The truth about the use of machine learning

BY: EMMANUEL URÍAS

Machine learning, Artificial intelligence, and Big Data are topics recently rising in interest among global companies. However, what do these names mean? Moreover, how do they form part of the daily life?

Many people relate them to science fiction movies like Spielberg’s already old “Artificial Intelligence” or the well-developed character of Jarvis in the Avengers. Maybe, even the conspiracies and fear raised by the stories of AI takeover come into the algorithm. However, it may be disappointing that we are currently very far to a point at which machines can actually reach the intelligence of a 4 years old child.

Even then, not everything is discouraging. The advances in Machine Learning, Artificial Intelligence, and Big Data are undeniable and are already part of everyone daily lives. The focus in this article will be Machine Learning although it terms as Big Data and Artificial Intelligence will appear constantly due to their closeness.

Defining Machine Learning and its’ applications

It is relevant to clarify that Machine Learning is a discipline in the context of Artificial Intelligence studies. In theory, it is a group of techniques and tools highly functional to enable machines “learning” automatically. That is to say, through a general data feedback and repetition, machines can obtain different results varying according to the interactions with the data itself.

In the actual scenario, Machine Learning is quickly developing and constantly taking more and more part of the daily life. Its applications are diverse, ranging from bank security, fraud detection and data protection tools to the most mundane things as the recommendations in the home pages of Youtube and Netflix customized to the user from the personal search history.

There lays the base for “Learning” in the Machine Learning name. Building from repetitive patterns and information segmentation, the machine learning algorithms set the system to answer in a distinct way for each user and case. For example, each time a user searches for a song in a platform that relies on Machine learning, the system search and sorts it by categories depending on genre, singer, views, date or even reviews made by other users.

Each characteristic of the song is stored in a Database, sometimes coinciding with multiple other songs. Then, after a considerable amount of sorting, the system craft a list of suggestions tailored for that particular user in order to make the best fit for the user’s search profile.

In the case of the user looking for something outside the common patterns, the machine should immediately generate a different kind of search. Associating a variety of data to obtain an answer fitting the user’s profile.

All of this is part of the Machine Learning utility. Analyzing sports, movies entertainment and marketing of diverse products, the machines or systems apply similar procedures and make databases of information to design repeatedly better predictions for the companies using Machine Learning.

In a more complex context, Machine Learning development focus point to security. Finance and information protection is of the highest importance and the best mathematicians, system, and informatics engineers work together to create algorithms that anticipate possible frauds and hacking attempts. News about the trial implicating Mark Zuckerberg with the inappropriate management of Facebook users’ information show dangers that are no longer fiction and have in many cases grave consequences.

For a common person, the dangers go from misuse of funds, electronic accounts stealing and private data theft. For companies, the losses can reach millions of dollars.

With all the risks and their possible consequences, surged enterprises dedicated specifically to develop software protection based on Machine Learning, Artificial Intelligence, and Big Data. With that, their systems are able to recognize and deal with varied threats classifying each according to its type and segmenting data to anticipate to any possible attack.

The most common method employed by Machine Learning Algorithms for protection is classification. Applying it to Intrusion Detection Systems (IDS), and varying on the coding level, it helps to solve problems like malware, spyware or spoofing.

An example is the prevention of phishing a user. This is to say determining if the identity of an email is of legitimate origin or if it is an impersonation.

The Machine Learning is of the kinds supervised or not. In the first case, the supervisor directs the searches made by the machines to look for something specific relying on the information to detect different kinds of data; the opposite is to avoid the risk of human bias on the equation and to save work for the personnel available.

This shows only a little of the Machine Learning possibilities. Sadly, not everything is good news. Just as Machine Learning improves security, the criminals can use it to recognize patterns of protection and to crash the safeguards. While Machine Learning detects 80% to 90% of the threats, the attacks are always evolving to adapt to the security systems and breach them. For this reason, the misused advance of Machine Learning, represent an important danger by itself.

Machine Learning challenges’ to overcome

Although Machine Learning is part of what many call the “fourth industrial revolution” together with Artificial Intelligence and Big Data, it certainly has limitations that delay the arrival of the greatly anticipated functional and automatic assistants.

The first problem is the huge amount of data generated by the systems. This data is stored on servers that are not only expensive but also require proper maintenance which results hard in many places so not every company can afford them.

Additionally, migrating the information to Big Data requires local processes and technology that many companies are yet to implement.

Representing with math the objectives and models desired of Machine Learning. The algorithms to predict specific answers and in turn generate adequate solutions are highly complex. In many occasions, the codes created by mathematicians and engineers are still not efficient enough and need improving to process the number of variables required to make answers adjusted to reality.

Another challenge is the results’ compression. On occasions, the algorithms and personnel in charge of decoding and interpreting the data, fail to do so properly; there are times when even after finding an answer, there is no way to determine why did that particular solution work or how was it achieved due to the growing intricacy and complexity of the nets and potential solutions. All of this increase proportionally with the capacity to multitask and processing of the machines.

In any case, even though the dreamt of Machine Learning is very far from its top potential, the companies are betting strongly on developing the necessary technology. That companies like Microsoft, Netflix, and Amazon invest huge amounts of money to develop Machine Learning is no coincidence at all. Nowadays, Machine Learning and Artificial Intelligence are slowly becoming an integral part in sectors as bank, security, medicine, pharmaceutics, and health to mention some.

Although many of its possibilities seem distant, at the end it is undeniable that the limits for the technology are also unknown and there is plenty of room for future development.