Before embarking on an analysis of potential applications of AI, it is important to understand what Artificial Intelligence encompasses, how it operates, and therefore its capabilities. Artificial Intelligence, described by one of the pioneers of the field is,
“the science and engineering of making intelligent machines, especially intelligent computer programs. It is related to the similar task of using computers to understand human intelligence, but AI does not have to confine itself to methods that are biologically observable.”
Artificial Intelligence can thus be considered as a field, occupying a very generic space of computer algorithms that are essentially able to draw inferences from observations. In fact, much of the advancements in Artificial Intelligence that we see today can be attributed to Deep Learning (DL), a subset of the field of AI (see figure below).
DL makes use of general but complex computational structures inspired by the physiological model of neurons to produce an output based on a particular input. These structures are composed of learnable parameters that are iteratively updated such that they better match observations that reflect the intended output of the model (backpropagation). In this way, deep learning models are ‘trained’ on data - the more complex the problem, the more data-hungry the model becomes. Once the model is sufficiently trained, it can begin automatically making inferences based on unseen input data, achieving state-of-the-art performance in a multitude of tasks including computer vision, natural language processing, speech recognition, and gaming.
Where previously traditional AI algorithms were covering every eventuality in a rule-based system when tackling problems, Machine Learning (ML) approaches introduced a new system of intelligence, where algorithms were designed to learn and make decisions from data without being explicitly programmed.
As seen in the figure above, and as described, DL is a subset of both ML and AI, it is concerned with designing general computational structures that can automatically capture and learn from complexities within data, a system of intelligence that has revolutionised the field, achieving state-of-the-art performance in a multitude of tasks. Although grounded in similar principles of learning, Deep Learning models can significantly vary in structure, and this of course is dependent on the task they are designed for.
In this Index we will focus on techniques/approaches sitting below the AI models shown above. The techniques are what directly enable use cases to be delivered and therefore we consider them as the right proxy to evaluate.
.png)
.png)
