A dictionary of AI terms with some fundamental concepts to avoid getting lost in the tide of a revolutionary technology.

Artificial intelligence is already an active part of many of our activities. Although we use it, we often don't really know how it works. Below you will find some concepts that can help you understand how it works.

Anchoring 

Anchoring is the process of anchoring AI systems to real-world experiences, knowledge, or data. The objective is to enhance AI’s understanding of the world, so it can effectively interpret and respond to user input, queries, and tasks. Anchoring helps AI systems become more context-aware, allowing them to provide more appropriate, relatable, and relevant responses or actions.

Automation

Automation refers to the use of technology to perform various tasks with minimal human intervention.

Artificial Intelligence (AI)

Technology that simulates human intelligence in machines that are programmed to think and learn like humans. Example: self-driving cars that can navigate and make decisions on their own using AI technology.

Artificial General Intelligence (AGI)

Artificial General Intelligence (AGI) refers to an AI system that possesses a broad range of cognitive abilities, much like human ones, allowing it to learn, reason, adapt to new situations, and develop creative solutions across a variety of tasks and domains, rather than being limited to specific tasks as narrow AI systems are.

Associative memory

Associative memory refers to a system's ability to store, retrieve, and process related information based on connections between elements, allowing it to efficiently identify and use relevant data for decision making.

AI plugin

AI plugins are specialized software components that enable AI systems to interact with external applications and services.

AI Agents

An artificial intelligence (AI) agent is a software that interacts with its environment, gathers data, and uses it to perform tasks aimed at achieving set goals. While humans define the goals, the AI independently decides the best actions to take.

AIOps

AIOps, or "artificial intelligence for IT operations," refers to platforms that use machine learning (ML) and analytics to automate various IT processes. By harnessing large volumes of data from operational devices, AIOps can quickly identify and respond to issues in real time.

Automatic voice recognition

Automatic speech recognition (ASR) is a technology that transcribes speech into text.

Backpropagation 

It is an algorithm frequently used in the training of artificial neural networks. It adjusts the weights of the connections between neurons to minimize the error in the prediction of a model. It is effective for, for example, pattern recognition or image classification.

Base model in AI

Base models are a broad category of AI models that include large language models and other types of models such as computer vision and reinforcement learning models. They are called “base” models because they serve as the building block on which to build applications, serving a wide range of domains and use cases.

Collective learning

Collective learning is an AI training approach that leverages diverse skills and knowledge across multiple models to achieve more robust intelligence.

Chatbot

An easy-to-use interface that allows the user to ask questions and receive answers. Depending on the back-end system powering the chatbot, it can be as basic as prescribed responses to a fully conversational AI that automates problem resolution.

ChatGPT

A chat interface built on top of GPT-3.5. GPT-3.5 is a large language model developed by OpenAI that is trained on massive amounts of text data from the internet and optimized to perform a wide range of natural language tasks. Example: GPT-3.5 has been fine-tuned for tasks such as language translation, text synthesis, and question answering.

Conversational AI

A branch of AI development focused on building systems that can understand and generate human-like language and carry out a two-way conversation. An example of this would be a chatbot that could understand and respond to customer queries in a manner similar to a human-to-human conversation.

Cross validation

It is a method used to evaluate the results of an analysis and ensure the accuracy of the predictions of machine learning models.

Data Augmentation

Data augmentation is the technique of generating additional data artificially from existing datasets, mainly to improve the training of machine learning (ML) models.

Data enrichment

Data enrichment is a technique used to artificially increase the size and diversity of a training set by creating modified copies of existing data. It involves making minor changes such as flipping, resizing, or adjusting the brightness of images, to improve the dataset and prevent overfitting of models.

Deterministic model

A deterministic model follows a specific set of rules and conditions to achieve a definitive result, operating on a cause and effect basis.

Discriminative model

Discriminative models are algorithms designed to directly model and learn the boundary between different classes or categories in a data set.

Deep learning

A subfield of machine learning that uses multi-layered neural networks to learn from data. An example would be a deep learning model that could recognize objects in an image by processing it through multiple layers of neural networks.

