Quick view: Key AI terms (USA)

Introduction

This Quick view provides in-house counsel, private practitioners, and compliance professionals with a glossary of must-know terms and explanations relevant to artificial intelligence (AI). There is a lot of jargon surrounding AI and the glossary aims to simplify the technical definitions and is intended to be a useful reference point. It combines definitions from terms that have been sourced from industry texts and reference guides.

This Quick view can be used in conjunction with the following How-to guides: Understanding AI-driven risks, Risks and liabilities of AI algorithmic bias; Checklists: Steps to mitigate risks associated with AI use in business, De-identification of data used by AI systems; and Quick view: Overview of AI in business.

Algorithm

A set of mathematical instructions or rules that, especially if given to a computer, will help to calculate an answer to a problem (Cambridge English Dictionary).

Algorithmic bias

When algorithms systematically interpret data in a way that produces unintended discriminatory results (Congressional Black Caucus Foundation). The term includes AI-driven decisions that are unfair to some people and have the potential to cause harm. For example, Amazon’s automated recruitment system for new hires judged an applicant’s qualifications by comparing qualifications with those of past candidates. Since prior candidates were overwhelmingly male, the algorithm assumed that male candidates were preferred, and so rejected applications from women.

Algorithmic efficiency

A measure of the average execution time necessary for an algorithm to complete work on a set of data (Oxford Reference). Since algorithms solve problems, efficient algorithms use the minimum possible resources and performance is measured against resources used in terms of time and memory space. The fewer resources the algorithm uses, the more efficient it is. An inefficient algorithm can lead to higher costs and delays in processing which are likely to directly impact the customer experience.

Alignment

Alignment is the process of getting AI to do the ‘right thing’ by encoding human values and goals into AI systems to mitigate risks of harm, and make the systems align with the needs and expectations of humans in a way that is ‘compatible with human moral values’ (Markkula Center for Applied Ethics, Santa Clara University).

Analytics

The method of logical analysis of data (Merriam-webster.com). Analytics are the process of interpreting data to make informed decisions and carry out instructions.

Artificial intelligence (AI)

A branch of computer science devoted to developing data processing systems that performs functions normally associated with human intelligence, such as reasoning, learning, and self-improvement.
The capability of a device to perform functions that are normally associated with human intelligence, such as reasoning, learning, and self-improvement (National Institute of Science and Technology). The autocorrect function on a word processing program or text messaging platform is an example of an application of AI.

AI ethics

AI ethics involves the considerations that stakeholders, including engineers and government officials, must address to ensure responsible development and use of AI technology. This involves implementing systems that promote a safe, secure, unbiased, and environmentally friendly approach to artificial intelligence (Coursera.org).

Autonomous agents

Software programs which respond to states and events in their environment independent from direct instruction by the user or owner of the agent, but acting on behalf and in the interest of the owner (ScienceDirect). Autonomous agents can plan and execute tasks end to end without human guidance. Self-driving cars are one example of autonomous agents entering the mainstream.

Big data

Data sets that are too large to process on a personal computer. Compared to traditional, smaller datasets that can be stored, analyzed, and easily managed on a personal computer, big data refers to datasets that are much larger, are created or added to more quickly, are more varied in their structures, and are stored on large, cloud-based storage systems (National Library of Medicine). Streaming media services such as Netflix or Spotify use big data when they analyze the choices made by millions of customers to make viewing or listening recommendations.

Black box AI

AI systems with internal workings that are invisible to the user. A user can feed them input and get output, but cannot examine the system’s code or the logic that produced the output (Scientific American). A black box is an ‘impenetrable system’ as in the absence of any explanation, it is not possible to understand how decisions have been reached. Even the designers cannot understand how the algorithms are making the predictions. While the black box AI model is very effective at making decisions, the lack of transparency presents ethical challenges such as concealing the data used by the system, so users and developers are unable to gauge the accuracy of the data, or whether the data will lead to biased or unfair results. There is also the risk that bad actors will take advantage of the lack of transparency to manipulate data.

