In the previous post of our Understanding machine learning series, we presented how machines learn through multiple experiences. We also explained how, in some cases, human beings are much better at interpreting data than machines. In many tasks machines still can’t replace humans, who understand surrounding reality better and can make more accurate decisions.
Machines can be given a list of symptoms of a suffering patient, compare it with the database of diagnosed patients and give their own predictions. However, an experienced medical doctor, probably having in mind a smaller number of cases can still put more reliable diagnosis. It’s because he takes into account many factors, that can’t be easily presented by data. The question arises – do we need machines and machine learning at all? Yep, we do and here is why.
Scale of the task
Even a very skilled bank employee will be able to assess, per day, the creditworthiness of only a few dozen clients. Besides, let’s face it – we are only humans – we get tired and sometimes have bad days which can affect our efficiency at work. While machine can make a similar decisions for the tens of thousands of potential borrowers, per day, with constant efficiency.
Machines vs. humans, 1:0
Imagine that you are an insurance agent and your job is to assess the likelihood of a car accident for a certain client. You have a variety of data, we live in the era of Big Data, like age, gender, occupation , income, frequency of travel, marital status, place of residence … .Stop ! And now try to assess the impact of these factors on the possibility of causing an accident by that person . At some point , we’ll give up and choose 2-3 factors that in our opinion are the most important. And machine? Machine will assess the impact of each variable (remember, there can be hundreds or thousands of them) for all policyholders, at the same time. And arrange them from the most important to the least important. And sometimes it happens in a blink of an eye.
Machines vs. humans, 2:0
Experts are still not reliable in terms of gauging exact probability and modifying their predictions based on exact cost analysis. Experts will assess whether a person will repay the loan and probably will make few mistakes, but all of these mistakes will be treated equally (giving credit to bad client or declining credit for a good one). However, if we define the costs for both situations (perhaps in 3:1 proportion), an expert will have a problem with changing his grading system to reflect that.
Machine will perfectly set its probability assessment to minimize the costs.
Machines vs. humans, 3:0. Sorry humans.
Each time we make our decisions we must face the fact, that we will make the wrong one. In the next post we will talk about how machines deal with that and minimize the risk of making a wrong decision.