The modern business world is on hype about data analytics and there is a reason. Digging deep into historical data strings is a way to valuable insights that might generate new revenue streams to boost efficiency and to discover improved ways of doing things. Does this mean data analytics cannot fail? Of course not.
In this blog post, we are highlighting 5 different cases where data analytics failed one or another way. Not matter it was because of wrong presumption, bad execution, missing variables or wrong numbers – the results were at least surprising and definitely not nice.
Donald Trump presidential run victory
Just as the Brexit, Donald Trump presidential victory was nothing less than surprise. Polls were widely generous on potential Hillary Clinton success as well as a Remain for Brexit. But the morning after things was way different. The truth is that polls are giving probabilities. FiveThrityEight` poll gave 85% chances for Clinton which mean that there was 15% probability for her to lose.
Another one is widely spreading the polls results. This might turn into bad favor for the poll winner because such information might discourage supporters to vote because they feel the victory is already secured. Such bias can’t be modeled or predicted.
According to Wall Street Journal, Trump`s campaign was alerted by its data analytics team within the morning of the election day that there is more than 50% chance for Trump to win the presidency. A day earlier the chance of winning was just 30%.
Google Flu Trends 2012/2013
Data without context might be less or more misleading. Just like Google Flu Trends in the winter season of 2012/2013. The tool relies on people`s location and flu-related search on Google search engine. The smart algorithms created by the data scientists are able to point the number of people with flu in the USA. In season 2012/2013, Google reported 11% of the USA population got the flu. The national Center for Disease Control and Prevention told the audience that the peak got 6% of the Americans infected with influenza. According to publication at New York Times, it seems that Google Flu Trends was led on by a large amount of flu-related media coverage.
Microsoft Chat Bot
This one is really funny! Microsoft worked hard to launch their AI Twitter bot which had to learn from Twitter users. The result was AI racist jerk arguing the existence of Holocaust, and also referred to women and minorities in an unacceptable manner. It took just 24 hours before the bot was disconnected for further adjustments. So this was a spectacular public fail but we are sure that the Microsoft guys are still trying to make the self-learning AI to behave in a mannered way.
The Samaritans suicide preventing app
The Samaritans, British NGO aiming to prevent suicides as much as possible created a free app to notify people whenever someone they followed on Twitter is posting potentially suicidal phrases like “hate myself”, “tired of being alone”, etc. Unfortunately, the tool aiming to let you know when a friend of yours need help is also empowering victim’s stalkers at crucial and vulnerable moments. The application lived just a week and a half and was shut down by the Samaritans.
Facebook napalm bomb picture
Facebook marketers are pretty familiar with the nudity rule which prevent uploading ads with too much bare skin. The Facebook community Users can not post pictures of nipples, bare buttocks, and genitalia. Sure, Facebook engineers created a data mining system identifying nudity on pictures and it blasted something which shouldn’t. The photography of nude and crying Vietnamese kid fleeing napalm bombs were anything but suggesting, hence child photography. Automated withdrawal of the pic caused public brawl which wasn’t really good for Facebook` AI reputation.
“The key thing to understand is that data science is a tool that is not necessarily going to give you answers, but probabilities,” said Erik Brynjolfsson, a professor at Massachusetts Institute of Technology recently quoted by NYTimes.
And that’s right.
