AI Glossary

A #

Activation Function: A mathematical function applied to a neural network’s node output to introduce non-linearity.

Active Learning: A machine learning approach where the model can query a user or some other information source to obtain the desired outputs for new data points.

Adversarial Attack: Techniques that manipulate input data to deceive machine learning models.

Algorithm: A step-by-step procedure or formula for solving a problem, typically implemented by a computer.

Artificial General Intelligence (AGI): A type of AI that possesses the ability to understand, learn, and apply intelligence across a wide range of tasks at a human level.

Artificial Neural Network (ANN): A computational model inspired by the human brain’s neural networks, used in machine learning.

Attention Mechanism: A technique used in neural networks, particularly in transformers, to focus on specific parts of the input sequence when making predictions.

Automation: The use of technology to perform tasks without human intervention, often enhanced by AI and machine learning.

Autonomy: The capability of a system to operate independently without human intervention.

Autoencoder: A type of neural network used for unsupervised learning of efficient codings.


B #

Backpropagation: An algorithm for training neural networks by adjusting weights based on error rates obtained in previous epochs.

Bayesian Network: A graphical model representing probabilistic relationships among variables.

Batch Normalization: A technique to improve the speed and stability of artificial neural networks by normalizing the input layer.

BERT (Bidirectional Encoder Representations from Transformers): A pre-trained transformer model designed for natural language understanding tasks.

Bias: Prejudices or favoritism in AI systems, often resulting from biased training data or algorithms.

Bias Detection and Mitigation: Techniques and processes used to identify and reduce bias in AI models.

Bias-Variance Tradeoff: A property of models in machine learning that reflects the tradeoff between bias (error from overly simplistic models) and variance (error from models that are too complex).

Big Data: Large and complex datasets that require advanced methods and technologies for analysis and processing.

Black Box: An AI system whose internal workings are not visible or understandable to users, raising concerns about transparency and accountability.

Bot: A software application that runs automated tasks over the internet, often leveraging AI to perform complex functions.


C #

Chatbot: An AI-driven software application that simulates human conversation, commonly used in customer service and virtual assistants.

Cloze Task: A type of fill-in-the-blank question used to evaluate language models like GPTs.

Clustering: A type of unsupervised learning where data is grouped based on similarity.

Context Window: The span of input tokens that a language model can consider at once when generating text.

Convolutional Neural Network (CNN): A class of deep neural networks primarily used for processing structured grid data like images.

Cross-Validation: A technique for assessing how the results of a statistical analysis will generalize to an independent dataset.

CRISP-DM (Cross-Industry Standard Process for Data Mining): A widely used methodology for data mining projects.

Cybernetics: The interdisciplinary study of regulatory systems, their structures, constraints, and possibilities, often focusing on feedback loops.

Cyberspace: The virtual environment created by interconnected digital technologies and networks.

Cyborg: A being with both organic and biomechatronic body parts, often used in discussions of post-humanism.

Culture Jamming: The act of disrupting or subverting media and cultural norms, often in relation to digital and cyber cultures.


D #

Data Augmentation: Techniques used to increase the diversity of data available for training models without actually collecting new data.

Data Labeling: The process of tagging or annotating raw data (images, text, videos) to prepare it for training AI models.

Data Mining: The process of discovering patterns and knowledge from large amounts of data.

Data Pipeline: A set of processes that transform raw data into a format suitable for analysis and machine learning.

Deep Belief Network (DBN): A type of generative neural network composed of multiple layers of stochastic, latent variables.

Deep Learning: A subset of machine learning involving neural networks with many layers, enabling the analysis of complex data.

Dialogflow: A natural language understanding platform for building conversational interfaces.

Dropout: A regularization technique used to prevent overfitting in neural networks by randomly setting a fraction of input units to zero at each update during training.


E #

Embedding Layer: A neural network layer that converts categorical data into continuous vector representations.

End-to-End Learning: A learning approach where a model is trained to perform a task directly from input to output, without intermediate steps.

Entity Recognition: The process of identifying and classifying key information (entities) in text.

Epoch: One complete pass through the entire training dataset during the learning process.

Ethics in AI: The study of moral values and principles in the design, development, and use of AI technologies.

Expert System: A computer system that emulates the decision-making ability of a human expert.

