Glossary

Voice recognition refers to programs that analyze human speech via a microphone and sound card before comparing the result with an internal database in order to distinguish individual words. For best results, speech recognition software is trained to recognize the voice of each individual speaker. Unrecognized terms can be added to the user dictionary. Voice recognition programs can be used for the voice control of computers, or for devices or voices used for inputting text into word processing programs.

Unsupervised learning is a method for data analysis in the field of artificial intelligence. In this process, an artificial neural network is guided by similarities within different input values. In unsupervised learning, the computer independently tries to recognize patterns or structures within the input values.

In supervised machine learning, developers provide algorithms with a prepared data set as a training source, with the result already known. The algorithms' task is to recognize the pattern: Why does this information belong in category A and not in category B?

Supervised learning is a technique used for teaching algorithms that are used for the categorization of natural data, for example, photos, handwriting, and speech.

Semi-supervised learning is a hybrid form. In this learning method, only part of the data set is labeled. The rest remains uncategorized with the goal being that these are assigned independently by the algorithms. An example of this can be found in Facebook's face recognition. It is sufficient to label a few pictures with the names of friends before the algorithm finds the remaining faces on its own.

Reinforcement learning is a machine learning option which requires no data for conditioning. The data is generated during training in a trial-and-error process and simultaneously labeled. The program goes through several training runs within a simulation environment to provide an appropriately accurate result. The desired result of this training is that the artificial intelligence is able to solve very complex control problems on its own without prior human knowledge. Compared to conventional engineering, reinforcement learning is faster, more efficient, and, when successful, delivers the best possible of all expected results.

Predictive maintenance refers to a maintenance process based on the evaluation of process and machine data that is found primarily in the linguistic context of Industry 4.0. This real-time processing of underlying data enables forecasts that form the basis of needs-based maintenance and thus the reduction of downtimes. In addition to the interpretation of sensor data, predictive maintenance requires a combination of real-time analysis technology and an in-memory database to achieve a higher access speed to the data compared to hard disk drives. If successful, it will in the future be possible on the production side to deploy a technician to fix a problem before it occurs.

Natural language processing (NLP) is a branch of computer science or, more precisely, a branch of artificial intelligence, that refers to teaching computers how to understand texts and spoken words as humans do.

NLP combines computational linguistics–the rule-based modeling of human language–with statistical modeling, machine learning, and deep learning. Together, these technologies enable computers to process human language in the form of text or speech data and "understand" its full meaning, including the intent and mood of the speaker or author.

Machine learning is a subfield of artificial intelligence which recognizes patterns and regularities as well as the subsequent derivation of suitable solutions. The basis of ML is formed by existing databases, which are needed to recognize patterns. The technology then generates artificial knowledge on the basis of experience already gained; this knowledge can then be generalized and used for solving future problems. Thanks to this approach, even unknown data can be processed and used quickly.

Inferencing is to process knowledge in the form of conclusions. Inferences are made independently of the experience an AI algorithm has accumulated or learned during training, while new knowledge or rules can be obtained through inference from a given database.

A distinction is also made between inductive (forward) and deductive (backward) inferences; two terms frequently used in the field of logic.

Data science is a method for gaining insights from structured and unstructured data. Various approaches are used in data science, from statistical analysis through to machine learning. Most companies use data science to turn data into added value through:

- Increases in revenue

- Cost reductions

- Business agility

- Optimized customer experiences or newly developed products

Data science gives purpose to the data which is collected by an organization.

Data mining is an interdisciplinary activity which uses information and knowledge from the fields of computer science, mathematics, and statistics for the computer-aided analysis of databases. Artificial intelligence methods are used in data mining to examine large data sets for new cross-connections, trends, or patterns. The term “data mining” is often used synonymously with "knowledge discovery in databases", but it is actually only a subarea of knowledge discovery in databases. Data mining automatically extracts correlations and makes them available to higher-level goals. Recognized patterns can help to facilitate decision making for specific problems.

Deep learning is a subcategory of machine learning that models data patterns as complex, multi-layered networks. This type of learning has the potential to solve complex problems for which neither traditional programming nor other machine learning techniques are suitable. It’s also able to create more accurate models than other methods and can be more quickly implemented. Despite these advantages, training deep learning models requires enormous computing power and interpreting the models is a difficult task.

The main feature of deep learning is that the trained models have more than one hidden layer between input and output. In most cases, these are artificial neural networks.