Blockchain

Blockchains are distributed digital ledgers of cryptographically (coded or enciphered) signed transactions that are grouped into blocks. Each block is cryptographically linked to the previous one (making it tamper-evident) after validation and undergoing a consensus decision (National Institute of Science and Technology).

Chatbot

A computer program that responds like a smart entity when conversed with through text or voice and understands one or more human languages by natural language processing (Artificial Intelligence Applications and Innovations, May 2020). Chatbots are designed to simulate human conversation via text or speech. A restaurant chain’s app that lets the customer order food remotely by voice command or text and then informs the customer when their order will be ready is an example of a common chatbot.

Cognitive computing

An intelligent system that converses with and mimics the human being in a natural form by learning at scale and reasoning with purpose (Cognitive Computing for Human-Robot Interaction). Rather than carrying out specific tasks, the system is trained to carry out certain tasks in a particular way. Cognitive computing encompasses AI and signal processing and could include automated advisors used by financial services companies to tailor portfolios based on an investor’s appetite for risk. A program that analyses patient medical records and makes treatment recommendations based on that analysis is another example.

Data augmentation

A method for artificially increasing the size of a dataset by producing more data points from existing data. Techniques include cropping, flipping, or rotating images, and adding noise. In the context of AI, data augmentation is used to diversify the training data available to machine learning models with the goal of improving the model’s ability to generalize unseen data and make accurate predictions.

Data integration

The process of combining data residing at different sources and providing the user with a unified view of this data (Data integration: a theoretical perspective). This makes data more accessible and easier to process. Consolidating data from different sources can produce useful business insight and enable better informed data driven decisions.

Data mining

An analytical process that attempts to find correlations or patterns in large data sets for the purpose of data or knowledge discovery (National Institute of Science and Technology). Companies use data mining to discover more about their customer base. For example, a retailer may use records of sales of diapers to deduce which customers are more likely to buy infant formula or clothes and use that deduction to target advertisements for those products to the correct buyers. AI is useful for data mining; in that it can rapidly analyze large sets of data to return useful knowledge to the user.

Data science

Data science is the discipline that uses analytics and machine learning techniques, along with subject matter expertise, to generate insights that drive actions. Data science is a diverse field that takes in a wide range of analytical and computer science methods. It is essential for developing AI solutions.

Data science is an interdisciplinary field that merges computer science, mathematics, statistics, and domain expertise to create data-driven insights and solutions for business, operational, and strategic challenges. Often, this involves leveraging machine learning and AI (Feinstein School of Social and Natural Sciences).

Data set

A collection of separate sets of information that is treated as a single unit by a computer (Cambridge Dictionary). These separate sets of data may come from any source, and are used to analyze information and train and test algorithms to find patterns and predict outcomes.

Deep learning

A method in AI that teaches computers to process data in a way that is inspired by the human brain. Deep learning models can recognize complex patterns in pictures, text, sounds, and other data to produce accurate insights and predictions (Amazon Web Services). Deep learning is powered by many layers of ‘neural networks’ which are algorithms trained to simulate the human brain and make predictions. Most people encounter deep learning daily when they are browsing the internet or their smartphones. One example of a deep learning system is a system that can identify an image and create a coherent caption with proper sentence structure for that image (eg, YouTube videos). In health care, deep learning models can complete classification tasks such as highlighting relationships between symptoms and outcomes within vast quantities of unstructured data. Deep learning has the added benefit of being able to make decisions with significantly less involvement from human trainers.

Deep seek

DeepSeek is an open-source large language model developed in China that relies on what is known as ‘inference-time computing’, meaning that only the most relevant portions of the model is activated for each query, which saves money and computation power (CBS News).

