A Matlab toolbox for pattern recognition Imported pages from 37Steps

Home   Guide   Software   Glossary   FAQ   Blog   About

<< Pattern recognition: for better or worse?

Post - Monday, June 17th, 2013 a


It starts as an innocent curiosity: humans recognize patterns in the surrounding world. It helps them to communicate and survive. How do they do this? How do they detect a pattern? How is it defined and recognized?

However, once it is understood somewhat, it is applied. Pattern recognition is mainly an engineering field after all. And any application has its proponents and its opponents. They start to dispute, to quarrel, to fight about the question whether the application is a blessing or a threat, whether it will make the world better or worse.

In the beginning the discussion seems rather harmless. Is it really needed that we are able to talk to a machine? And why should it be able to read my handwritten notes? It will generate errors as a price of our laziness.

Applications like automatic sorting and industrial inspection have dramatically increased the speed and reliability of several processes in our society, at the cost of, sometimes boring, jobs.  But is it also needed that more intelligent tasks like the microscopic analysis of human tissues for cancer detection are taken over by machines? The gain of speed will allow for mass screening on a much larger scale. Not everybody is convinced that this is desirable as it will also introduce much larger numbers of false positives.

To my family and friends I explained my work by examples like the prediction of volcano eruptions from the analysis of seismic signals and the monitoring of bio-diversity by automatic counting of animals. Such positive examples are needed as they also read about cruise- missiles and killer robots that find an eliminate the enemy by pattern recognition.

The recent news about the NSA storing all internet traffic, email connections, bank transfers and travel data, made my friends even more skeptic about my activities. The news papers explained that once such data has been stored it is a matter of pattern recognition to trace people with an uncommon behavioral pattern. My earlier explanation that pattern recognition is far from trivial and will always make errors was not of much help. If it would have been perfect it was bad as everybody wants to be unique which makes them suspicious to such a system. If the system on the other hand is unreliable, it is even worse. Now also not-unique, middle-of-the-road people may by accident become suspected. Everybody has the risk of becoming the victim of such a system.

Does pattern recognition make the world better, or worse? I still believe the first, but this will not happen automatically. Human action remains needed. We need to make the dangers clear. Every politician deciding for such a system or relying on it should understand the high probability and risk of errors.