Machine learning and artificial intelligence, these buzzwords have seemingly been on everyone’s lips over the past few years and keep popping up on job descriptions everywhere. I have recently been scouting job openings in various sectors, one of which is the gaming sector, and it seems as though these terms have been used interchangeably quite a bit…however just because the personnel who publish these job posts use these terms interchangeably it does not necessarily mean that they mean the same thing. In fact, more often than not, these job posts are written by HR personnel or recruitment agencies who are not well-versed enough in this sphere to fully distinguish between these terms.
This topic has been discussed frequently by bloggers and scientists in this field, one of which is Chris Nicholson, CEO of Skymind and co-creator of Deeplearning4j, who sums up the relationship between these terms as follows: ” Machine learning is a subset of AI, which is an umbrella term for any computer program that does something smart. In other words, all machine learning is AI, but not all AI is machine learning, and so forth. ”
Let us now delve into Artificial Intelligence, which as the term suggests, is an area of computer science that seeks to mimic or achieve similar results to human intelligence. There are many AI methodologies (many of which are defined in this glossary by G2 Learning Hub) which can be used to simulate human intelligence, with some methods proving more effective than others. There exist three main approaches to AI: namely statistical methods, computational intelligence, and traditional symbolic AI. Statistical methods and traditional symbolic AI are rather straight-forward concepts: Statistical methods are mathematical formulas and/or models which are formulated by analysing a sample of data and which are used to make generalisations about other subsets of data in a similar format. A statistical method approach is a common approach to AI used to predict stock market prices; Traditional symbolic AI, or good old-fashioned AI (GOFAI), can be a group of if-then statements, or for the laymen reading this article: a set of logical steps or rules, which mimic the way a human would carry out a manual process. An example of such a process could be the calculation of an individual’s income tax based on local tax laws and their family structure. The definition of computational intelligence is somewhat more complex, and it is often argued that computational intelligence is a field within itself. For all intents and purposes of this article, computational intelligence refers to the application of soft computing to AI, which refers to a discipline of computing that deals with mathematical problems that cannot be clearly defined by hard rules (such as in the example of the income tax computation). For this reason, soft computing seeks to make assumptions about the data it is analysing. Computational intelligence is frequently used to calculate risk scores for diseases.
Now for machine learning. Daniel Faggella, founder and CEO of artificial intelligence market research firm Emerj put together a very pragmatic review of the different meanings experts have given to machine learning in his article “What is machine learning?” which is truly worth a read as it delves into this topic in great detail. In short we can say that machine learning focuses on the development of computer programs which access and analyse data automatically, without the explicit coding of rules. Such machine learning programs are self-learning and improve their ability to identify trends within the data they are analysing in an iterative manner. Initially the computer program defines a set of generic rules that define the distribution of the dataset. These rules are refined further at each iteration. Machine learning techniques can be a combination of statistical methods and computational intelligence depending on the machine learning algorithm implemented.
So there you have it, machine learning and AI are not interchangeable terms. Machine learning is a set of algorithms which only form a very small part of AI. There exist a plethora of machine learning techniques, which I will explore in another article, with different techniques being better suited for different problems. In the meantime I have left the comment section open as I would really appreciate any feedback on this article and whether there are specific topics that you would like me to discuss.