Artificial Intelligence vs. Machine Learning vs. Deep Learning - What’s the difference?
What's the difference between Artificial Intelligence, Machine Learning, and Deep Learning?
The terms Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) are often used synonymously. Strictly speaking, however, this is wrong. DL is a subfield of ML, while ML is a subfield of AI. This article explains the terms and their differences in more detail.
Artificial intelligence is an umbrella term and describes the rough approach of using machines to imitate intelligent human behavior to solve problems. The term “artificial intelligence” was first introduced in 1956 at a workshop at the Dartmouth Conference. However, AI has only become relevant in recent years, especially from 2015, mainly due to the significantly improved computing power in recent years, and the significantly increased availability of data (e.g. images and videos on the Internet ). Typical examples of AI are spam filters, smart assistants, Google maps applications, face detection and recognition, text editors and autocorrect, chatbots, search and recommendation algorithms, e-payments and banking, self-driving cars, and so on. But what exactly happens in AI? In AI the developer writes a code/program to execute a certain task, e.g. the spam filter. The spam filter sorts out e-mails based on specific characteristics such as sender address, particular words in the subject line, external links, etc. However, this AI would be very limited in its capabilities, because what would happen if the senders of spam emails change one of the aforementioned characteristics? The spam filter would no longer work. So the technology needs to learn by itself and adapt accordingly. This is called machine learning.
Machine Learning is a technology used to achieve Artificial Intelligence. First, data (e.g. images, videos, audio files, statistics, etc.) are collected and fed into the program - this is called input. This data is then passed to the program, which uses a more complex algorithm to analyze the data and make predictions or decisions. The special feature is that machine learning programs learn without human intervention. Netflix or Youtube recommendations are an example of ML. Input for the recommendations on these platforms is the media that you have already seen and liked. The recommendation algorithm then analyzes and compares tens of thousands of media pieces and finally recommends the best fit for you or what you supposedly would most like to consume. Other examples of ML are image recognition, speech recognition, medical diagnosis, predictive analytics, and so on. A machine learning system learns and becomes more intelligent with more data, ergo the more data the algorithm receives, the more precise the result. With increased data availability, optimized computing power through the further development of so-called GPUs (Graphics Processing Unit), and more refined algorithms, a new subfield has emerged - Deep Learning.
Deep learning is the evolution of machine learning. The technology makes use of so-called neural networks (or artificial neural networks). These Artificial Neural Networks (ANN) are originally inspired by the human brain and consist of artificial neurons. It has so-called input and output neurons. In between are several layers of intermediate neurons. The input neurons can be linked to the output neurons by learning in various ways via the intermediate neurons. The more neurons and layers there are, the more complex issues can be mapped. So the concept behind DL is that DL teaches machines to learn. The machine can improve its capabilities independently and without human intervention. This is achieved by extracting and classifying patterns from existing data and information. The knowledge gained can in turn be correlated with data and linked to a broader context. And finally, the machine is able to make decisions based on these links. These information links are given certain weightings by continuously questioning the decisions: If decisions are confirmed, their weight increases; if they are revised, the weight decreases. Between the input layer and the output layer, more and more levels of intermediate layers and links are created. So that the actual output is determined by the number of intermediate layers and their linking. To achieve accurate results, deep learning requires enormous amounts of data and therefore extreme computing power, which we are only just achieving with the current state of technology. Examples of DL are virtual assistants, vision for driverless, autonomous cars, speech recognition, language translation services, etc.
AI is constantly evolving, as can be seen in ML and DL. With DL, many practical applications of machine learning and extensions of the field of AI become real, because DL handles almost all tasks making any kind of machine assistance seem possible. AI, ML and DL are an integral part of the present and future. They pave the way for new advanced technologies. So look out for the future as it will be exciting.