Dell Enables an Industrial Digital Twin Proof of Concept with Artificial Intelligence Technology

AI has been around for some time, but recent developments have made it one of the hottest topics in tech. Although AI is becoming mainstream, the technology is still new to many, and many of the related concepts and terminology remain unclear. This Futurum Group and Signal65, insight looks to demystify AI basics including machine learning and deep learning.

Analyst Take

Over the past year, AI has been everywhere. It’s become the most hyped topic in the tech world and beyond and seemingly every organization has an initiative to adopt AI in some form or another. And while the field of AI has been around for some time, the technology – and the concepts, vocabulary, and buzzwords that come along with it – are unfamiliar to many. So, what exactly is AI and is it different from machine learning? What’s the difference between supervised and unsupervised learning? And what exactly makes deep learning so “deep”?

Artificial Intelligence

AI (Artificial Intelligence) is a broad term that is used to describe a wide range of applications and capabilities – so many, in fact, that the exact meaning behind AI can become lost. The term AI has been associated with everything from virtual assistant devices, to self-driving cars, to sci-fi style robots set on world domination. In some cases, AI has been used merely as a marketing style buzzword and it may not really mean anything at all. Most recently, AI has become synonymous in the minds of many with generative AI chat applications such as ChatGPT.

The reality is that AI is a field of technology focused on machines that are capable of achieving intelligent, human-like actions and decisions. AI stems from the even broader field of computer science, and while recent innovations have brought AI into the spotlight over the last few years, it has been around for decades. Perhaps the first real conceptualization of AI (outside of pure science fiction) dates back to 1950 in Alan Turing’s paper titled Computing Machinery and Intelligence which opens by posing the question “Can machines think?”. The first use of the actual term “Artificial Intelligence” is credited to John McCarthy only a few years later during a conference at Dartmouth College in 1956. From there, the field of AI has been on a tumultuous journey with several breakthroughs and setbacks. But the roots of the technology are not new – in fact, neural networks (more on these later) stem from even earlier research that attempted to replicate human neurons in the 1940s

So, if AI technology is not new, why then is it hyped as the latest technological innovation? Although research and development of AI has been in progress for decades, earlier iterations of AI applications typically had a limited scope of abilities. AI applications that can beat a human in the game of chess, such as IBM’s Deep Blue developed in 1997, for example, are certainly impressive, but they only really apply to the game of chess. The excitement around newer AI applications, much of which started with the release of OpenAI’s ChatGPT, stems from their broad applicability and ability to generate new information – termed Generative AI or Gen AI.

These new Gen AI applications represent a significant breakthrough in AI – finally, AI applications appear to be truly “intelligent” and the possibilities for how they may be applied is seemingly endless. Today’s AI applications typically utilize machine learning and deep learning techniques – and like the general concept of AI – these techniques have actually been around for some time. The rapid advancement seen in the field of AI however is made possible with greater computing power, such as GPUs, greater access to large scale data sets, and innovative new architectures such as transformer neural network models.

Machine Learning

Machine Learning (ML) is the main technique currently used within the field of AI. ML, as the name suggests involves providing data to a machine that allows it to learn some behavior. ML models use a sample data set, called training data, to learn the patterns and characteristics relevant for completing a desired task. Once the model is trained, it can apply what was learned to complete the task using new data. Machine learning algorithms differ from traditional algorithms in that the outcomes are determined by the data it was trained on, rather than executing specific programming instructions.

ML algorithms can be used for a number of tasks including categorization, regression, pattern recognition, and prediction. Examples of popular machine learning algorithms include linear regression, decision trees, native bayes algorithm, K nearest neighbors, K-means, and random forests. ML algorithms are categorized as Supervised, Unsupervised, or Semi-Supervised.

  • Supervised Machine Learning refers to algorithms which are trained with labeled data. These algorithms learn by comparing the characteristics of the data with what its corresponding label output. The trained model can then be used to predict labels for new data, making it useful for classification and prediction applications.
  • Unsupervised Machine Learning refers to algorithms that are trained on unlabeled data. These algorithms recognize patterns and similarities between the data, without having a labeled output to rely on. Unsupervised learning can be used to recognize new insights and patterns that exist within datasets.
  • Semi-Supervised Machine Learning refers to an approach that combines labeled and unlabeled datasets. Semi-Supervised algorithms can create predictive models, similar to supervised machine learning, while only a small subset of the data contains labels. These models will use the labeled data to predict labels for the unlabeled data, and then incorporate the new data that was labeled accurately back into the model for further training.

Deep Learning

Another key term within the field of AI is deep learning, which is a subset of machine learning that uses artificial neural networks to solve problems. Artificial neural networks, often referred to as simply neural networks, mimic the structure of the human brain by connecting nodes in a similar pattern as neurons. The networks consist of multiple layers – an input layer, one or more hidden layers, and an output layer – and it is this “deep” layered architecture that gives deep learning its name.

Neural Network Architecture Diagram
Figure 1: Example Neural Network Architecture

The connection between the nodes in a neural network are assigned weight values that signify the strength of the connection between nodes. As the neural network is trained on a dataset, these weight values are adjusted to better fit the relationship between nodes. A full explanation of the underlying calculations that make neural networks work is beyond the scope of this insight, but the result is a computer model that can input an arbitrary number of features, learn the relevant characteristics, and provide an intelligent output.

Neural networks are incredibly flexible and powerful tools which can be trained to understand various forms of data including text, images, and audio, making deep learning an effective method for a wide range of challenges including natural language processing, image classification, recommendation systems, and generative content. An advantage to deep learning is in the ability of neural networks to automatically learn features within the data, requiring less human intervention and feature engineering compared to other machine learning methods. This makes deep learning specifically advantageous for recognizing patterns within large, complex datasets.

A key challenge with neural networks, however, is their computational complexity. Neural networks rely on computations such as matrix multiplication and gradient calculations which can be computationally expensive. The amount of these calculations required is driven by the number of parameters used in the model. Complex models with large numbers of parameters are continuously being developed to create more powerful AI applications, however their computational complexity is quite high, often requiring high end hardware such as GPUs. The computational complexity involved in such models additionally translates to energy consumption, leading to concern around the sustainability of deep learning applications.

Final Thoughts

While the field of AI has been around for some time, recent advances have brought a new wave of excitement to the technology. As is the case with most technologies, AI is filled with techniques, concepts, and buzzwords, that can be confusing to those who are unfamiliar with the topic. The field of AI is quite broad and it is currently moving very fast. While there are several more in-depth concepts and techniques relevant to the world of AI, it is important to have an understanding around the fundamental concepts of AI as a field, machine learning, and deep learning.