AI glossary: Key terms to help you understand artificial intelligence

Artificial Intelligence - what you need to know

AI is transforming how we live, work, and interact with technology. In its most basic sense, Artificial Intelligence refers to creating computer systems that can perform tasks that generally require human intelligence - such as learning, reasoning, problem-solving, and decision-making. AI systems today employ complex algorithms and vast amounts of data to interpret information, recognize patterns, and make predictions or decisions, often without being explicitly programmed to perform the specific task.

From computer vision, which enables machines to interpret visual information, to natural language processing, which allows computers to understand and generate human language, Artificial Intelligence (AI) is at the heart of some of the most sophisticated technologies nowadays.

AI Glossary: Essential Terms

To stay up to date in the Artificial Intelligence solutions era, it's critical to become acquainted with the fundamental terms, concepts, and jargon that characterize this fast-developing discipline. Whether you're a novice venturing into AI for the first time, a data science hobbyist, or an expert dealing with machine learning models, this glossary will guide you through the intricate world of AI technologies. Discover the foundational concepts of neural networks, deep learning, natural language processing, and beyond. It will help you comprehend how AI systems read and analyze data, perform tasks, and innovate industries better.

Basic AI Terms

1. Artificial Intelligence (AI) is the simulation of human Intelligence in machines that are programmed to think and mimic human actions, like learning and problem-solving.
2. Artificial General Intelligence (AGI) is a theoretical form of AI with capabilities comparable to human Intelligence across a wide range of tasks. Unlike narrow AI, AGI would be able to understand, learn, and apply knowledge in different domains, but it remains a research goal and has not yet been achieved.
3. Generative AI (GenAI) is a group of AI systems capable of creating new content, such as images, text, or music, based on patterns learned from existing data.
4. Machine Learning (ML) is a subset of AI where computers learn to recognize patterns and make decisions based on data, without being explicitly programmed.
5. Large Language Models (LLMs) are a group of models focused on understanding and generating human-like text, which are trained on vast amounts of text data (i.e., Wikipedia or books).
6. A Neural Network is a computational model inspired by the human brain that processes information through interconnected nodes (neurons), allowing machines to learn from data.
7. Deep Learning (DL) is a type of machine learning inspired by the structure and function of the human brain, using artificial neural networks with many layers to learn from large amounts of data.
8. Deep Neural Network is a neural network with multiple hidden layers, capable of modeling complex data relationships and improving pattern recognition in tasks like image processing and gameplay.
9. Deep Learning Model is a model based on deep learning architectures, often used in generative AI and trained on large datasets to create complex, original content. Deep Learning Models are responsible for advanced AI applications such as image recognition, language understanding, and generative content creation, leveraging deep neural networks for high performance.
10. Deep Learning Algorithms are advanced machine learning techniques inspired by brain-like structures, enabling the analysis of complex data and automation of tasks without human intervention.
11. An AI System is an integrated technology designed to perform specific tasks, such as generating content or making predictions. AI systems consist of components like models, algorithms, and data, and are defined by their capabilities and functional roles.
12. AI Agents are autonomous software programs capable of perceiving their environment, making decisions, and performing actions to achieve specific goals. They are used in applications like virtual assistants, chatbots, autonomous vehicles, and agentic AI systems.
13. Foundation Models are large, versatile machine learning models trained on vast datasets that can be adapted for various applications, including natural language processing, image, and audio generation. They are an intermediate step between task-specialised AI and AGI.  
14. Machine Learning Algorithms are specific algorithms that enable AI systems to learn from data, improve prediction accuracy, and reduce errors. They are fundamental to predictive analytics and automation.
15. Machine Learning Techniques encompass various algorithms and models, such as neural networks, decision trees, and support vector machines, that enable AI systems to learn from data and perform tasks like classification or pattern recognition.
16. A Machine Learning System is a complex AI architecture that learns from data to perform specific tasks. These systems often raise issues of transparency, explainability, and reliability in decision-making processes.
17. Training Models refers to the process of training AI and machine learning models using data, rewards, and feedback mechanisms to improve their performance over time.

