With the upheavals caused by ChatGPT and its ilk, AI is throwing the company into disarray. The opportunity to go back to the fundamentals: what are the differences between AI, machine learning and deep learning? Do you think you know everything? It's time to confront your knowledge with reality!
Just read an article about the prowess of ChatGPT or MidJourney to encounter the terms Artificial intelligence, deep learning, machine learning. Their definition and meaning are sometimes mixed and often misunderstood. However, they designate very distinct concepts.
AI, machine learning, deep learning, what are we talking about?
Artificial intelligence
Artificial intelligence (AI) is the ability of a system to simulate human intelligence. The American pioneer of AI, Marvin Minsky, had a more precise definition: AI is “a science whose goal is to have a machine carry out tasks that humans accomplish using their intelligence. »
This therefore covers very diverse systems: expert systems, chatbots, machine learning, natural language processing (NLP), computer vision, autonomous agents, neural networks, genetic algorithms (optimization method inspired by the process of natural selection), etc.
Machine learning
Le machine learning (ML), or automatic learning, is a branch of artificial intelligence that allows computers to learn without having to be explicitly programmed. This definition is that given in 1959 by one of its pioneers, Arthur Samuel.
Machine learning works from examples. It uses algorithms to statistically analyze data and identify patterns. These models are then used to predict outcomes, gain a better understanding of the processes that generate this data or to make decisions.
Its uses :
- Classification: classify files according to their content, detect anomalies on a production line, detect spam, etc.
- Regression: predicting a numerical value, useful for knowing the evolution of the weather or the price of a stock
- Grouping: grouping customers by persona and purchasing habits
- Search engines, recommendation engines…
- Conversational agents (chatbots)…
Deep learning
Le deep learning (DL), or deep learning, is a subset of machine learning. It uses artificial neural networks (AAN), algorithms that are inspired by the functioning of the human brain, mimicking the way neurons send signals to each other. These neurons are organized in interconnected layers with a certain level of depth (deep learning). Each depth level helps optimize and refine the accuracy of the results.
Deep learning algorithms are very suitable for solving complex problems. They require a large volume of data and therefore very high computing power to process them.
Its uses : image recognition (health, industry, etc.), automatic translation (Google Translate, DeepL, etc.), voice recognition (Siri, Alexa, etc.), financial services (fraud detection, predictive analyses, risk assessment), autonomous driving cars , robotics (teaching robots to perform complex tasks), etc.
Let's sum it all up with a sketch.
The differences between machine learning and deep learning
Types of learning: supervised, unsupervised, reinforced
Machine learning and deep learning algorithms need to learn from example data to adjust their parameters, called training datasets (train datasets). Without training, artificial intelligence is nothing. The quality of learning conditions the quality of results. There are several types of learning: supervised, semi-supervised, unsupervised or reinforcement.
Supervised learning
Supervised learning involves one or more humans helping the computer by providing training data labeled with the correct answer to a question.
For example, is this email spam or not? Thanks to statistical analysis, the algorithm then understands what characteristics allow it to classify these emails. So, as it is presented with new emails, it will be able to identify them and assign them a probability score whether or not they are spam. The human will be used to correct its errors during the learning process so that it improves over time.
Unsupervised learning
Unsupervised learning applies when the answers we seek to obtain are not available in the dataset: the data is not labeled. The algorithm works without human intervention. It teaches itself to discover information from a set of data. Its results may be less precise than supervised learning.
Unsupervised learning is used to:
- data clustering operations based on their similarities or differences. For example, grouping bank customers according to their profile.
- association operations to identify relationships between variables in a data set.
Other types of learning
Semi-supervised learning involves learning labels from a partially labeled dataset. The advantage is that this avoids having to label the entire training dataset. This is often the case for processing an image bank.
Reinforcement learning consists of letting the computer learn from its experiences thanks to a reward and penalty system if the action taken was a good or bad choice. The goal of the algorithm will then be to define a strategy to maximize its reward. The main applications of this type of learning are games (chess, go, etc.) and robotics.
Types of data: structured and unstructured
Another big difference between machine learning and deep learning: the type of data input to the dataset.
Le machine learning deals with structured data, data organized according to a predefined model, easily indexable like a table or a database, as well as unstructured data, data that does not follow a particular pattern. Unstructured data can be text, images, videos, audio, etc.
Le deep learning used to process and analyze unstructured data.
Let's summarize with a table
Finally
AI, machine learning and deep learning are three distinct elements. AI is the discipline and by extension and metonymy, the products or services based on AI: Siri, ChatGPT, the AI of self-driving cars,Radiology AI that detects more and more cancers...
Machine learning and deep learning are two machine learning techniques used by AI. However, their uses are different. Machine learning algorithms will process quantitative and structured data while deep learning algorithms will focus on processing unstructured data, such as sound, text or images.
And what are generative AIs like ChatGPT, Bard, DALL-E or MidJourney? They produce text, images or code, or are even multimodal, and are based on LLMs (large language models). They therefore use deep learning and neural networks to process billions of unlabeled texts.
If you too want to create your own AI models, learn to work with AI or simply find out more about this revolution, we offer you around sixty seminars and training courses to take you into this exciting world!
