“Tell me how the house you live in looks and I’ll tell you how much chance you have of having a car accident.” It seems a meaningless phrase and instead has its own logic, to listen to a new study – quite surprising! – by Łukasz Kidziński of Stanford University (Usa) and Kinga Kita-Wojciechowska of the University
Their research was based on the observation of images of Google Street View, and it was intended to find a possible correlation between the type of house a person lives in and the probability that, the same person, he or she is involved in a car accident.
The Google Street View Theorem. The whole thing started by analysing the data of 20,000 people who had signed a car insurance in Poland between 2013 and 2015: researchers and their team searched for the address of their home in Google Street View for each insured person and, after Finally, they cross-referenced these data with those provided by the insurance company to extrapolate any correlation with the likelihood that an insured person would file a complaint.
An image taken from the study that identified a link between the type of dwelling and the probability of incurring a road accident. © University of Warsaw
Surprise effect. The results? Surprising. Kidziński and Kita-Wojciechowska have discovered that the type of house in which an insured person lives is a good predictor of the probability that he will file a claim for insurance: •The visible features Even knowing the state of the dwelling could increase the probability that an insurer will be able to identify a customer who is likely to complain. And according to researchers, the current study is only a “test”: its accuracy could be improved by using a greater amount of data and using a more precise analysis.
What about privacy? Using data such as the facade of the house where you live – with current laws – goes far beyond the limits of what can be granted insurance against those who choose it. “The consent given by customers to the company to store their addresses does not necessarily mean consent to store information about the appearance of their homes,” the researchers say. And the correlation between home and accidents could open a Pandora box: other companies could also take advantage of it, with good privacy “The insurance industry could be quickly followed by banks, since there is a proven correlation between insurance risk models and credit risk score
Vote for Google Street! This is not the first time Google Street View data has been used to retrieve sensitive and/or confidential information. Two years ago another Stanford researcher, Timnit Gebru, used Google Street View images to figure out how they voted for some cities in the United States, starting from photos of cars on the stand. Thanks to an algorithm, the Gebru team managed to grasp the correlation between vehicle types and US census data and the voting methods of the 35-city presidential elections in each district examined.
The question they wanted to answer was: given the scheme of vehicles in an area, could the algorithm accurately predict the demographic data recorded in the US census and presidential vote data? The answer was yes: • Looking at motor vehicles classified in each neighborhood, we can deduce a wide range of demographic statistics, socio-economic attributes and political preferences of its residents,” explained the researchers. In this case, unlike the study that links housing with the profiles of car insurance clients, some more details had been provided: for example, the sedans were more closely associated with the Democrats, while the pickups “matched” more with the “We found that by traveling a city for 15 minutes with sedan and trucks, it is possible to reliably determine whether the city voted democratic or republican,” says Gebru.
Results that raise questions in both cases about how from seemingly innocent data sets can filter out personal information and how much organizations should be able to use it. And, most importantly, what else can you discover about us using a simple web app or social network?