AI Glossary 111 Terms Explained

An activation function is a mathematical function applied to the output of each neuron in a neural network that introduces non-linearity, enabling the network to learn complex, non-linear relationships in data. Without activation functions, a neural network, no matter how deep, would behave like a simple linear model.

Agentic AI refers to artificial intelligence systems capable of autonomously planning and executing multi-step tasks to achieve a goal, without requiring human input at every step. These systems can use tools, browse the web, write and run code, and loop through actions until a task is complete.

An AI agent is an autonomous software system that perceives its environment through inputs, makes decisions based on that information, and takes actions to achieve a specified goal. Agents can operate independently, use tools, and adapt their behavior based on feedback from the environment.

AI alignment is the research field and engineering challenge of ensuring that AI systems pursue goals and exhibit behaviors that are beneficial and consistent with human intentions and values, especially as AI systems become more capable.

AI analytics is the application of artificial intelligence and machine learning techniques to data analysis, enabling automated pattern discovery, predictive forecasting, anomaly detection, and natural language querying of data at a scale and speed far beyond manual or traditional BI approaches.

AI as a Service (AIaaS) is the delivery of artificial intelligence capabilities, including language models, computer vision, speech recognition, and machine learning tools, through cloud-based APIs and platforms, allowing businesses to access powerful AI without building or maintaining their own models.

AI automation is the use of artificial intelligence to perform tasks, make decisions, and execute workflows that previously required human effort, with minimal or no human intervention. Unlike traditional rule-based automation, AI automation can handle unstructured data, adapt to new situations, and improve over time.

An AI benchmark is a standardized evaluation dataset or test suite used to measure and compare the capabilities of AI models on specific tasks. Benchmarks provide a common reference point for tracking progress, identifying weaknesses, and making informed choices between competing models.

AI content creation is the use of generative AI models to produce written, visual, audio, or video content, either fully autonomously or in collaboration with human creators. It encompasses AI-generated articles, marketing copy, images, social posts, scripts, product descriptions, and more.

An AI copilot is an AI-powered assistant designed to work alongside humans in professional contexts, augmenting their capabilities by automating routine tasks, providing intelligent suggestions, and enabling people to focus on higher-value work.

AI customer service is the application of artificial intelligence, primarily conversational AI and large language models, to handle, assist with, or augment customer support interactions. It encompasses AI chatbots that resolve issues autonomously, AI copilots that assist human agents in real time, and intelligent routing systems that ensure queries reach the right resource.

AI decision-making refers to the use of artificial intelligence systems to automate or augment complex choices by analyzing large volumes of data, identifying patterns, and applying learned models to recommend or execute decisions faster and more consistently than humans can manage manually.

AI ethics is the branch of ethics that examines the moral questions raised by artificial intelligence, including issues of fairness, privacy, accountability, autonomy, and the broader societal impact of AI systems and their deployment.

AI governance is the set of policies, processes, standards, and oversight structures that organizations and governments establish to ensure AI systems are developed, deployed, and used responsibly, safely, and in alignment with stated values and legal requirements.

AI guardrails are a set of technical and policy controls designed to constrain AI system behavior, ensuring outputs remain safe, accurate, and aligned with intended use. They include input filters, output classifiers, system prompts, reinforcement from human feedback, and monitoring systems.

AI personalization is the use of machine learning algorithms to dynamically tailor content, product recommendations, communications, and experiences to individual users based on their behavior, preferences, and context. It enables one-to-one relevance at massive scale that would be impossible through manual segmentation.

AI regulation refers to the body of laws, policies, and standards that governments and regulatory bodies create to govern the development, deployment, and use of artificial intelligence systems to protect individuals and society.

AI robotics is the field that combines artificial intelligence, including computer vision, machine learning, and language models, with robotic hardware to create systems that can perceive their environment, make intelligent decisions, and perform physical tasks with greater flexibility and adaptability than traditional programmed robots.

AI safety is an interdisciplinary research field focused on identifying and mitigating risks from AI systems, encompassing both near-term harms from current AI tools and longer-term risks from increasingly capable and autonomous AI systems.

AI search refers to search systems that use large language models and AI to understand queries in natural language, synthesize information from multiple sources, and provide direct, conversational answers rather than a ranked list of links.

