An AI Glossary for Business Leaders

Some companies have created special groups to aid executives in implementing AI initiatives. Farmers Insurance, for instance, runs two AI “councils.” One is focused on identifying business applications and use cases for AI; the other council is IT‑focused and helps identify the right enterprise tools, data sets, and other technical requirements for building AI systems.

AIOps refers to AI used in IT operations. IT Ops pros use AI to analyze large amounts of operational data, then apply machine learning to automate specific tasks such as performance monitoring or managing service requests.

A method for testing whether underlying algorithms in machine learning applications are compromised by human or other biases. Engineers can audit the code itself or feed it a variety of inputs and look for problematic patterns in its decision‑making.

“Black box” generally refers to complex algorithms used in AI applications. They are typically proprietary algorithms and inaccessible to outside review or management.

Also called spiders, scrapers, or crawlers, bots are software applications that automate a set of tasks. Search engines developed some of the first bots to “crawl” websites for content and data that helped improve the quality of search indexing and rankings.

Also called virtual assistants or voice assistants, chatbots are software programs that create conversational experiences with people through text or voice via web browsers, mobile apps, messaging applications and other means. They can answer questions and automate tasks much more efficiently than people.

Also known as voice AI, conversational UI is software, powered by natural language processing, that allows computers (and chatbots, smart speakers, and other devices) to understand and interact with human language.

An advanced form of machine learning through which computers train themselves to perform human‑like tasks, such as understanding speech or making predictions. Instead of using predefined algorithms to draw conclusions from data, deep learning uses multiple processing layers (called neural networks) that allow computers to recognize patterns and learn on their own.

Some AI programs can be trained to identify human emotional states, such as stress. Deep learning algorithms can be trained to recognize a range of human emotions by studying patterns in a person’s facial expressions and speech.

Machine bias occurs when algorithms in AI systems draw erroneous conclusions from data, either due to human error or intent, or as a result of insufficient analysis.

The ability of computers to learn from experience. Machine learning algorithms are typically designed for specific tasks, such as facial recognition or malware detection. The algorithms can be “trained” on large volumes of data.

Also known as computer vision, machine vision allows computers to understand images and other visual content by capturing and analyzing data via cameras or other means.

NLP uses sets of algorithms to allow computers to understand human language as it’s spoken or written. Using deep‑learning models, NLP algorithms analyze unstructured data to interpret ambiguous and complex linguistic structures, including social context, slang, and regional dialects.

The foundational element required in deep learning. Modeled loosely on the human brain, a neural network isn’t an algorithm, but rather a multi‑layered framework that allows machine learning algorithms to work in concert to analyze and learn from complex data sets.

An emerging computer architecture modeled on the biological network of the human brain, designed to overcome the limits of the traditional Von Neumann architecture that defines most computing today. Along with other next‑gen architectures like quantum computing, neuromorphic architectures could power the demanding AI workloads of the future.

Tools that analyze vast quantities of data to predict an outcome based on data patterns and other inputs. These analyses are possible because computers are increasingly good at learning from experience and understanding how certain conditions lead to specific outcomes.

The process of training or teaching workers new skills to adapt a company’s workforce for changing needs and technologies, especially AI. With automation able to perform an increasing number of tasks that were once done by humans, re‑skilling can help workers transition into new jobs and roles.

As part of the European Union’s General Data Protection Regulation (GDPR) that took effect in 2018, companies are required to provide an explanation to customers whenever they are subject to decisions made by machine learning applications, AI, or other advanced technologies. The regulation also mandates aggressive privacy and transparency standards for personal data.

Software applications designed to instruct other applications to automate specific business processes such as clearing transactions, processing data, or communicating with other computer systems.

Computer systems that require little if any human intervention but are designed by humans to advance desired outcomes. These automated systems may draw on elements such as machine learning to improve their performance over time.

An advanced cybersecurity technique, increasingly used in concert with machine‑learning and AI capabilities, that combines automation of rote tasks with the ability to manage security processes and workflows across disparate networks, locations, and platforms.

Algorithms used to test the fidelity of other algorithms. Shadow algorithms can help detect and fix problems that arise when data scientists unwittingly introduce biases and mistaken assumptions into their algorithms.

Data that is not organized according to a specific framework or defined variables that conventional computers can understand. Examples include text, photos, and videos. The ability to process and work with unstructured data, or to transform it into structured data, is a key component of advanced AI systems.