Myths about machine learning

Early on Machine learning used to be something that happened behind the scenes. Large companies invested massively in creating smart search engines, which displayed recommendations according to your previous clicks and they would place ads where it’s the most likely for the consumer to click. But now machine learning gets all the attention and debates take place on national television discussing this incredible tech.  Cars, amazingly smart applications, translations and everything machine learning can or can’t-do is being discussed. Be ready, machine learning is one of those technologies which will positively shape our future.

Unfortunately, because everyone is talking about it, some information has become distorted and now we have several misconceptions and myths around machine learning. Here is a list of the most common myths.

1. Machine learning algorithms just discover correlations between different events. Well, this is the most mentioned fact about machine learning in the media. It’s like a flu virus spreading throughout the news. It is not necessarily false, but most learning patterns can discover more productive forms of information and knowledge, like new algorithms that are based on deep learning methods which can extract information from unstructured data.
2. Machine learning can’t discover causal relationships between data. As a matter of facts, one of the most popular types of machine learning consists in testing different actions and observing their results and consequences, which is the essence of causal findings. Let’s take an example. An e-commerce site can build distinct ways of displaying the products on its pages, but then it will target the ones that increase purchases. Each of us most likely have participated in these experiments without even knowing it.
3. The more data you have, the more probable it is for you to imagine patterns. For example, the more phone calls the NSA listen in on, the more significant the probability of discovering innocent people as potential terrorists from using the terrorist detection rule. Mining more attributes of the same entity could indeed increase the risk of “hallucinating”, but machine learning specialists are doing their best to keep the danger at the lowest level. If we interpret the same mining activity differently and search for more entities with the same set of attributes, we can decrease the risk, because the rules will have stronger support. Machine learning algorithms can find these kinds of patterns involved in multiple entities, which makes them even more trustworthy. If a man is creating a video in a large train station, he may not be suspicious, neither a person who is buying a large package of ammonium nitrate, but if those two are close by and are having phone conversations, then police have a strong motive to investigate the situation.

Most of the myths above are pessimistic, they assume machine learning to be more limited that it is in reality. But there are also some optimistic misconceptions around the technology:

1. Machine learning will soon create superhuman intelligence. The movie industry has exploited the idea of machine learning taking over the human race. This technology is growing at an exponential and while the thought of machines taking over the race in the next several years is highly unlikely it is only a matter of time before the technology evolves and the machines are thinking and growing on their own.
2. Simpler models could be more accurate. Simple is preferred because it is easy to understand and remember. But sometimes, a simple hypothesis consistent with the data could be less reliable in making predictions than a more complicated one.

These are the 5 myths surrounding machine learning that we have been seeing lately. ML is now more powerful than it used to be and we need to accept that it will be helping us make more intelligent decisions.

If you have questions about the topic, feel free to comment in the section below.

Photo source: www.pexels.com