An AI strategy is an organization's deliberate plan for adopting, developing, and leveraging artificial intelligence technologies to achieve competitive advantage, improve operations, and create value. It encompasses decisions about which AI use cases to prioritize, what capabilities to build versus buy, how to manage data and risks, and how to develop internal AI literacy.

An AI watermark is a hidden or visible signal embedded in AI-generated content, such as text, images, audio, or video, that identifies the content as machine-generated and can be used to trace it back to a specific model or provider. Watermarking is a key tool for AI content provenance and combating disinformation.

An algorithm is a finite set of well-defined instructions or rules that a computer follows to solve a problem, complete a task, or make a decision, forming the logical foundation of all software and AI systems.

Algorithmic fairness is the field concerned with identifying and mitigating discriminatory outcomes in AI and machine learning systems, ensuring that automated decisions are equitable across different demographic groups.

Anomaly detection is the AI and machine learning task of identifying data points, events, or observations that deviate significantly from expected patterns or the norm, signaling potentially significant, rare, or suspicious activity.

An API (Application Programming Interface) is a set of rules and protocols that allows different software systems to communicate and share functionality. In AI, APIs enable applications to access AI model capabilities, such as language generation, image analysis, or speech recognition, without building or hosting those models directly.

Artificial General Intelligence (AGI) is a theoretical form of AI that would possess the ability to understand, learn, and apply intelligence across any intellectual task at a level equal to or exceeding human capability.

Artificial intelligence (AI) is the branch of computer science focused on building machines that can perform tasks that typically require human intelligence, such as learning, reasoning, problem-solving, understanding language, and recognizing patterns.

An autonomous system is a system capable of perceiving its environment, making decisions, and taking actions to achieve goals without continuous human direction. Autonomous systems range from self-driving vehicles to industrial robots to AI agents, unified by their ability to operate independently in dynamic, real-world environments.

An autonomous vehicle (AV) is a vehicle capable of sensing its environment and navigating without human input, using a combination of AI, sensors, and computing systems to perceive, plan, and act in real-world driving scenarios.

Backpropagation is the algorithm used to train neural networks by calculating how much each parameter (weight) in the network contributed to the prediction error, then using those gradients to update the weights in a direction that reduces the error. It makes training deep neural networks computationally feasible.

Batch size is the number of training examples processed together before a model's parameters are updated. It is a fundamental hyperparameter that controls the tradeoff between training speed, memory usage, and the quality of parameter updates during machine learning model training.

Bias in AI refers to systematic errors or unfair outcomes in AI systems caused by flawed assumptions, unrepresentative training data, or problematic design choices that lead the model to disadvantage certain groups or produce inaccurate results.

Chain-of-thought (CoT) is a prompting technique that encourages an AI model to work through a problem step by step before giving a final answer, similar to showing your work in math. This intermediate reasoning process significantly improves performance on complex logical, mathematical, and multi-step tasks.

A chatbot is a software program designed to simulate human conversation through text or voice, using pre-programmed rules, machine learning, or large language models to understand and respond to user inputs.

Classification is a supervised machine learning task where a model learns to assign input data to one of several predefined categories or classes, such as identifying whether an email is spam or whether a tumor is malignant.

Cloud AI refers to AI computing resources, services, and pre-built AI capabilities delivered over the internet through cloud platforms. It allows organizations to train and deploy AI models at scale without owning or managing physical hardware, paying instead for the compute they consume.

Clustering is an unsupervised machine learning technique that groups data points into clusters based on similarity, such that points within a cluster are more similar to each other than to points in other clusters - without using predefined category labels.

Computer vision is a field of artificial intelligence that enables computers to interpret, analyze, and make decisions based on visual information from images and videos, mimicking and often exceeding human visual perception for specific tasks.

A context window is the maximum amount of text (measured in tokens) that a language model can process at a single time, determining how much information the model can reference when generating a response.

Conversational AI is a category of artificial intelligence technology that enables computers to engage in natural, human-like dialogue through text or voice. It combines natural language understanding, dialogue management, and language generation to create interactive AI systems that can handle multi-turn conversations.

Cross-validation is a statistical technique for evaluating machine learning models by dividing the dataset into multiple subsets, training and testing the model on different combinations, to produce a more reliable estimate of real-world performance.

