Computers solve problems using algorithms. These algorithms are step-by-step instructions for the computer to sequentially follow. Algorithms are used to process a set of inputs into a set of outputs. These algorithms are typically written line-by-line by computer programmers. But what if we don’t have the fundamental understanding of a problem to be able to write the algorithm to automate it?
For example, consider filtering spam emails from genuine emails. For this problem, we know the input (an email) and the output (identifying it as spam or genuine) but we don’t know how to define what actually classifies it as a spam email – we just use our intuition. This lack of logical understanding often arises when there is some intellectual human involvement in the problem we are trying to solve. In this example, the human involvement is that a human-being wrote the original spam email.
Similarly, humans are involved in handwriting recognition, interpreting words (language) and facial recognition. It is clear that these problems are something that our subconscious is able to handle effortlessly yet we don’t consciously understand the fundamentals of the process. On the other hand, for sequential logical tasks, like sorting a list alphabetically, we consciously understand the fundamental process and therefore can program a solution (algorithm). This isn’t possible for more complex tasks like spam, handwriting recognition and interpreting language.
Machine learning is what gives us the tools to solve these ‘black box’ problems. We know what goes in and what comes out, so we can reverse engineer a ‘black box’ model of how to get there.
“What we lack in knowledge, we make up for in data”.
Let’s go back to the spam example. Our goal is to be able to identify whether a new incoming email is genuine or spam. We can use a data set of millions of emails, some of which are spam, in order to ‘learn’ what defines a spam email. The learning principles are derived from statistical approaches to data analysis. In this way, we do not need to understand the process/understand why it’s spam, but we can construct an accurate and functional model (a ‘black box’) to approximate the process and identify what is spam. Whilst this doesn’t explain the why’s, it can identify some patterns and regularities that allow us to determine whether the email is genuine or spam. Problem solved.
Artificial intelligence was conceived in the mid-20th century but it was not until the 1980s that the more statistical branch, machine learning, began to separate off and become a field in its own right. Machine learning developed a scientific approach to solving problems of prediction and finding patterns in data. This quickly had value in industry which fuelled the academic exploration further. But entering the 21st century we have seen rapid rise in machine learning popularity. This is largely due to the emergence of large data sets and the demand for data mining processes to extract knowledge from them. Machine learning has since established itself as a leading field of computer science with applications ranging from detecting credit card fraud to medical diagnosis.
