The Future of Artificial Intelligence and Machine Learning

By Arnaldo Pérez Castaño

Artificial Intelligence (AI) is a term widely used today to refer to an area of science that is getting not only a lot of attention from the common audience but also tons of research time and interest from Computer Scientists around the globe. For many people it may seem like a distant topic, something drawn out of science fiction stories or dreams of a more sophisticated form of living, but the reality is that impressive advancements have been made in AI during the last 20 years, to the point where many of our human dreams have become reality.

AI is an area of Computer Science that studies the multiple ways humans (programmers, scientists) can impregnate intelligence on a given software component by making it mimic the way humans react or act. AI includes the study of Agents, Multi-Agent Systems (MAS), Coordination and Communication between agents, Simulation and all topics related to Machine Learning.

Machine Learning (ML) is the study of how to make machines learn in various ways; it could be after some training, with no training at all but learning the structure of a data set, or it could be by learning over time by means of reward and punishment mechanisms. ML is probably the hottest topic nowadays on AI.

The first and very popular approach of ML is known as Supervised Learning (SL); in SL you start with a dataset that it is used in the training phase to train and adapt some structure to the dataset. Once this structure has been adapted to the dataset then new examples of the same type of the dataset are submitted to the structure for prediction, this second phase is known as prediction phase. Since the structure was train to correctly predict data in the dataset then it will now be able to predict with a certain degree of success all incoming examples. Typical examples of SL structures are Support Vector Machines, Neural Networks or Decision Trees.

A second approach of ML is that of Unsupervised Learning (UL); in UL, an algorithm is provided with a dataset and then using a certain UL technique it’s capable of learning the structure of that dataset. Learning the structure of a dataset simply means that patterns or features are learned from this dataset. A classic problem of Computer Science known as Clustering is a well-known example of an UL technique and has applications in many areas such as Psychology, Astronomy, Business, etc. It’s a very important technique in Data Mining and can discover frauds like those committed in the insurance field. In Clustering, the goal is to partition a set of n objects into k classes using some criteria, trying to guarantee that objects of the same class are as similar as possible and objects of dissimilar classes are as dissimilar as possible. Said in a practical example, if we have n pens and we wish to partition them using as criteria ‘color’ then we would have classes ‘red’, ‘blue’, … etc.  each with pens of the color described by their class.

The last approach, which is consider by many scientists as the future of AI and ML is that of Reinforcement Learning (RL). In RL, we want to mimic the way humans learn as closely as possible, we want machines to learn by ‘reward’ and ‘punishment’ lessons and over time. AIs implemented using the RL paradigm have already achieved great success, beating in many occasions the World Champion of different games for which they were created, something no other AI type had achieved prior to them. RL relates to topics such as Markov Decision Processes, Dynamic Programming, etc. Why is RL considered the Future of ML and AI?? Because being able to incorporate the ability of learning over time and learning by actions in terms of reward and punishment and on their own, represents an unbelievable learning model that we can now incorporate to machines. Imagine in a future a machine able to use RL to learn how to cook over time, or how to love, or drive (learning in a safe environment); the possibilities are in general outstanding. 

In the book ‘Practical Artificial Intelligence: Machine Learning, Bots and Agent Solutions using C#’ the reader will be able to find multiple algorithms and practical examples implemented in C# not only of the three approaches of ML previously mentioned but in general of many others related to AI (Agents, Multi-Agent Systems, Mars Rover, Logic & AI, Automated Theorem Prover, Simulation, etc.).

About the Author

Arnaldo Pérez Castaño is a computer scientist based in Cuba. He’s the author of PrestaShop Recipes (Apress, 2017) and a series of programming books—JavaScript Fácil, HTML y CSS Fácil, and Python Fácil (Marcombo S.A.)—and writes AI related articles for MSDN Magazine, VisualStudioMagazine.com, and Smashing Magazine. He is one of the co-founders of Havana Digital Enterprises and the creator of their first websites, http://www.havanaclassiccartour.com and www.havanadanceclass.com. His expertise includes Visual Basic, C#, .NET Framework and Artificial Intelligence.  Arnaldo offers his services through freelancer.com and served as reviewer for the Journal of Mathematical Modelling and Algorithms in Operations Research. Cinema and music are some of his passions.

This blog post was contributed by Arnaldo Pérez Castaño, the author of “Practical Artificial Intelligence: Machine Learning, Bots, and Agent Solutions Using C#”.