Glossary of Terms

Original: 02/01/20
Revised: 04/03/21

(This is a compilation of technical terms, as used in our articles. Sometimes the terms we use have meanings a bit different from the standard ones. When in doubt, or just for supplementary and somewhat different definitions, reader should consult Wikipedia)

Artificial Consciousness (AC): The consciousness which may be part of the machines (hardware or software) that we will produce in the future. This is the most difficult subject on our website. How and if such an AC may occur takes us into an intersection of many other fields of study: Consciousness. [Wikipedia]
Artificial Intelligence (AI): The intelligence present in the machines (hardware or software) we produce. The AI Components diagram shows the relationship of AI with Machine Learning and Deep Learning, the two most important (currently) subfields of AI. Deep Learning is the focus on our website. [Wikipedia]
Artificial General Intelligence (AGI): The AI functioning at the same level with human intelligence, capable of solving all tasks that humans do. The relationship with the other types of AI, is shown here. [Wikipedia]
Artificial Narrow Intelligence (ANI): Also known as Weak AI, this is AI that is implemented for one narrow task; it can achieve a higher level of intelligence than humans, on that narrow task. The relationship with the other types of AI, is shown here. [Wikipedia]
Artificial Neural Network (ANN): The artificial neural networks are learning systems which do their learning in a manner borrowed from the biological neural networks. They are most responsible for the current success of AI, deep learning is nothing but learning with such networks that are "deep" in the sense of having more than one hidden layer. [Wikipedia]
Artificial Super Intelligence (ASI): The AI which, after having surpassed the level of human intelligence, functions at such high level that it is unimaginable to humans. The relationship with the other types of AI, is shown here. [Wikipedia]
Artificial Wide Intelligence (AWI): This term is not standard, we simply need it in order to anchor and make more precise the AI systems on which we focus on our website. AWI systems work on large graphs, are based in large data centers, and accomplish many ANI tasks. They only have cognitive powers, and while they would understand human emotions and human consciousness, they will not exhibit those properties themselves. The nodes in their graphs are the digital twins of large populations and they tap into the collective intelligence of those nodes. AWI shows a different possible progression from ANI to ASI, without having to go through AGI, as shown here.
Biometric Identifier: A measurable characteristic of an individual human, which is distinctive enough that it can identify this individual. There are two types of such biometrics, physiological and behavioral. They have many uses in AI: identification, access control, or the profiling needed in various digital twins. Among the biometrics we mention in our articles are the brain biometric, the voice biometric, and the facial biometric. [Wikipedia]
Blockchain: A linked list of chronologically ordered records, with each link based on cryptography, and the technology used to produce, verify and distribute these lists. [Wikipedia]
Church-Turing Thesis: The assertion that anything that can be computed, can be computed by Turing machines or one of the other equivalent mathematical formalisms. [Wikipedia]
Computation Theory of the Mind (CTM): The theory that all aspects of the human mind (including consciousness) are some form of computation; it posits that all neuronal activity can be explained as computation; the possibility that the computation may be a quantum computation is included. [Wikipedia]
Convolutional Neural Network (CNN): A special class of artificial neural networks, most effective such networks for image recognition and classification; we mention them in the context of the AlphaGo architecture. [Wikipedia]
Cryptography: The collection of techniques to achieve secure communications between parties; most of our interest is in asymmetric cryptography also known as public-key cryptography. [Wikipedia]
Data Center: We are mostly interested in very large data centers around the world. Such a data center is spread over many buildings, houses thousands of computers, telecommunications equipment, cooling systems, and has special operational requirements, among them 24x7 availability and security. [Wikipedia]
Deep Learning: A subfield of Machine Learning focusing on artificial neural networks having more than one hidden layers. The AI Components diagram shows the relationship of AI with Machine Learning and Deep Learning, the two most talked about subfields of AI. Deep Learning is the focus on our website. [Wikipedia]
Deep Reinforcement Learning: The type of Reinforcement Learning done with the use of deep neural networks. [Wikipedia]
Explainable AI: This is AI whose models can be understood by humans (explained to humans). But most of the spectacular success of AI nowadays has been with "black-box" AI where explainability plays no role and results are left to speak for themselves; that is especially the case with neural networks. There is however a strong movement nowadays away from the "black-box" approach and towards a much more transparent AI, sometimes known as interpretable AI. All three approaches (black-box, explainable, and interpretable) have their adherents. [Wikipedia]
Extended Church-Turing (ECT) Thesis: The ECT thesis tightens the Church-Turing thesis; it says that anything that can be computed efficiently, can be computed efficiently by a Turing machine or the other equivalent mathematical formalisms. Efficiency means polynomial computational complexity, i.e., the time that the computation takes is on the order of a polynomial of the size of its input. [Wikipedia]
Generative Adversarial Network (GAN): A deep learning system composed of two neural networks, a generator and a discriminator, opposed to each other as two players in a game. The generator network generates data with the same characteristics as the training data, while the discriminator tries to figure out if data is generated by the generator. The two networks in AlphaGo, although not designed as a GAN, nevertheless can be looked at as a GAN, as mentioned here. [Wikipedia]