Datasets

1. Train/validation/test set: Train Set is the subset of data used to train a machine learning model, teaching it to recognize patterns or perform tasks.
2. The validation set is the subset of data used to evaluate a machine learning model during the training process. This part of the data set is often used to choose the best-performing version of the model. Test Set is a separate subset of data used to evaluate the performance of a machine learning model after it has been fully trained, to ensure it can generalize to new, unseen data.
3. Training Data is the dataset used to train machine learning models, including deep learning and generative AI. The quality and representativeness of training data are crucial for AI performance and fairness, with issues like overfitting, bias, and ethics being important considerations.
4. Labeled Training Data is data that has been annotated with correct answers or labels, used in supervised learning to help algorithms recognize specific patterns.
5. Unlabeled Data is data without pre-existing labels, used in unsupervised learning to identify patterns, groupings, or clusters without guidance.
6. Structured Data refers to organized, searchable datasets, often formatted into rows and columns (like spreadsheets), which are easy to analyze and play a critical role in training machine learning models.
7. Unstructured Data is data that lacks apparent organization, such as text, images, or audio. Special techniques like deep learning are required to extract meaningful insights from unstructured data.
8. Big Data refers to large, complex datasets that are difficult to process using traditional methods. Big data is important for analyzing patterns, informing business decisions, and advancing AI development. Thanks to Big Data, we entered the successful era of LLMs. 
9. Historical Data is past data used in predictive analytics to forecast future trends, behaviors, or outcomes across various industries.
10. Data Processing is the automation and streamlining of data handling using AI, which eliminates manual errors and improves workflow efficiency.
11. AI Techniques are the range of methods, algorithms, and tools that enable machines to learn from data, perform tasks, and improve over time.
12. Generative AI Tools are AI systems that create original content such as text, images, and videos, powered by models like large language models and transformer-based neural networks. Examples include ChatGPT, MidJourney, and Cursor.
13. A Self-Driving Car is an example of AI technology capable of acting independently, replacing human intervention, and making autonomous decisions.
14. Human Learning relates to the process of learning in humans. This is often contrasted with machine learning, as AI algorithms mimic human learning processes.

Division of the ML world

1. Supervised Learning is a machine learning approach where the model is trained on labelled data. Typical examples are Classification or Regression.
2. Classification is a Supervised Learning task of predicting the category or class of an item based on its features.
3. Regression is the Supervised Learning task of predicting a continuous value based on input features.
4. Unsupervised Learning is a machine learning approach where the model is trained without labeled data, allowing the model to identify patterns and relationships in the data on its own. Typical examples are Clustering or Dimensionality Reduction.
5. Supervised and Unsupervised Learning are fundamental types of machine learning models. Supervised learning uses labeled data for specific input-output mappings, while unsupervised learning identifies patterns and groups within unlabeled data, both playing key roles in AI training and pattern recognition.
6. Clustering is the Unsupervised Learning task of grouping objects based on their observations.
7. Dimensionality Reduction is the process of reducing the number of random variables under consideration by obtaining a set of principal variables.
8. Reinforcement Learning is a type of machine learning where an agent learns to make decisions by performing actions and receiving feedback in the form of rewards or penalties. It was used, for example, in AlphaGo. It is also widely used in robotics.
9. Training Models refers to the process of training AI and machine learning models using data, rewards, and feedback mechanisms to improve their performance over time.
10. Natural Language Processing (NLP) is a branch of AI that aims to enable computers to understand, interpret, and generate human language.
11. Natural Language Processing Tasks are tasks performed by large language models and other AI systems, such as text generation, understanding, translation, and summarization, enabled by training on massive datasets.
12. Computer Vision (CV) is a branch of AI that enables computers to see and reason, just like humans.
13. Automatic Speech Recognition (ASR) is a branch of AI that focuses on enabling computers to listen to and understand human speech.
14. A Convolutional Neural Network (CNN) is a deep learning algorithm that can take in an input image, assign importance to various aspects/objects in the image, and differentiate one from the other.
15. Recurrent Neural Networks (RRN) are neural networks designed to process sequence data and retain information over time, making them suitable for applications like natural language processing and speech recognition.
16. A Generative Adversarial Network (GAN) is a class of machine learning frameworks in which two neural networks compete to generate new, synthetic instances of data that can pass for real data.
17. Transformer is a family of models (popular in NLP but also in CV) that uses self-attention as its main building block, which allows the model to weigh the importance of different parts of the input sequence dynamically.
18. Bias is an error caused by inaccurate assumptions in the learning algorithm. It can also refer to the unfair and discriminatory prioritization of certain outcomes over others.
19. Overfitting is a modelling error that occurs when a function is too closely aligned to a limited set of data points and may fail to generalize well.

(EXTRA) Underfitting occurs when a model is too simple to capture the underlying pattern in the data and cannot perform well on the training set or new data.

1. A feature is a characteristic of the observation that is used by the Machine Learning algorithm to provide an output. For human identification, it can be height, weight, hair colour, age, name, and surname.
2. Model fine-tuning is a technique where a pre-trained neural network is adapted to perform a new, specific task. This process leverages the knowledge gained from a large, general dataset (used for pre-training) and applies it to a more specialized dataset for the target task.
3. Data augmentation is a process of creating more data by performing some transformations on it so that it appears similar but is not exactly the same, aiming to increase the robustness of the models.

