Glossary

Agentic AI refers to autonomous or semi-autonomous AI systems (often called "agents") that can reason, plan, and take action to achieve specific goals. These agents can break down tasks, make decisions, use multiple tools or data sources, and execute steps without needing continuous human direction.

AI Ethics encompasses the principles, guidelines, and frameworks that ensure artificial intelligence is developed, deployed, and used responsibly. It focuses on preventing bias, discrimination, and harm while promoting fairness, transparency, accountability, privacy, and societal well-being.

AI hallucination occurs when a generative AI model produces information that sounds plausible but is factually incorrect, misleading, or entirely fabricated. It reflects the model's tendency to generate confident answers even when reliable data is lacking.

Artificial Intelligence refers to the simulation of human cognitive processes by machines, particularly computer systems, that can perform tasks such as learning, reasoning, problem-solving, perception, and decision-making.

Computer vision is a field of artificial intelligence that enables computers to interpret and understand visual information from the world. It involves teaching machines to process images or videos, recognize patterns, identify objects, and make decisions based on what they see. In practical terms, computer vision powers things like facial recognition, barcode scanning, medical image analysis, self-driving cars, and quality inspection in manufacturing.

Data bias occurs when the training data fails to accurately represent the real-world population or context, causing AI models to learn distorted patterns. This often leads to skewed, unreliable, or unfair outputs that negatively affect certain groups or scenarios.

Data labeling is the process of annotating raw data, such as text, images, audio, or video, with meaningful tags or classifications so it can be used to train supervised machine learning models. These labels provide the "ground truth" that helps models learn to recognize patterns and make accurate predictions.

Deep learning is a specialized subfield of machine learning that uses multi-layered neural networks (often called "deep" neural networks) to learn and represent complex patterns in large datasets. These models excel at tasks involving images, audio, text, sensor data, and other unstructured information because the layered architecture automatically extracts increasingly abstract features as data moves through the network.

Embeddings are mathematical representations that convert text, images, or other types of data into numerical vectors. These vectors capture the semantic meaning and relationships between pieces of information, enabling AI models to compare, search, and understand data based on similarity rather than exact wording or appearance.

Explainable AI (XAI) refers to techniques and systems that make the decisions and behaviors of AI models transparent, interpretable, and understandable to humans. Its goal is to reveal how and why a model arrives at specific outcomes, helping users build trust, validate results, and identify potential errors or biases.

Generative AI refers to artificial intelligence models designed to create new content, such as text, images, audio, video, or code, by learning patterns from large training datasets. These models respond to user prompts to produce original outputs that resemble the data they were trained on, enabling applications like content creation, design, simulation, and automated decision support.

Large Language Models (LLMs) are advanced AI systems trained on massive collections of text to understand context, follow instructions, and generate human-like language. They can perform a wide range of language tasks, from answering questions and summarizing content to writing, coding, and reasoning. Common examples include OpenAI's GPT models and Google's Gemini models.

Definition: Machine Learning is a subset of AI that uses statistical algorithms to enable systems to automatically learn from data and improve their performance without being explicitly programmed to do so.

Model drift occurs when an AI model's performance declines over time because real-world data begins to differ from the data it was originally trained on. As patterns shift, the model becomes less accurate, making ongoing monitoring and retraining essential to maintain reliability.

Model inference is the process of using a trained AI model to generate predictions, classifications, or insights from new data it hasn't seen before. It represents the model's real-world application, where learned patterns are applied to produce useful outputs.

Model training is the process of feeding data into an algorithm so it can learn the patterns, relationships, and rules needed to make accurate predictions or classifications. During training, the model adjusts its internal parameters to reduce errors, improving its performance over time.

Multimodal AI processes and interprets multiple types of data at the same time — such as text, images, audio, video, and sensor signals — to generate richer, more context-aware outputs. By integrating insights across different modalities, these systems can understand complex scenarios more holistically than models limited to a single data type.

NLP is a field of AI that enables computers to understand, interpret, and generate human language in both text and voice formats. It combines linguistics, machine learning, and deep learning to process text or speech in a way that is meaningful and useful.

Neural networks are computing systems inspired by the structure of the human brain. They consist of interconnected nodes (neurons) that process input data and generate outputs based on learned weights. During training, the network adjusts the strength of connections (weights) between neurons to reduce errors in its predictions. Neural networks are designed to learn complex relationships in data, especially when the patterns aren't easily captured by traditional algorithms.

Reinforcement Learning (RL) is an AI training approach in which an agent learns optimal behaviors through trial and error. By interacting with an environment and receiving rewards or penalties based on its actions, the agent gradually discovers strategies that maximize long-term success.

Supervised Learning trains models using labeled data, where the correct outputs are already known. While unsupervised Learning works with unlabeled data, allowing models to discover hidden patterns, relationships, or groupings without predefined answers.

Tokenization is the process of breaking text or other data into smaller units, called tokens, so that AI models can analyze and process the information. These tokens may represent characters, words, subwords, or symbols, depending on the model's design.

Transfer learning allows a pre-trained AI model to be adapted for new tasks with relatively little additional training. By leveraging knowledge learned from a large, general dataset, the model can achieve strong performance on a related but more specific task using far less data and compute.