Geometric Deep Learning: The datasets on which AWI systems work are structured as graphs; graphs have a non-flat geometry, as opposed to the structure exhibited by most datasets of current AI applications. This (non-euclidean) geometry of graphs demands a set of new algorithms, algorithms which would take advantage of the structure; Geometric Deep Learning is the development of learning algorithms done on graphs (and other non-euclidean structures); the area is mostly research but evolving rapidly. [geometricdeeplearning.com]
Global Workspace Theory: A theory of consciousness based on the idea of a high-speed memory (a cache in computer speak) which holds the temporary imprint of consciousness. The contents of this memory (called the global workspace) is continuously swapped in and out of larger memory banks in the brain, such as the short-term and the long-term memories. The theory posits that our momentary inner experiences are due to the quick neural connections between this global workspace and the sensory organs, especially vision. [Wikipedia]
Graph Database: These are specialized databases that facilitate the storage and the querying of graph data in a format which is intimately related to the graph structure. They are the fastest growing segment of the DB industry at this time. [Wikipedia]
Integrated Information Theory (IIT): A theory of consciousness which starts with a definition of a concept of consciousness, focuses on the properties of such concept and only afterwards asks what physical substrates exhibit such properties. Although the human brain is the main substrate of interest, it may not be the only one. So the theory may be interestingly tied to metaphysics. The theory has a strong mathematical foundation, which heightens the interest in it.[Wikipedia]
Interpretable AI: This is AI used for high stakes decisions where results have to be justified by humans for humans. Most of the spectacular success of AI nowadays has been with "black-box" AI, especially with deep learning. Then we had explainable AI, whose models can at least be explained by humans to humans. Interpretable AI is an even more transparent form of AI, one whose models can not only be explained but built from the beginning with constraints that ensure that results can be understood by humans. [Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead]
Large Language Model (LLM): This is AI used for high stakes decisions where results have to be justified by humans for humans. Most of the spectacular success of AI nowadays has been with "black-box" AI, especially with deep learning. Then we had explainable AI, whose models can at least be explained by humans to humans. Interpretable AI is an even more transparent form of AI, one whose models can not only be explained but built from the beginning with constraints that ensure that results can be understood by humans. [Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead]
Machine Learning: The subfield of AI studying the statistics-based algorithms which develop models they learn from data. The AI Components diagram shows the relationship of AI with Machine Learning and Deep Learning, the two most talked about subfields of AI. Deep Learning is the focus on our website. [Wikipedia]
Microtubule: Microtubules are hollow tubes (made of proteins) which are part of cellular structure; they provide for transport of material between cells and participate in a variety of other cellular processes, like movement and division. [Wikipedia]
Natural Language Processing (NLP): The field of NLP is a multidisciplinary field, but we are only interested in the AI algorithms that are used for various NLP tasks, like language translation, chat bots, personal assistants, etc. In particular, neural networks and word embeddings based on these neural networks have completely transformed the field and account for most of its current spectacular successes. [Wikipedia]
Neuromorphic Computing: The neuromorphic chips compute in a fashion similar to biological neural networks; they contain a mixture of analog and digital circuitry to mimic that biological functionality. [Wikipedia]
Quantum Computing: A form of computing done with processors and memory based on qubits, not bits, qubits being devices which use using principles from quantum physics, like superposition of states. [Wikipedia]
Quantum Supremacy: The solution of a problem with quantum computing, for which problem there would provably be no possible or practical equivalent solution on a standard computer. [Wikipedia]
Reinforcement Learning: There are three main methods of learning in AI: supervised, unsupervised, and reinforcement learning. Reinforcement learning is built around the concept of an agent which takes actions in order to maximize a reward. It is based on two distinct operations: one of exploration of the effects of possible actions and one of exploitation of the knowledge it has already accumulated. [Wikipedia]
Residual Neural Network (RNN): A special class of artificial neural networks, their architecture contains layers in which the neurons may also connect to other layers ahead, not just the following layer ; we mention them in the context of the AlphaGo architecture. [Wikipedia]
Supervised Learning: There are three main methods of learning in AI: supervised, unsupervised, and reinforcement learning. Supervised learning is used to build models from training data which has been labeled by humans; through this labeling of data, humans supervise the machine. [Wikipedia]
Type 1 Concerns About AI: People have many concerns about the current and future development of AI, and its impact on our civilization. The concerns of Type 1 are immediate and not speculative. Job losses due to automation are an example of Type 1 concerns.
Type 2 Concerns About AI: People have many concerns about the current and future development of AI, and its impact on our civilization. The concerns of Type 2 are about a more distant future and are speculative in nature. The potential destruction of humanity by a malevolent AI is an example of Type 2 concerns; the potential achievement of human immortality is another.
Unsupervised Learning: There are three main methods of learning in AI: supervised, unsupervised, and reinforcement learning. Unsupervised learning is used in algorithms which find patterns in data without previously having seen that type of data labeled by humans in any which way. [Wikipedia]