A data pipeline is an automated set of processes that collect, transform, validate, and move data from source systems to destinations where it can be used for AI model training, inference, or analytics. Data pipelines are the infrastructure that ensures AI systems have access to clean, timely, and appropriately formatted data.

Data preprocessing is the set of techniques used to clean, transform, and organize raw data into a format suitable for machine learning model training, directly impacting model quality and reliability.

Data privacy in AI refers to the rights of individuals to control their personal information and the obligations of organizations to protect that information when collecting, using, and sharing data for AI training and deployment.

Deep learning is a subset of machine learning that uses artificial neural networks with many layers to automatically learn hierarchical representations of data, enabling breakthroughs in image recognition, language understanding, and more.

A deepfake is a piece of synthetic media - typically a video, audio recording, or image - created using artificial intelligence to convincingly depict a real person saying or doing something they never actually said or did.

A diffusion model is a type of generative AI model that creates images, audio, or other data by learning to reverse a process of adding random noise, gradually transforming noise into coherent, high-quality outputs guided by text or other conditioning.

A digital twin is a dynamic virtual representation of a physical object, process, or system that is continuously updated with real-time data from its real-world counterpart. Digital twins use AI and simulation to model behavior, predict outcomes, and optimize performance without intervening in the physical system directly.

Dimensionality reduction is a set of techniques that transform high-dimensional data into a lower-dimensional representation while preserving as much meaningful structure as possible, making data easier to visualize, analyze, and use for machine learning.

Edge AI refers to the deployment of artificial intelligence models directly on local devices, such as smartphones, IoT sensors, cameras, and embedded systems, rather than sending data to a central cloud server for processing. This enables real-time, low-latency AI inference with improved privacy and offline capability.

An embedding is a dense numerical vector that represents a piece of data, such as a word, sentence, image, or user, in a high-dimensional space where semantically similar items are positioned close together. Embeddings allow AI systems to work with complex, unstructured data using the mathematical operations that machine learning models are designed for.

In machine learning, an epoch is one complete pass through the entire training dataset during model training. Training a model typically involves multiple epochs, allowing the model to see each training example many times and progressively refine its parameters toward better performance.

Explainable AI (XAI) is a set of methods and techniques that make the decisions and outputs of artificial intelligence systems understandable and interpretable to human users and stakeholders.

Facial recognition is a biometric AI technology that identifies or verifies a person's identity by analyzing their facial features and comparing them against a database of known faces. It combines computer vision and machine learning to map and match unique facial characteristics.

Feature engineering is the process of using domain knowledge to select, transform, and create informative input variables from raw data to improve a machine learning model's predictive performance.

Feature selection is the process of identifying and selecting the subset of input variables (features) that are most relevant and informative for a machine learning model, removing redundant or irrelevant features to improve performance and efficiency.

Few-shot learning is a machine learning approach where a model learns to perform a new task from only a small number of labeled examples, typically between two and twenty. In the context of large language models, it refers to providing a few input-output examples in the prompt to guide the model's behavior.

Fine-tuning is the process of taking a pre-trained machine learning model and continuing to train it on a smaller, task-specific dataset to adapt its capabilities for a particular domain, style, or application.

Generative AI is a category of artificial intelligence systems capable of creating new, original content - including text, images, audio, video, and code - by learning patterns from existing data and generating novel outputs based on prompts.

GPT (Generative Pre-trained Transformer) is a family of large language models developed by OpenAI, trained on vast text datasets to generate coherent and contextually appropriate text across a wide range of tasks.

A GPU (Graphics Processing Unit) is a specialized processor originally designed for rendering graphics that has become the dominant hardware for training and running AI models. Its architecture of thousands of small parallel cores makes it exceptionally efficient at the matrix operations that power deep learning.

Gradient descent is an iterative optimization algorithm used to train machine learning models by adjusting model parameters in the direction that most reduces prediction error, repeating until the model reaches its best performance.

AI hallucination is a phenomenon where a language model generates text that sounds plausible and confident but contains factually incorrect, fabricated, or nonsensical information not supported by its training data or the provided context.

AI image generation is the use of generative AI models to create original images from text descriptions, reference images, or other prompts. Models like DALL-E, Midjourney, and Stable Diffusion can produce photorealistic images, artwork, and illustrations on demand.