Exploratory Data Analysis (EDA): Analyzing datasets to summarize their main characteristics, often using visual methods.


F #

Feature Engineering: The process of using domain knowledge to extract features from raw data for machine learning.

Feature Map: A representation of the input data in a convolutional neural network layer.

Few-Shot Learning: A machine learning approach where the model is trained to recognize patterns with only a few examples.

Fine-Tuning: Adjusting a pre-trained model to specialize it on a new task or dataset.

Flowchart Automation: Using AI to create and manage flowcharts that automate decision-making processes.

F1 Score: A measure of a test’s accuracy, considering both the precision and the recall.


G #

GenAI: Generative Artificial Intelligence.

Generative Adversarial Networks (GANs): A class of machine learning frameworks where two neural networks compete to generate realistic data.

Generative Model: A type of model that generates new data instances based on the training data distribution.

Generative Pre-trained Transformer (GPT): A type of language model developed by OpenAI that uses transformer networks to generate human-like text.

Governance: The frameworks and policies that regulate the development and deployment of AI technologies.

Gradient-Based Optimization: Techniques used to minimize or maximize an objective function by iteratively adjusting model parameters.

Gradient Descent: An optimization algorithm used to minimize the cost function in training machine learning models.

Graph Neural Network (GNN): A type of neural network that operates on graph structures.

Ground Truth: The accurate, real-world data or labels used to train and evaluate machine learning models.


H #

Human-in-the-Loop (HITL): An approach where human judgment and input are integrated into the training and operation of AI systems.

Hybrid Intelligence: The combination of human and artificial intelligence to enhance decision-making and problem-solving.

Hyperparameters: Configuration settings used to control the learning process of machine learning models, such as learning rate and batch size.

Hyperparameter Tuning: The process of optimizing the hyperparameters of a machine learning model to improve performance.

Hyperreality: A condition in which the distinction between reality and simulation becomes blurred, often discussed in cybercultural studies.


I #

Image Recognition: The process of identifying and detecting an object or feature in a digital image or video.

Informatics: The study of information processing, particularly with respect to systems and technologies.

Instance Segmentation: A computer vision task that identifies objects in an image and segments them at the pixel level.

Intent Recognition: Identifying the underlying intention behind a user’s input in natural language processing tasks.

Interactive AI: AI systems designed to interact with users in real-time, often used in customer service and entertainment.

Iterative Prompting: A technique in prompt engineering where prompts are refined iteratively to improve the quality of generated responses.

IoU (Intersection over Union): A metric used to evaluate the accuracy of object detection systems by comparing the overlap between the predicted and ground truth bounding boxes.

Intelligent Agent: An autonomous entity that observes and acts upon an environment to achieve specific goals.


J #

Job Scheduling: Automating the process of planning and allocating resources for tasks, often using AI to optimize efficiency.

Just-in-Time Learning: An educational approach that delivers information as it is needed, often facilitated by AI.

Jupyter Notebook: An open-source web application that allows the creation and sharing of documents containing live code, equations, visualizations, and narrative text.


K #

Keyword Extraction: Identifying significant words or phrases in a text, commonly used in natural language processing.

K-Means Clustering: A method of vector quantization that partitions n observations into k clusters.

k-Nearest Neighbors (k-NN): A simple, instance-based learning algorithm that classifies data points based on the k closest training examples.

Knowledge Distillation: A technique where a smaller model learns to mimic a larger, more complex model to achieve similar performance with fewer resources.

Knowledge Representation: The field of AI concerned with how knowledge can be represented symbolically and manipulated in automated ways.


L #

Language Model: A model that predicts the probability distribution of words in a sentence, used in natural language processing.

Latent Space: The abstract multi-dimensional space where input data is projected and analyzed by a machine learning model.

Learning Algorithms: Algorithms that enable machines to learn from data and improve their performance over time.

Learning Rate: A hyperparameter that controls how much to change the model in response to the estimated error each time the model weights are updated.

Linear Automation: The sequential automation of tasks, typically without conditional branching or complex decision-making.

Liquid Modernity: A concept describing the fluid and ever-changing nature of modern life, relevant in post-digital and post-humanist theories.

Logistic Regression: A statistical method for analyzing a dataset in which there are one or more independent variables that determine an outcome.

