A software to predict disasters. The future according to Microsoft

Can you predict the future based on past events? In some cases it seems that this is the case, since there are patterns of behavior that can be repeated under certain circumstances. This is the belief of Microsoft and the Technion (Israel Institute of Technology), which are developing a technology capable of predicting disasters, based on information from the past.

Using new software that uses articles from the New York Times and up to 90 different sources available online, you can issue warnings about epidemics, natural disasters, and violence , thus being able to avoid or alleviate its consequences.

Regarding the New York newspaper, 22-year-old articles are used, and other sources include: DBPedia, WordNet and OpenCyc. In the words of Eric Horvitz, co-director of Microsoft Research, the system could one day help NGOs and others to be more proactive in fighting disease outbreaks or other problems.

The system has provided amazing results when tested against historical data. For example, from reports of droughts in Angola in 2006, an alert was deduced about possible cholera outbreaks in the African country, since previous events had taught the system that cholera outbreaks are more likely in years after droughts. .

In similar tests on disease, violence, and number of significant deaths, system warnings have been between 70% and 90% correctof the occasions. These are very high percentages.

Horvitz has stated that performance is good enough to suggest that a more refined version could be used in real world environments, to help in tasks of planning and preparation of humanitarian aid in case of disasters.

The use of cross-referenced information from various sources provides valuable context that is not available in a news article, and is necessary for find out the general rules of the events that precede the others.

For example, the system could infer connections between events in Rwanda and Angolan cities, based on the fact that both countries are are in Africa, have a similar GDP, and other factors. This approach led the software to conclude that, in predicting cholera outbreaks, one should consider the country or location of a city, the proportion of land covered by water, population density, GDP, and whether there had been a drought the previous year.

The idea of ​​establishing predictions based on information from the past is not new, there are already companies that use similar techniques based on information and online statements, whose clients include government intelligence departments.

While writing this article, I remembered hearing on occasion that the use of cross-referenced information from various sources is also used to make predictions about the behavior of stock markets.

Microsoft has no plans to commercialize this research, which I welcome because information is an extraordinary source of power, and tools can be fantastic or diabolical, depending on who wields them.

The project will continue, including more digitized newspapers and books, further refining the system, which will be able to make better predictions reliable.

Via | MIT Technology Review Image | Amit Chattopadhyay, Michael Gray, Ciprian Popescu In Xataka Windows | The future according to Microsoft