Extensibility in Artificial Intelligence

Extensibility in AI refers to the ability of AI systems to expand their capabilities to new domains, tasks, and data sets without requiring complete retraining or major architectural changes.

Enhanced learning 

A form of machine learning where a model learns to make decisions by interacting with its environment and receiving feedback through rewards or penalties. For example, GPT uses boosted learning from human feedback. When tuning GPT-3, human annotators provided examples of the desired model behavior and classified the model outputs.

Fine tuning

Fine-tuning involves adapting a pre-trained model for a specific task by training it on a smaller dataset. For example, an image classification model trained on all photos of intersections can be fine-tuned to detect when a car runs a red light.

GPT-3 

GPT-3 is the third version of the GPT-n model series. It has 175 billion parameters (adjustable knobs) that can be weighted to make predictions. Chat-GPT uses GPT-3.5, which is another iteration of this model.

GPT-4 

GPT-4 is the latest model addition to OpenAI’s deep learning efforts and is a significant milestone in the scalability of deep learning. GPT-4 is also the first GPT model that is a large multimodal model, meaning it accepts both image and text inputs and emits text outputs.

Generalization from weak to strong 

Weak-to-strong generalization is an AI training approach that uses less capable models to guide and constrain more powerful ones toward better generalization beyond their narrow training data.

Generative AI 

Generative AI models create new data or content by discovering patterns in data inputs or training data. For example, creating an original short story based on the analysis of existing published short stories.

Generative adversarial networks

Generative adversarial networks are a powerful example of a neural network, capable of generating new, never-before-seen information that closely resembles the training data.

Hallucination 

Hallucination refers to a situation where an AI system, especially in cases where it is tasked with natural language processing, starts generating output that may be irrelevant, nonsensical, or incorrect based on the input provided. This often occurs when the AI system is unsure of the context, relies too heavily on its training data, or lacks a proper understanding of the subject matter at hand.

Intelligent Document Processing IDP

Intelligent document processing (IDP) refers to the technology that extracts and structures data from documents to enhance business process automation.

Inference 

The process of making predictions with a trained machine learning model, that is, putting into practice what the AI has learned during its training.

Latent space 

In machine learning, this term refers to the compressed representation of data that a model (such as a neural network) creates. Similar data points are closer together in the latent space.

Large Language Models (LLM) 

A type of deep learning model trained on a large dataset to perform natural language understanding and generation tasks. There are well-known models such as BERT, PaLM, GPT-2, GPT-3, GPT-3.5, and the innovative GPT-4. All of these models vary in size (number of parameters that can be tuned), breadth of tasks (coding, chat, scientific, etc.), and the knowledge they have been trained on.

Machine learning 

A subfield of AI that encompasses the development of algorithms and statistical models that enable machines to improve their performance with experience. An example would be a machine learning algorithm that could predict the likelihood of customer churn based on past behavior.

Model chaining 

Model chaining is a technique in data science where multiple machine learning models are linked in sequence to make predictions or analysis.

Multi-hop reasoning 

Multi-hop is a term often used in natural language processing, and more specifically in those tasks related to machine reading comprehension. It refers to the process by which an AI model retrieves answers to questions by connecting multiple pieces of information present in a given text or across multiple sources and systems, rather than directly extracting information from a single passage.

Natural Language Processing

A subfield of AI that involves programming computers to process massive volumes of linguistic data. It focuses on transforming free-form text into a standardized structure.

Neural network 

A machine learning model inspired by the structure and function of the human brain, which is composed of layers of interconnected nodes or “neurons.” They allow pattern recognition and problem solving, based on machine training or deep learning, seeking to model the analysis mechanisms of the human brain.

OpenAI 

This is the company that developed ChatGPT. OpenAI is, broadly speaking, a research company that aims to responsibly develop and promote friendly AI. One example is their GPT-3 version of OpenAI, one of the largest and most powerful language models available for natural language processing tasks.

Overfitting 

A problem that arises when a model is overly complex, performing well on the training data but struggling with unseen data. For example, a model that memorizes the training data instead of learning general patterns, and thus performs poorly on new data.

