In the coronavirus spread scenario, artificial intelligence cannot correctly predict the scale of the pandemic and provide recommendations, using only the data on past epidemics, the expert says
Specifics of the artificial intelligence technologies based on big data and machine learning, which are widespread today, are ill-fitted to predict the coronavirus spread and to develop optimal management decisions; the scientists need to look for other methods, says Maksim Fyodorov, head of the Skoltech’s science and engineering big data computation center.
The machine learning-based AIs must collect and analyze a large amount of data first. Even then, their algorithms are only suited for solving "template" tasks with a minimum of unknown variables. In the coronavirus spread scenario, artificial intelligence cannot correctly predict the scale of the pandemic and provide recommendations, using only the data on past epidemics, the scientist says, adding that the ongoing situation is far from being a "template."
"The entire perceived ongoing artificial intelligence revolution is merely a breakthrough in one narrow technology: the machine learning-based AIs. These solutions work well with [multiple similar] tasks, such as recognition of faces that have specific number of eyes and ears, for example. The coronavirus is a task of a different nature, which involves unique events and a lot of new variables, so an AI, based on historic data, works poorly here because it needs a lot of data of similar kind for learning. This pandemic story has shown that we must look for other technologies and approaches," Fyodorov said.
The expert believes that methods of mathematical modelling could prove relatively effective in making management decision. In these methods, the scientists "play" different event scenarios, calculating the results on supercomputers in detail.
"These methods were popular at the time, and we should return to this scenario approach, but using a new algorithmic and computational base. [We should] play scenarios, including the most implausible ones. Having this library of scenarios, the scientists will be able to access it promptly in unforeseen situations. And then, one must turn on the human mind," he said, adding that the science evolves like a spiral.
The machine learning-based AI, in its modern shape, will be able to "learn" from the coronavirus spread data and be useful to humanity only if the ongoing pandemic repeats itself in the future, Fyodorov says. He also believes that there is hope that another epidemic could be avoided.
In late December 2019, Chinese officials notified the World Health Organization (WHO) about the outbreak of a previously unknown pneumonia in the city of Wuhan, in central China. Since then, cases of the novel coronavirus - named COVID-19 by the WHO - have been reported in every corner of the globe, including Russia.
On March 11, 2020, the WHO declared the coronavirus outbreak a pandemic. According to the latest statistics, over 2,484,000 people have been infected worldwide and more than 170,000 deaths have been reported. In addition, so far, over 652,000 individuals have recovered from the illness across the globe.
To date, a total of 52,763 coronavirus cases have been confirmed in Russia, with 3,873 patients having recovered from the virus. Russia’s latest data indicates 456 fatalities nationwide. Earlier, the Russian government set up an Internet hotline to keep the public updated on the coronavirus situation.