Decision intelligence is an umbrella term used for a wide range of decision-making techniques that bring together several traditional and advanced disciplines in order to design, model, align, execute, monitor, or tune decision-making models and processes. These disciplines include decision management (including advanced non-deterministic techniques such as agent-based systems) and decision support, as well as activities such as descriptive, diagnostic, and predictive analytics.

Taking the form of human-to-human communication or human-software dialogue, conversational commerce is the interaction between customers and companies in real time with the help of technical means such as messaging services (WhatsApp, Facebook Messenger, etc.), digital voice assistants, or chatbots. The aim of this interaction is to simplify and accelerate the customer's contact with a company in order to allow them to find out about certain products or services, use customer service, or buy a product. Through direct customer contact, conversational commerce enables companies to respond to the individual wishes or preferences of their customers, to understand customer needs better, and to win new custom in the long term.

Computer vision is an area of artificial intelligence (AI) that enables computers and systems to extract meaningful information from digital images, videos, and other visual inputs and take actions or make recommendations based on that information. If AI enables computers to think, then computer vision enables them to see, observe, and understand.

Computer vision works similarly to human vision, but humans have had a head start. Human vision has been trained over millennia to distinguish between objects, judge distances, detect movement, and spot irregularities in an image.

Computer vision trains machines in these skills, using cameras, data, and algorithms instead of retinas, optic nerves, and a visual cortex.

Cloud computing is the provision of IT resources via the internet. Users can access cloud computing regardless of their device, and pay for the service according to their consumption. Typical IT resources offered as cloud computing are servers, workstations, storage space, software, databases, and AI algorithms.

A chatbot is a dialogue system with natural language skills of a textual or auditory nature. Chatbots are used on websites or in instant messaging systems, where they explain the products and services of their operators and promote or attend to the concerns of prospects and customers, often in combination with static or animated avatars. In some instances, they simply serve to entertain and reflect. Social bots appear in social media and can function as chatbots. Sometimes the term chatbot is used so broadly that it also includes voice assistants and voice bots.

Big data refers to large amounts of data that is stored, processed, and evaluated with special solutions. Among other things, it is about grid search, (inter)dependency analysis, environment, and trend research as well as system and production control.

The “big data concept” is described using the four Vs:

- Volume: This describes the large amount of data. Ever larger amounts of data must be stored and processed.

- Variety: This refers to the variety of different file structures, such as structured, semi-structured, and unstructured. A large part of stored data is in unstructured formats such as texts, images, or videos. Big data makes this data analyzable through machine learning.

- Velocity: This describes the increased speed by which data must be generated and simultaneously processed.

- Veracity: This refers to the uncertainty (truthfulness) of data and data quality. Data from different sources sometimes does not arrive in the desired quality and therefore cannot be used as intended or must be processed with great effort.

Algorithms are rules for solving a class of problems, with each algorithm consisting of a finite series of steps which enables new output data to be explicitly computed from known input data. An algorithm has five properties:

- Deterministic: Given multiple repetitions of the same input values and the same constraints, the algorithm always produces the same result.

- Correctness: The result of an algorithm must always lead to a correct result.

- Finiteness: Algorithms complete a task in a finite number of steps.

- Uniqueness: The sequence of steps in an algorithm is always the same and always leads to an unambiguous result.

- Generality: An algorithm is designed to solve a class of problems and must be able to solve all situations relevant to that class of problems.

Artificial intelligence (AI) is the generic term for applications in which machines demonstrate human-like intelligence, for example by learning, making judgments, or problem solving. Machine learning technology, a subset of artificial intelligence, teaches computers to learn from data and experience, as well as how to perform tasks better over time. Sophisticated algorithms can recognize patterns in unstructured data sets such as images, texts, or spoken languages, and make decisions independently of them.

This type of learning is made possible thanks to Natural Language Processing (NLP), among other technologies. An example of NLP would be when texts and natural human speech are processed and used in Amazon's Alexa voice service. Deep learning (DL), which uses very deep neural networks with multiple layers and large amounts of data, is currently considered the most promising method of machine learning.

Artificial general intelligence (AGI) is the representation of generalized human cognitive abilities in software which enable an AGI system to find a solution to an unknown task. The goal of an AGI system is to perform any task that a human is capable of.

Definitions of AGI vary because of the array of expert perspectives defining human intelligence. Computer scientists often define human intelligence as the ability to achieve goals, whereas psychologists typically define general intelligence in terms of adaptability or survivability.

AGI is referred to as strong artificial intelligence (AI). Strong AI contrasts with weak or limited AI, which is the application of artificial intelligence to complete specific tasks or solve identified problems.