Artificial Intelligence – Software that understands no longer a man – Süddeutsche.de

Recommendations Amazon seem harmless. But what happens when algorithms over the lending decision? “Machine Learning” raises ethical questions.

There is a misconception that most people are subject to, when it comes to their computers and smartphones. You think to be at the controls of legendary knowledge tradition machines. In this case, the opposite is the case. . The devices are always mysterious operators who disguise as they actually come to what they serve there

That this is no longer just a theoretical problem, take two quotations: “We are fast becoming a point at which we no longer understand how software works, “said Peter Purgathofer, professor at the Vienna University of Technology recently at a conference that dealt with the algorithms and the control they exert over our lives. “People overestimate the extent to which IT companies understand how their own systems to work.” That says Andrew Moore. He should know. He is Dean of the Institute of Computer Science at Carnegie Mellon University and was until last year vice president at Google.

One of the most promising new fields in the IT industry is that which Informationsethikern prepares the most worried. Machine Learning states; expressed simplistic, these are a precursor to the artificial intelligence. This refers to the ability of software to analyze ludicrous amounts of data to make a decision. A researcher in Moore Institute has developed a system that can process billions of different data points. Since you can as a human ever lose track.

Moore illustrates his concerns the example of film recommendations, as we know them from Netflix or Amazon. “Machine-learning models used for recommending large amounts of information, the former users have left,” says Moore. “This can range from the color of the movie poster up to the proximity to other people who have this TV evaluated positively, include everything.”

This may seem harmless, as long as it’s all about movies. But if companies rely to make their business decisions on algorithms that ensures significant ethical problems. The best example is the US patent number 9100400, filed in August by Facebook. It bears the name “Authorization and Authentication based on on individual’s social network.” A sample application that provides Facebook in the patent relates to lending. “If a person wants to take out a loan”, it says, “examines the lender’s credit standing of members of the social network of the individual.” If the average credit rating of these members at a minimum, the application is rejected.

The systems that decide what happens to us on the web, so are not only more complex, but by also increasingly dangerous. Because when the system decides that the sample from the network sufficient? The computer draw the user for something to account, on which they can have no more influence.