Building Blocks of Neural Networks and Machine Learning

1. Hyperparameters are configuration settings used to structure the machine learning process. They are set before the learning process starts and control the training algorithm's behavior.
2. The loss function is a function controlling the learning process, which is minimized or maximized during training to find the best-performing model.
3. Ensemble Learning is the method of combining multiple learning algorithms to obtain better performance than with a single model.
4. Backpropagation is the process by which neural networks update their weights based on the error rate obtained in the previous iteration.
5. The Activation Function is a function that introduces non-linearity into the model, helping the neural network learn complex patterns in the data.

Evaluation series

1. Accuracy is the ratio of correct predictions to the total number of predictions made.
2. Precision is a fraction of relevant instances among retrieved instances. Recall is a fraction of all relevant instances retrieved.
3. A False Positive is an error when the algorithm gives a positive output while the label is negative.  False positives are critical in contexts like medical testing, spam detection, and security, where they can lead to unnecessary actions or concerns.
4. A False Negative is an error when the model gives a negative output while the label is positive. False negatives can be particularly problematic in situations such as medical diagnoses, security screenings, and fraud detection, where missing a positive case can have serious consequences.

Chatbots and NLP series

1. Hallucination refers to a mistake or error made by a model that generates incorrect or nonsensical outputs that do not align with reality or the intended task.
2. A Token in NLP refers to a unit of text, typically a word or a subword, that is processed as part of the input data by an AI model.
3. The prompt is a specific instruction or input given to an AI model to generate a desired output, guiding its behaviour and responses.
4. Retrieval Augmented Generation (RAG) is the concept of finding relevant information in a knowledge database and then adding it as an input to the text generation model, consequently simplifying the task from “understanding everything” to “reasoning” over given text.
5. Chain of Thought (CoT) is a concept where a model to provide an answer is asked to explain how they arrived at such a conclusion, presenting intermediate steps in its reasoning
6. Prompt engineering is a concept of correctly instructing text generation models to create reasonable output according to our needs.
7. Few-shot learning is a type of machine learning where a model is trained to recognize and classify new categories or tasks using only a small number of examples.
8. Zero-shot learning is a type of machine learning in which a model can correctly predict new, unseen classes. Instead of learning from labelled data, it uses other information, such as semantic embeddings, descriptions, or relationships between classes.
9. The agentic approach is an idea where, instead of using one model for a complex task, you decompose it into smaller tasks, all of which are handled by different specialized instances of the models.
10. A Mixture of Experts (MoE) is a type of ensemble learning technique that divides the problem space into subspaces and assigns specialized models (experts) to each subspace, with a routing mechanism on which expert to consult.
11. GPT Models are large language models (LLMs) based on the transformer architecture, pretrained on vast datasets to generate human-like text. They are used in chatbots, AI services and can handle multiple data types in multimodal applications.

Others

1. Explainable AI (XAI) refers to AI methods and techniques that make AI behavior and decisions transparent to users, allowing them to comprehend, trust, and effectively manage these systems.
2. MLOps, or Machine Learning Operations, is a set of practices for reliably and efficiently deploying and maintaining machine learning models in production.
3. Multimodality is the idea of using more than one modality (i.e., tabular data, image, infrared image, text, audio, speech) for your model to widen the context for the model.
4. Self-supervised learning is a type of machine learning in which the system learns to predict part of its input from other parts of its input without needing labelled data. It leverages the inherent structure in the data to create pseudo-labels, which can then be used to train the model, e.g., having a similar output for transformed images or predicting the next frame/word based on previous ones.
5. Diffusion models are a class of generative models that, during training, learn to remove noise from progressively more noisy images, while during inference, they iteratively create images from random noise.
6. A bounding box (bbox) is a rectangular box that is used to define the location of an object within an image.

Conclusion: Navigating the Future of AI

With Artificial Intelligence developing at a breakneck pace, learning the primary terms and concepts behind AI systems is now more important than ever. From large language models and deep neural networks to the latest developments in natural language processing and computer vision, AI technologies are transforming industries, enhancing decision support systems, and enabling machines to perform tasks previously considered to be within the realm of man alone.

Monitoring the evolving jargon of Artificial Intelligence not only helps you stay updated on the newest developments but also empowers you to maximize the usage of AI tools, generative AI models, and data science techniques in your own job or routine tasks. As increasingly sophisticated AI models become integrated into everything from voice assistants to self-driving cars, a solid grasp of AI terminology will allow you to tackle the future with confidence and curiosity. Continue to learn, continue to explore, and stay at the forefront of the world of Artificial Intelligence.