Inference in AI is the process of using a trained machine learning model to generate predictions, classifications, or outputs from new, unseen input data - the deployment phase that follows model training.

A language model is an AI system trained on large amounts of text to learn the statistical patterns of language, enabling it to predict likely word sequences, understand context, and generate coherent text.

A large language model (LLM) is a type of AI model trained on massive amounts of text data with billions or trillions of parameters, enabling it to understand, generate, and reason about human language across a wide range of tasks.

A loss function is a mathematical function that measures the difference between a model's predictions and the actual correct values during training. It produces a single number, the loss or error, that quantifies how wrong the model currently is, and optimization algorithms use this signal to adjust the model's parameters to improve performance.

Machine learning is a subset of artificial intelligence in which systems automatically learn and improve from experience by analyzing data, without being explicitly programmed for every possible scenario.

Mixture of Experts (MoE) is a neural network architecture where a large model is divided into many specialized sub-networks called 'experts,' with a gating mechanism that routes each input to only the most relevant experts. This allows models to scale to enormous parameter counts while keeping inference costs manageable.

MLOps, short for Machine Learning Operations, is the discipline of applying DevOps practices to the machine learning lifecycle, encompassing the processes, tools, and culture needed to reliably build, deploy, monitor, and maintain machine learning models in production.

An AI model is a mathematical system that has been trained on data to recognize patterns and make predictions, decisions, or generate outputs - the end product of the machine learning training process.

Model collapse is a phenomenon where AI models trained on data generated by other AI models progressively lose diversity and accuracy, converging toward a narrower, lower-quality output distribution. It occurs because each generation of training data amplifies errors and discards rare but important patterns from the original data.

Model deployment is the process of making a trained AI model accessible in a production environment where it can receive real inputs and generate outputs for users or systems. It encompasses serving infrastructure, latency optimization, monitoring, versioning, and the operational processes needed to keep a model running reliably at scale.

Model training is the process by which an AI model learns to perform a task by repeatedly adjusting its internal parameters in response to training data. The model makes predictions, compares them to correct answers, measures the error, and updates its weights via an optimization algorithm until performance reaches an acceptable level.

Multimodal AI refers to artificial intelligence systems that can understand and generate content across multiple types of data, including text, images, audio, video, and code. These systems integrate information from different modalities together rather than treating each type separately.

Named entity recognition (NER) is a natural language processing task that identifies and classifies named entities in text - such as people, organizations, locations, dates, and quantities - into predefined categories.

Natural language processing (NLP) is a branch of artificial intelligence focused on enabling computers to understand, interpret, manipulate, and generate human language in both text and speech forms.

A neural network is a computational system loosely modeled on the human brain, consisting of interconnected layers of nodes (neurons) that process and transform data to recognize patterns, make predictions, or generate outputs.

Object detection is a computer vision task where an AI model identifies and localizes multiple objects within an image or video frame, drawing bounding boxes around each detected object and classifying what each object is.

Open-source AI refers to artificial intelligence models, frameworks, and tools whose code, and in many cases model weights, are made publicly available for anyone to inspect, use, modify, and distribute. This stands in contrast to proprietary AI systems accessible only through commercial APIs with closed weights.

Overfitting is a machine learning problem where a model learns the training data too well, including its noise and random fluctuations, resulting in excellent performance on training data but poor generalization to new, unseen data.

Predictive analytics is the practice of using statistical algorithms, machine learning, and historical data to identify patterns and forecast future outcomes, trends, or behaviors.

Prompt engineering is the practice of crafting and optimizing the inputs given to AI language models to elicit more accurate, useful, and relevant outputs for specific tasks or applications.

A recommendation system is an AI system that predicts and suggests items, content, or actions that a specific user is likely to find relevant or valuable, based on their past behavior, preferences, and patterns from similar users.

Regression is a supervised machine learning task where a model learns to predict a continuous numerical output, such as a house price, stock value, or temperature, based on input features.

Reinforcement learning is a machine learning paradigm in which an agent learns to make decisions by interacting with an environment, receiving rewards for desirable actions and penalties for undesirable ones, gradually optimizing its behavior.