LSTM (Long Short-Term Memory): A type of recurrent neural network capable of learning long-term dependencies.


M #

Machine Learning (ML): A subset of AI focused on building systems that learn from data and improve their performance.

Media Archaeology: The study of media history and technology from a non-linear perspective, often used in cybercultural studies.

Meta-Learning: A machine learning approach where algorithms learn to learn, improving their ability to adapt to new tasks.

Minibatch: A subset of the training data used to update the model parameters in an iteration of gradient descent.

Model Compression: Techniques used to reduce the size of machine learning models while maintaining performance.

Model Drift: The degradation of model performance over time due to changes in the underlying data distribution.

Multi-Task Learning: Training a model on multiple tasks simultaneously to improve generalization and efficiency.

Multilayer Perceptron (MLP): A class of feedforward artificial neural network consisting of at least three layers of nodes.

Model Compression: Techniques used to reduce the size of machine learning models while maintaining performance.

Multi-Task Learning: Training a model on multiple tasks simultaneously to improve generalization and efficiency.


N #

Neural Architecture Search (NAS): The process of automating the design of artificial neural networks.

Natural Language Processing (NLP): The field of AI that focuses on the interaction between computers and human languages.

Neural Network: A computational model inspired by the human brain, consisting of interconnected nodes (neurons) that process information.

Normalization: Scaling input data to improve the performance and training stability of machine learning models.


O #

Ontology: In AI, a structured framework to categorize and define the relationships between concepts within a domain, often used in AI to enhance understanding and reasoning.

Open Source AI: AI technologies and systems whose source code is made available for public use and modification.

Optimization: The process of adjusting model parameters to minimize or maximize an objective function.

Overfitting: When a model learns the training data too well, capturing noise and details that negatively impact performance on new data.

Optimizer: Algorithms or methods used to change the attributes of the neural network, such as weights and learning rate, to reduce losses.


P #

PCA (Principal Component Analysis): A dimensionality-reduction technique used to reduce the complexity of data while retaining most of the variation.

Perceptron: The simplest type of artificial neural network, consisting of a single layer of output nodes.

Post-digital: A perspective that examines the integration of digital technologies into all aspects of life, moving beyond the novelty of the digital.

Post-humanism: A philosophical approach that explores the boundaries and relationships between humans and technology.

Predictive Analytics: Techniques that use data, statistical algorithms, and machine learning to identify the likelihood of future outcomes.

Preprocessing: The process of transforming raw data into a format suitable for machine learning.

Pre-trained Model: A machine learning model that has been previously trained on a large dataset and can be fine-tuned for specific tasks.

Prompt Engineering: The design and refinement of input prompts to guide AI models, particularly language models, to generate desired outputs.

Provenance: Tracking the origin and history of data, including how it has been processed and transformed, to ensure transparency and accountability in AI systems.


Q #

Q-Learning: A model-free reinforcement learning algorithm that seeks to learn the quality of actions, telling an agent what action to take under what circumstances.

Quantization: Reducing the precision of the numbers used in a machine learning model to decrease its size and improve efficiency.

Quantum Computing: A type of computing that uses quantum-mechanical phenomena to perform operations on data, potentially revolutionizing AI.

Query Expansion: Enhancing a search query with additional terms to improve retrieval performance.


R #

Recurrent Neural Network (RNN): A class of neural networks where connections between nodes form directed cycles, used for sequence prediction.

Reinforcement Learning (RL): A type of machine learning where an agent learns to make decisions by performing certain actions and receiving rewards or penalties.

Regularization: Techniques used to prevent overfitting by adding additional information to the model.

Reproducibility: The ability to consistently reproduce the results of a machine learning experiment or study, ensuring reliability and validity.

Robotics: The branch of technology that deals with the design, construction, operation, and application of robots.

Rule-Based System: An AI system that uses predefined rules to make decisions or solve problems, often contrasted with learning-based approaches.


S #

Singularity: A hypothetical future point where technological growth becomes uncontrollable and irreversible, often associated with superintelligent AI.

Semi-Supervised Learning: A machine learning approach that combines a small amount of labeled data with a large amount of unlabeled data during training.

Social Computing: The study of how social behaviors and interactions can be modeled and facilitated by computational systems.

