Imagine a system designed to provide information that can be useful for you. This kind of system would be analyzing your behavior and gathering data in order to provide selected advertisements from the data. Of course this is not novel. You need imagine nothing. You only need to open a social network to watch a real one. This kind of systems is around us in everything we do. For instance, cable TV companies analyze what programs you are seeing in order to improve their TV offer.
In previous times, the information we received was not strongly driven towards the individual, however, day by day, more received information is driven by automatic systems deciding what is interesting for us and what is not.
Social networks have increased hugely the complexity of our societies, but I was pointing out that there is an additional and not analyzed complexity that proceeds from the use of data mining systems.
Our brain feeds from the perceived information and works in a reinforcement scheme. This concept is very easy to understand. All of us know that we usually remember better the things that are repeated many times.
From a systemic viewpoint, those systems seem to be increasing the efficiency in the process of information distribution for the society; however, they have an additional effect for the individuals. They are reinforcing our preferences in a natural way.
Think about this example. Imagine a family where the parents are educated people and they usually spend their time watching scientific programs at the TV. A system with driven advertisements would usually show ads about scientific documentaries. An interest about science would be reinforced in the children of the family although they only see cartoons yet. This would not happen in a family where the adults are not interested in that kind of programs. This system, due to the reinforcement mechanism of the brain, would be contributing to increase the future differences among people and to create different social classes.
An effect of mass media is to produce uniformity at the society because the information received by all the people in that society is the same. However, this can be over with the new sources of information.
The structured information produces the specialization of the nodes processing information, and the brains of people are acting as processing information nodes. In our modern societies, general knowledge and information are being diminished while specific information is increased. This fact can increase the differences among people in the future, and this can produce less manageable societies because a government is more effective when all people receive a similar concept about the challenges, objectives and methods of the society.
Human brain filters well a random noise however a structured noise with the same energy feels very unpleasant. If the system is not effective in gathering our preferences, it will feel very unpleasant. The effect on benefit of the advertisement company could not be affected due to a little group of individuals wrongly classified; however, the effect on those individuals can be very negative because they only receive wrong information.
Machine learning systems are designed to cope with lack of information in unexpected situations. When there is not information enough the system will complete it through interpolation or extrapolation of previous data. In normal conditions machine learning algorithms works well enough but they can fail under unexpected situations too, depending on the used algorithm. Back to the cable TV example, imagine a certain channel is never seen in a certain home but is very accepted by most clients, and one day is connected for two hours. This can be due to an odd visit or due to a change of preferences. If the system takes the average of this event with home data, this would not be noticeable and it would be dischargeable, however, if the system links this event to the global data, it would understand a change in the preferences. As both hypotheses are possible, both schemes can be either right or wrong. This is unavoidable because our natural brain would act in the same way. The system is acting like a complement or even a substitute of our brain, making decision about our preferences instead of us. As we can see, structured information distribution can be positive working without uncertainty but it can have negative effects working under uncertainty that is the common situation and the reason why machine learning systems are being developed and improved.