Prejudice or bias in Artificial Intelligence

Inductive bias is the set of assumptions that a machine learning algorithm makes about the underlying distribution of the data. A “bias-variance trade-off” is the balance that must be struck between the assumptions a model makes about the data and the amount that a model’s predictions change, given different training data.

Parameters 

In machine learning, parameters are internal variables that a model uses to generate predictions. They are derived from training data during the training process.

Prompt engineering 

Identifying inputs (prompts) that result in meaningful outputs. So far, prompt engineering is essential for LLMs. LLMs are an amalgamation of layers of algorithms that consequently have limited controllability with few opportunities to intervene and override behavior. One use case for prompt engineering would be to create a collection of templates and wizards to control a business copywriting application.

RAG (Retrieval-Augmented Generation)

Retrieval-Augmented Generation (RAG) optimizes a large language model's output by referencing an authoritative external knowledge base before response generation which enhance the LLM's results.

Recurring prompts 

Recursive prompts are a strategy for guiding AI models like OpenAI’s GPT-4 to produce higher-quality results. It involves providing the model with a series of prompts or questions that build on previous answers, refining both the context and the AI’s understanding to achieve the desired outcome.

Reasoning in Artificial Intelligence

AI reasoning is the process by which artificial intelligence systems solve problems, think critically, and create new knowledge by analyzing and processing available information, enabling them to make well-informed decisions across a variety of tasks and domains.

Instruction tuning 

Instruction tuning is an approach where a pre-trained model is tailored to perform specific tasks by providing a set of guidelines or directives that describe the desired operation.

Stacking 

Stacking is a technique in AI that combines multiple algorithms to improve overall performance. By combining the strengths of multiple AI models, stacking compensates for the weaknesses of each model and achieves more accurate and robust output in various applications such as image recognition and natural language processing.

Sequence modeling 

A subfield of natural language processing that focuses on modeling sequential data such as text, speech, or time series data. In this case, a sequence model would be one that can predict the next word in a sentence or generate coherent text.

Singularity 

In the context of AI, singularity (also called technological singularity) refers to a hypothetical future point in time when technological growth becomes uncontrollable and irreversible, leading to unforeseeable changes in human civilization.

Stable Diffusion 

It is an open-source artificial intelligence system that uses deep learning to generate images from natural text input.

Text to speech 

Text-to-speech (TTS) is a technology that converts written text into spoken voice output. Users typically hear the written content read aloud, typically using a synthetic voice.

Tokenization 

The process of breaking text into individual words or subwords to feed them into a language model. Example: tokenizing a sentence “I am ChatGPT” into the words: “I,” “am,” “Chat,” “G,” and “PT.”

Transformer in Artificial Intelligence

It represents a type of neural network architecture designed to process sequential data, such as text. The transformer architecture learns context and therefore meaning. It is used in models such as ChatGPT for natural language processing tasks.

Unsupervised learning 

A type of machine learning that trains a model using unlabeled data to find patterns or features in the data. Example: An unsupervised learning algorithm that can group similar images of handwritten digits based on their visual characteristics.

Unstructured data 

Unstructured data is any information that is not organized into a predefined model or structure, making it difficult to collect, process, and analyze.

Validation data 

A subset of the dataset used in machine learning that is separate from the training and testing datasets. It is used to tune the hyperparameters (i.e. the architecture, not the weights) of a model.

Voice processing 

Speech processing in AI refers to the sequence of speech-to-text conversion followed by text-to-speech synthesis.

Zero-shot learning or zero-shot training 

Zero-training learning is a technique in which a machine learning model can recognize and classify new concepts without any labeled examples.

Ready to leverage your infrastructure foundations for AI success?

Download our whitepaper to discover how a robust IT infrastructure and data strategy can help you navigate AI complexities. PlatformDIGITAL®™ equips enterprises, service providers, and hyperscalers with optimized AI architectures tailored to meet your AI application and enterprise storage challenges. Digital Realty’s AI-Ready High-Density Colocation data centers to support your legacy systems and large AI workloads.

Wanscale kiest ServiceFabric® van Digital Realty voor hun connectiviteitsplatform