Supervised Learning: A type of machine learning where the model is trained on labeled data, learning to map inputs to known outputs.

Support Vector Machine (SVM): A supervised learning model used for classification and regression analysis.

Surveillance Capitalism: A critique of how data and information are commodified and used for profit, often implicating AI technologies.

Swarm Intelligence: The collective behavior of decentralized, self-organized systems, often used in AI to solve complex problems.


T #

TensorFlow: An open-source library for machine learning and artificial intelligence developed by Google.

Tokenization: The process of converting a sequence of text into individual units (tokens), such as words or subwords, for processing by machine learning models.

Transfer Learning: A technique where a pre-trained model is reused on a new, but related problem.

Transformer: A type of neural network architecture designed for processing sequences of data, particularly effective in NLP tasks.

Transhumanism: A movement advocating for the transformation of the human condition through advanced technologies.

Trustworthy AI: AI systems designed to be reliable, transparent, and aligned with human values and ethical principles.

Turing Test: A test proposed by Alan Turing to determine whether a machine exhibits intelligent behavior equivalent to, or indistinguishable from, that of a human.


U #

Underfitting: When a model is too simple to capture the underlying structure of the data, resulting in poor performance.

U-Net: A convolutional neural network architecture designed for biomedical image segmentation tasks.

Update Step: In machine learning, the process of adjusting the model parameters based on the gradient of the loss function.

Unsupervised Learning: A type of machine learning where the system learns patterns from untagged data without explicit instructions.


V #

Validation Set: A subset of the data used to evaluate the model’s performance during training and tune hyperparameters.

Vanishing Gradient Problem: A difficulty encountered when training deep neural networks where gradients used for updating weights diminish exponentially as they propagate back through layers.

Variance: The amount by which the model’s predictions would change if we used a different training dataset, reflecting the model’s sensitivity to the specific training data.

Variational Autoencoder (VAE): A type of autoencoder that learns a probabilistic mapping from input data to a latent space and can generate new data samples.

Vectorization: The process of converting data into a vector (numerical format) that can be processed by machine learning algorithms.

Vernacular AI: AI technologies adapted to local languages, cultures, and contexts, emphasizing inclusivity and accessibility.

Virtual Assistant: An AI-powered system designed to assist users by performing tasks or providing information, often through natural language interaction.

Virtual Reality (VR): A simulated experience created by computer technology, immersing users in a virtual environment.


W #

Wearable Technology: Electronic devices worn on the body, often incorporating AI to monitor and enhance user experiences.

Weak AI: AI systems designed for narrow, specific tasks, lacking general cognitive abilities.

Weight Initialization: The process of setting the initial values of the weights in a neural network before training begins.

Weight Sharing: A technique in neural networks where multiple nodes share the same weights, reducing the number of parameters and computational complexity.

Word2Vec: A set of models that produce word embeddings by training words against their contexts, capturing semantic meanings.

Word Embedding: A type of word representation that allows words to be represented as vectors in a continuous vector space.

Workflow Automation: The use of AI to automate complex business processes, improving efficiency and accuracy.


X #

XAI (Explainable AI): AI systems designed to be transparent and understandable, enabling users to comprehend how decisions are made.

XGBoost: An optimized gradient boosting framework designed to be highly efficient, flexible, and portable.

XOR Problem: A problem in machine learning that demonstrates the limitations of a single-layer perceptron, where the output is true if and only if the inputs differ.


Y #

Yield Curve: A graphical representation of interest rates across different contract lengths for a similar debt contract, used metaphorically in AI to describe performance over time.

Yield Optimization: The use of AI to improve the efficiency and productivity of processes, particularly in manufacturing and agriculture.

YOLO (You Only Look Once): A real-time object detection system that predicts bounding boxes and class probabilities directly from full images in a single evaluation.


Z #

Zero-day Exploit: A vulnerability in software that is unknown to the developers and exploited by hackers, highlighting cybersecurity concerns in AI.

Zero-Shot Learning: A machine learning task where the model is required to make predictions on new classes that were not observed during training.

Zettabyte: A unit of digital information storage equivalent to one sextillion (10^21) bytes, illustrating the scale of big data in AI.

Z-score Normalization: A statistical technique used to standardize data by subtracting the mean and dividing by the standard deviation.