Digital currency

A digital representation of value that exists electronically and that functions as a medium of exchange, a unit of account, and a store of value. Digital currencies are not issued or backed by government or other public authority, but are so by their developers and are used and accepted among the members of a specific virtual community (Digital Pound Foundation). In the United States, digital currency is referred to as a central bank digital currency (CBDC) and the Federal Reserve is actively exploring the topic.

Distributed AI

An approach to artificial intelligence in which processing takes place not in a single algorithm but is distributed across a number of agents, possibly many (Oxford Reference). Tasks, such as processing data, are distributed among a number of different actors, working in tandem with each other.

Explainable AI (XAI)

This type of artificial intelligence systems focuses on providing understandable explanations for decisions and actions by implementing techniques and methods to ensure that each decision made can be traced and explained (IBM).

Extractive AI

A subfield of natural language processing (NLP), this type of artificial intelligence system focuses on identifying and extracting specific information from existing data, rather than generating new content (Medium). An extractive system can be useful for locating information or data within a document or database and for document summarization.

Foundation models

AI models that are trained on a broad set of unlabeled data that can be used for different tasks, with minimal fine-tuning (IBM). Foundation models are capable of a wide range of general tasks, including being able to generate outputs (see also Generative AI). ChatGPT is an example of a chatbot which is powered by foundation models.

Frontier AI

Models that are at or beyond the current cutting edge, which ‘expand the frontier’ or ‘push into the unknown’, in that they are pushing the limits of what AI can do. They are also considered to be above a certain threshold of riskiness (Center for Security and Emerging Technology). ‘Frontier AI’ is a fluid concept, and particular examples are best defined by reference to the current state of technology.

Generative AI (GenAI)

A machine-learning model that is trained to create new data, rather than making a prediction about a specific dataset. A generative AI system is one that learns to generate more objects that look like the data it was trained on (MIT). The types of new content generated includes text, images, audio, video or other types of media. An example is an app designed for visually impaired individuals that converts images into text instantly. Users are able to send images through the app and obtain immediate identification and interpretation of those images, and also obtain conversational visual assistance. Other examples of GenAI include chatbots, photo filters, and virtual assistants.

Issue tree/logic tree

Issue trees are decision-trees that are used to break down complex problems into smaller, more manageable component parts, which helps manage analysis and problem-solving. Creation of issue trees provide AI professionals with a structured framework of issues which helps address challenges requiring analysis or resolution (Open Train AI).

Knowledge-based system

A computer program that uses a centralized repository of data known as a knowledge base to provide a method for problem-solving (TechTarget). An AI-powered customer service system is an example of a knowledge-based system, where AI and a base of expert knowledge from a central knowledge base can be useful to help customers solve their problems and help agents get answers quickly.

Large language model (LLM)

A category of AI model trained on immense amounts of data making them capable of understanding and generating natural language and other types of content (IBM). ChatGPT is an example of a system that uses LLMs.

Limited memory AI

A type of machine learning model that uses historical data and pre-programmed information to make predictions and perform classification tasks (Emerging Information & Technology Conference). . The models learn from previously acquired information, data, or past events and therefore learn from past experiences. An example is the virtual voice assistant (eg, Alexa or Siri) that uses past interactions and user preferences to provide personalized recommendations (‘Shall I play more music like that?’).

Machine learning

The field of study that gives computers the ability to learn without explicitly being programmed (MIT Sloan School of Management). Machine learning systems are based on algorithms that learn and have the capacity to make decisions by finding patterns in complex data. By learning from data, machine learning enables computers to improve through experience thus enabling predictions and decisions to become more accurate over time. Foundation models and LLMs are examples of machine learning models. A spam filter for email that finds patterns in objectionable email content is an application of machine learning.

Natural language processing (NLP)

A machine learning technology that gives computers the ability to interpret, manipulate, and comprehend human language (Amazon Web Services). This enables computers and smart devices to recognize and analyze language in text or spoken form. It can be used to streamline business operations, and includes digital assistants, chatbots, automatic translations, and voice-operated GPS.