Reinforcement Learning from Human Feedback (RLHF) is a training technique that uses human evaluators to rate model outputs, then trains a reward model on those ratings, and finally uses reinforcement learning to fine-tune the AI model to maximize the learned reward. RLHF is the primary method used to align language models with human preferences for helpfulness, honesty, and safety.

Responsible AI is a framework of principles and practices for developing, deploying, and governing AI systems in a way that is ethical, fair, transparent, accountable, and beneficial to individuals and society.

Retrieval-Augmented Generation (RAG) is a technique that combines a language model with a retrieval system, allowing the AI to search a knowledge base for relevant documents before generating a response. This grounds the output in real, up-to-date information rather than relying solely on what the model memorized during training.

Robotic Process Automation (RPA) is a technology that uses software bots to automate repetitive, rule-based digital tasks by mimicking human interactions with user interfaces, such as clicking buttons, entering data, reading screens, and moving files between applications. RPA does not require changes to existing systems because it operates at the UI layer.

Sentiment analysis is a natural language processing technique that automatically identifies and extracts the emotional tone or opinion expressed in text, typically classifying it as positive, negative, or neutral.

A small language model (SLM) is a language model with significantly fewer parameters than frontier large language models, typically ranging from 1 billion to 10 billion parameters, designed to be faster, cheaper to run, and deployable on devices with limited compute resources while still performing well on targeted tasks.

Speech recognition (also called automatic speech recognition or ASR) is AI technology that converts spoken language into text by analyzing audio signals and matching them to linguistic patterns learned from training data.

Supervised learning is a machine learning paradigm in which a model is trained on a labeled dataset, learning to map input data to correct outputs by studying input-output pairs provided by a human supervisor.

Synthetic data is artificially generated data that mimics the statistical properties of real-world data, created algorithmically rather than collected from actual events or people. It is used to train, test, and augment AI models when real data is insufficient, too sensitive to use, or too expensive to collect.

Temperature is a parameter in language model inference that controls the randomness of text generation - lower values produce more predictable, conservative outputs, while higher values produce more creative, diverse, and unpredictable outputs.

Text classification is a natural language processing task that automatically assigns predefined categories or labels to text documents based on their content, enabling large-scale automated organization and routing of textual information.

Text generation is the AI capability to automatically produce human-readable text - such as articles, code, summaries, or responses - by predicting and outputting sequences of words that are coherent and contextually appropriate.

A token is the basic unit of text that language models process, typically corresponding to a word, part of a word, or a punctuation character, used as the fundamental input and output element in language model computations.

Tokenization is the process of splitting text into smaller units called tokens - such as words, subwords, or characters - that serve as the basic inputs for natural language processing models.

A TPU (Tensor Processing Unit) is a custom application-specific integrated circuit (ASIC) developed by Google specifically to accelerate machine learning workloads, particularly the matrix operations at the heart of deep learning. TPUs are optimized for the specific computational patterns of neural network training and inference, offering higher throughput and energy efficiency than general-purpose GPUs for compatible workloads.

Training data is the collection of examples, labels, and information that a machine learning model learns from during the training process, directly determining how well the model performs on real-world tasks.

Transfer learning is a machine learning technique where a model pre-trained on a large dataset is adapted for a different but related task, leveraging learned knowledge to achieve high performance with much less data and training time.

A transformer is a deep learning architecture that uses self-attention mechanisms to process entire sequences of data in parallel, revolutionizing natural language processing and becoming the foundation for all modern large language models.

Unsupervised learning is a machine learning paradigm where a model discovers hidden patterns, structures, or groupings in data without any labeled examples or predefined correct answers to guide the learning process.

A vector database is a specialized database system designed to store, index, and efficiently search high-dimensional numerical vectors called embeddings. It enables semantic similarity search, allowing AI systems to find information based on meaning rather than exact keyword matches.

A voice assistant is an AI system that uses speech recognition, natural language understanding, and text-to-speech technology to interact with users through spoken language, enabling hands-free control of devices and access to information.

A word embedding is a dense numerical vector representation of a word that encodes its semantic meaning, allowing machine learning models to process text and understand relationships between words mathematically.

Zero-shot learning is the ability of an AI model to perform a task it has never seen an explicit example of, relying entirely on its general knowledge and the description of the task provided in the prompt. No task-specific examples are given to the model at inference time.