Why hasn't AI helped so much in the Covid-19 pandemic?

The world is facing the biggest health crisis in decades, but AI, one of the world's most promising technologies, has yet to play an important role in the way some people might hope. Well-known AI laboratories such as DeepMind, OpenAI, Facebook AI Research and Microsoft remain quite silent while the Covid-19 translation has spread globally.

'The silence of the AI ​​is remarkable. AI is great, and the technology will be useful someday, but it's no surprise we're back to techniques that were tested during the pandemic. '' Neil Lawrence, former chief executive machine learning director at Amazon Cambridge, now a professor of machine learning at Cambridge University.

Experimental techniques include old statistical techniques and mathematical models to create epidemiological models, predicting how the disease will spread in the community. At the moment, this technique is more useful than areas in AI such as intensive learning and natural language processing.

Of course, there are still a few useful AI projects going on here and there in the world. In March 2020, DeepMind announced it had used a machine learning technique to describe in detail the structure of the six proteins involved in the SARS-CoV-2 virus. Israeli-based startup Aidoc is using AI images to mark abnormalities seen in the lungs. Another UK startup also uses AI to search for drugs to treat disease.

A facial recognition system developed by British company SCC has also been adapted to detect people infected with the virus instead of using it to identify terrorists as before. Exscientia Company in Oxford (UK) is screening more than 15,000 drugs to consider the ability to treat cases.

However, the role of AI in dealing with the Covid-19 epidemic is not profound, not as much different as some have predicted. AI will not soon bring the world out of its current state. 'This seems to show how well the AI ​​has been hyped,' says Prof Lawrence.

AI researchers rely on large amounts of data to train their algorithms, but currently there is not enough reliable data about the SARS-CoV-2 virus to do that. 'AI learns from a large amount of manually labeled data, which is a time-consuming and expensive job. In addition, it takes time to build, test and deploy AI in the real world. As the world changes, as it is happening, the challenge for AI lies in how it is possible to collect enough data to learn, build and deploy technologies quickly to create impact. '' Catherine Breslin, a machine learning consultant who worked on Amazon Alexa, said. Breslin agrees that AI technologies have their own roles, but 'they are not silver bullets', the only powerful weapon to directly end the epidemic.

The AI ​​community is currently trying to think about ways to make it more useful. Last week, Facebook AI announced a number of partnerships with scholars across the United States. Meanwhile, Demis Hassabis, co-founder and CEO of DeepMind, is helping the Royal Society, the world's oldest independent science academy, in a new multidisciplinary project called DELVE. Data and Learning on Epidemiology).