On January 9, 2020, the World Health Organization (WHO) warned the world: In the Chinese province of Hubei, an unusually large number of patients with persistent pneumonia are hospitalized. There are many indications that they were infected with a new virus. This may have jumped onto people at a fish and poultry market in the provincial capital Wuhan. The US epidemic center for disease control, CDC, had issued a similar outbreak announcement three days earlier. The warnings were mainly based on public information from the Chinese health authorities. The artificial intelligence of the Canadian start-up BlueDot had tracked down the epidemic significantly earlier than the health officials in Geneva and Washington: on New Year’s Eve 2019.
Data from around the world
The algorithm of the epidemiological early warning system collects data from regional news reports, official statements, health databases as well as veterinary and phytosanitary forums around the clock and worldwide. For a fee, BlueDot reports to its customers, including many western health authorities, where new sources of danger are emerging. With the help of flight ticket data, the system also predicts where the disease will spread in the near future. In the case of the new virus in Wuhan, the data-learning system tapped Bangkok, Seoul, Taipei and Tokyo. In retrospect, this turned out to be 100 percent correct.
We have all read and learned a lot about epidemics in the past few months. Even as a medical layperson, we now know that every day counts when it comes to containing a dangerous virus like Covid-19. The earlier doctors and health authorities identify the first cases of infection, send them to quarantine and thus interrupt the chains of infection, the fewer people have to go out of their lives before their time. Every day counts in order to prevent the exponential growth curve of new infections country by country, which then almost inevitably takes its ominous rise and unfolds the enormous destructive force that we feel so violently almost everywhere in the world today. Had health authorities, politicians and finally all of us listened to the warning from BlueDot-AI, drawn the right conclusions and translated it into consistent action. Of course that’s true. And of course things are more complicated here too.
The pandemic would hardly have stopped a week more reaction time. Artificial intelligence and machine learning are not a silver bullet against the pandemic. You can only support us in our fight against the virus if we use it intelligently. Data-rich early warning systems like the one from BlueDot are just one tool in the information technology toolbox. AI systems in Chinese hospitals can already use x-rays of the lungs to determine with a high degree of certainty whether a patient is infected with Covid-19. The system can then provide doctors with valuable information on how likely pneumonia is likely to develop and which therapy they should use.
Search for vaccines
The recently published, anonymous Google movement data shows health politicians in which regions the initial restrictions are largely adhered to, and where people break them en masse. From this, they can derive health policy decisions. AI-powered search engines are used to identify promising drug candidates in medical databases. Machine learning can also help and accelerate the development of a vaccine, for example through improved DNA analysis. But it is also important to know that the actual battle is fought by pharmacologists in petri dishes, not by algorithms in an artificial neural network.
Much is still unknown
Many of us medical laypersons currently look at the infographic preparation of the infection data of the Johns Hopkins Center for Systems Science and Engineering every day. Looking at the graphs gives us the feeling of being able to observe the health status of the world in real time. At the same time, in the eye of the virological storm, epidemiologists are just learning that even in times of big data, we live in bitter data shortage on many key epidemic issues. We still do not know exactly how the virus is transmitted, whether the manipulations on the shopping trolley are dangerous and in which situations a mouth guard reduces the risk of transmission. We don’t know why mortality rates are so different in different countries. Is this because of different counting methods or because the population in some countries is better immunized by other flu waves? We don’t know who dies with and who dies because of Corona.
Covid-19 was a worst-case scenario for data-based decision making, particularly in the early months of the crisis. The unusually long incubation period of the virus means that infected people go about their everyday life undetected and infect many others. The corona symptoms are too ambiguous, too many infected people do not develop any symptoms at all, so that the majority of the cases can be grasped by simple diagnostics, let alone by a smartphone. In some countries, apps are in use that use traffic lights to indicate how high a person’s risk of being infected with Covid-19. “Red” means quarantine, “green” means move freely.
In many respects, these traffic lights are more of a non-statistic, more apparent evidence for informed decisions than real evidence. Because apart from those who tested positive for Covid-19 and those who have had the disease and are immune, all other citizens would have to be set to “yellow”. GPS data on smartphones have an accuracy of around two meters and do not show a third dimension. They cannot be used to clarify whether people keep a safe distance and whether they are located one above the other on different floors in a high-rise building. The pronouncements with which Chinese authorities advertise the efficiency of social tracking systems as the health variant of social scoring systems should also be treated with caution.
Covid-19 tests were developed in record time, but the record time was still too long to end the blind flight of health authorities in time in many countries without secured infection data. Until April, the availability of tests was too low, so that no country was tested statistically representative, that is, a sufficient number of random samples across the population, including antibody tests of the already immunized.
But only in this way would scientists, doctors and politicians have had a real overview of the state of health in the country. The crucial lesson here is that knowing what you don’t know is helpful evidence. In future crises, health politicians will hopefully recognize earlier what information data scientists need to feed their models in a meaningful way and what measures they can use to generate the necessary data earlier. This can also mean prioritizing statistical testing and quarantining more suspicious cases for longer.
Of course, it is still too early to take stock of what we can learn from this health crisis for upcoming epidemics. But a lesson related to the use of AI can already be recognized: Systems learning from data are always particularly helpful if they have learning data from similar situations available. One of the essence of major crises is that we are dealing with a largely unknown situation. There is a structural lack of data for the big questions that have to be clarified under time pressure. Only when we understand this will we use digital tools as what they are good for: reinforcing human intelligence and effectively implementing socially sensible measures.
With these tools, we don’t have to send people suspected of having corona to the doctor in the waiting room, but can use telemedicine to decide whether a saliva sample should be sent in. We can send palliative medications via an online pharmacy, use a fitness band for medical checkups and, in an emergency, a doctor can be referred online to a suitable hospital, which reports its capacities to a central system in real time.
The difference is not made by artificial intelligence, but by digital competence when using the available tools.
Your opinion counts! Write us! Dear readers of this newspaper, what do you think of this article? Send us your opinion! “Mannheimer Morgen”, debate, PO Box 102164, 68021 Mannheim, email: [email protected]
© Mannheimer Morgen, Saturday, April 11th, 2020