which helps state officials to keep a track on patients using a map-like interface that displays relevant information. This interface would also be colour-coded to depict hotspots and containment zones in an area. Additionally, a group of Mumbai-based biotechnology students and one professor has recently claimed to have developed an artificial intelligence-based tool that tests COVID-19 through voice-based diagnosis using a smartphone.
Treatment: Though we have not witnessed a vaccine or drug which can cure a COVID-19 patient, however, many research organisations are heavily relying on AI for drug discovery to fight against this pandemic. Google’s Deepmind has done the computational prediction of protein structure associated with COVID-19 using its latest version of AlphaFold system which could be useful in developing the vaccines. Similarly, some of the other research organisations have published the results using ML models to identify some of the existing drugs which can potentially be repurposed to treat COVID-19.
Crowd management and control: AI tools have provided real-time monitoring and tracking information about infected citizens and their whereabouts. While governments and municipalities use this information to identify zones based on severity and prepare policies, residents can individually use real-time information to assess the threat and prepare better for the pandemic. These tools can also use big data to analyse travel histories of individuals, and how much time each individual has spent in spots infected with the virus, and potential exposure to people that carry the virus, etc.
The challenge of huge data sets, unproven claims and more
Whilst several AI tools have certainly proven to be advantageous, many experts feel that a lot of historical evidence was already available and that companies are just building upon what we already know using AI.
Jha clarifies this saying, “Yes, we do have lots of historical health data, but we don’t have endless, huge datasets about the spread of coronavirus or epidemics that are similar enough to this. So, we don’t know enough to learn exclusively from historical data. In addition to the above, some of the other challenges are:
- Unavailability of feedable data to AI systems
- Inconsistencies in the information provided by governments and various news outlets
- Data privacy issues with sharing personal health information data
- Health data being bound geographically due to the laws and regulations of various countries
Now, the question that lingers in mind is that isn’t it important to be wary of the broad, unfounded claims about what AI can do? And isn’t it important to question whether companies really have the data and expertise to ensure that the application has proven outcomes?
“Epidemic at this scale was not seen earlier, especially in the last few decades. Humankind or the healthcare sector has not experienced anything like this before. So, we can’t actually rely exclusively on the past data or inputs. We need to be really careful while relying completely on the results of AI (as it’s not backed up by enough data at this moment) as the applications are relatively new and aren’t robust enough to give us the right answer every time. The potential of AI cannot be denied, but with a novel disease like COVID-19, not having enough data will lead to false negatives. It can’t replace human physicians, but when combined with physicians and subject matter experts, can definitely drive innovation towards a clinically-viable solution in a shorter timeframe,” Jha warns.
Some of the key points to note are:
- AI can be useful but it’s not ready to replace human in the battle against COVID-19
- Lack of datasets about the spread of coronavirus and epidemics that are similar in nature. Learning and accuracy cannot be predicted in the absence of historical data.
- Challenges related to AI such as accuracy and biases always act as hurdles in implementation
- Misjudgement and overconfidence of AI capabilities could be counterproductive and the expense of proven interventions programme
The worst example of AI application
To justify these points, here are some of the worst examples of AI that’s been touted as effective in response to COVID-19, identifies Jha. According to him, these applications and tools need to be scrutinised by government bodies before they are open to the market.
- People claiming about attaching various sensors to the drones and detect everything about the virus. Companies are claiming that drones can be used not only for thermal imaging to detect fever, but also to get a sense of respiratory rate and heart rate.
- Justifying a substantial surveillance opportunity in the problematic areas which may affect people’s behaviour
- Mortality rate prediction are prone to bias and gives a potentially harmful and a different perspective
- One of the examples of AI bias which we have seen in the case of Optum algorithm (a mortality-risk algorithm) which is biased against African-Americans for reasons that weren’t relevant to health, but more relevant to finances and socio-economic status. Same could be possible in case of COVID as well.
The points raised by Jha in this conversation ring a bell of caution. Although being a tech expert and his business thriving on AI application, he does believe that AI applications and tools need vigilance. This also brings us back to the point of the much-needed regulations in the stream of Health IT and its various applications. We are much aware that AI and Deep Learning (DL) have the potential to propel the healthcare industry into the future, with great experimental results and a variety of critical applications such as improved cancer diagnosis, genetic healing and more. Moreover, in these times of crisis, technology has brought in a lot of hope. To prepare the industry for a whole new world dominate by technology without exposing patients to unnecessary risks, regulatory entities will need to make drastic changes to their frameworks to ensure they are better equipped to deal with Software-as-a-Medical-Device solutions. FDA has already begun considering it. When will Indian regulatory bodies bring this issue to the fore?
- Advertisement -