Node

A data point in a network or graph. Nodes can be connected to other nodes by edges, which represent relationships between the data points (Autoblocks). Nodes have the capacity to receive and process data or signals. They serve as building blocks for various AI algorithms and can be a simple connection of two data points or a complex network of related data. In blockchain transactions, the nodes are computers that are a part of the chain that run the chain’s software to validate transactions and keep the network secure.

OpenAI

A private AI research and deployment company. Its stated mission is ‘to ensure that artificial general intelligence benefits all of humanity’ (OpenAI).

Reactive machine

An AI system that has no memory and is task-specific, meaning that an input always delivers the same output (Coursera). It is the most basic type of AI system as it does not have the capacity to make decisions or predictions. An example is the ‘recommendation’ feature on an entertainment streaming platform that recommends programs and movies based on the user’s prior viewing habits.

Robotics

Technology dealing with the design, construction, and operation of robots in automation (Merriam-Webster.com). When robotics and AI work together, robots are programmed to undertake tasks, and AI is used as the program or algorithm that directs the robot to perform the tasks.

Self-aware

In the AI context, self-awareness involves creating AI systems capable of being conscious of themselves. The AI system would be able to comprehend its own existence, have a sense of identity, and be aware of its thoughts and emotions (Medium). This means it would be aware of emotions and mental states to an almost human-like level. Self-awareness is largely theoretical.

Smart contract

Computer code that automatically executes all or parts of an agreement and is stored on a blockchain-based platform. The code can either be the sole manifestation of the agreement between the parties or might complement a traditional text-based contract and execute certain provisions (Harvard Law School Forum on Corporate Governance). Smart contracts are typically used to automate workflows where actions and outcomes are triggered when a set of predetermined conditions are met. Smart contracts have a wide variety of uses including supply chain management contracts that operate automatically without the need for continued human involvement, or a streaming music service that makes royalty payments to artists immediately when a song is played.

Strong AI

An AI system that can perform a variety of functions, eventually teaching itself to solve new problems. While human input accelerates the growth phase of Strong AI, it is not required, and over time, it develops a human-like consciousness instead of simulating it (IBM). Strong AI will therefore eventually be able to think, reason, and communicate like humans; however, strong systems have not yet reached that level of development.

Theory of mind

The ability to comprehend, predict, and respond to the mental states of humans and other AI agents (Medium). In relation to AI, theory of mind uses this information to draw inferences or make predictions, and a computer equipped with these skills will be able to understand the entities and individuals it interacts with. Theory of mind has not been fully developed and is still being researched and developed.

Turing test

A test to establish the existence of AI in which questions from an interrogator are answered by an unseen person and computer with the understanding that if the interrogator is unable to correctly identify which responder is human, the computer has demonstrated thinking ability comparable to a human's (Merriam-Webster.com). This test of the machine’s ability to demonstrate human-like intelligence was first introduced by Alan Turing as the ‘imitation game’ in 1950. An example is the 'CAPTCHA' checkbox on a webpage that allows the user to prove that they are not a robot (CAPTCHA is an acronym for ‘Completely Automated Public Turing Test to Tell Computers and Humans Apart’).

Weak AI

A system that focuses on performing a specific task, such as answering questions based on user input, or playing chess following the well-defined rules of the game. It can perform one type of task, but not both. Weak AI is also known as narrow AI (IBM). Weak AI is built on algorithms which have been programmed to read and interpret data and act in accordance with how they have been taught to approach problems or perform tasks. Examples include Meta’s (Facebook) news feed, or Amazon’s suggested purchases.

Additional resources

Sara Brown, Machine learning, explained
Nick McCullum, Deep learning neural networks explained in plain English
Jennifer Monahan, Artificial intelligence, explained

Filed under

Topics

Artificial intelligence