[CG] Look in a dictionary before choosing words to be defined

Alex, Gary, Dan B. Before writing any detailed comments, I want to emphasize three points: (1) Major software systems survive in one form or another for 40 years or more. Few, if any precise definitions from the early days remain unchanged for more than a tiny fraction of that time. As an example, IBM developed the first Airline reservation system for American Airlines in the 1960s to run on the IBM 7094. An updated version of that became IBM's airline reservation system running on System/360. The ontology and terminology of that system became the industry-wide basis for all reservations for hotels, cars, and any kind of services that travelers might need. The ontology and choice of word definitions that IBM adopted in collaboration with American Airlines has become the universal world-wide standard. The formal definitions change with every update, but the choice of words and their translations from English to other languages do not change. (2) The researchers and programmers working on the details of any system may understand the formal details, but the top-level managers, the great majority of the users, and the investors who have money will never see or understand the details of those definitions. They will interpret the terminology according to the way those words are used in everyday life. If the formal definitions diverge too far from common usage, the result will be confusion and repeated errors. (3) Any attempt to edict an official, precise definition for all terms will guarantee that whatever system uses those terms exactly as defined will become obsolete within a few years. Please note that the manuals for every product -- from a refrigerator to a programming language -- will have a new manual with new definitions of key terms for every update. IBM used the term 'functionally stabilized' for any hardware or software system whose terminology would never change. That term was a synonym for "obsolete". IBM would continue to sell those obsolete systems to customers who could not afford to update their systems to accommodate the new products. Microsoft, for example, just recently stopped producing and delivering updates for System/95 (wjocj was introduced in 1995).. Alex> Is there a chance to have one world wide dictionary for every science and technology? You can define it, if you like, but it is guaranteed to become obsolete with the first new discovery in science or new development in engineering. And even if you define it, 99.999% of the people in the world would never use more than a tiny percentage of the words as defined. Alex> AI is first of all summa technologiae, each with its own glossary. There is no universal glossary of AI. New terms are constantly being defined by people who never read or understood similar terms that had been defined and published before. AI terminology changes very rapidly because many AI people never read anything that is more than five years old. Alex> Why is the theory of directed graphs with composition of arrows called category theory? For historical reasons. Mathematicians, unlike Ai people, cite publications of any date and make updates compatible with the original definitions. Alex> Why did the DBMS guys call their company Oracle? Because it answered questions, like an oracle. There are many horror stories about compatibility in DB systems, but they developed in different ways than AI for different reasons -- mostly bad: preserving incompatibility. Preserving incompatibility was also one of the worst reasons for Windows 95. But that is another story. Dan> In general, ML-AI terminology is a mess. Eg Labelled/unlabelled data, unsupervised/supervised learning, giving way (thankfully) to the otherwise wordy “self-supervised”. And the word “inference” is used in ways that might make some ontolog-forum readers splutter their coffee. That's a good answer to Alex's questions. Gary> One may leverage results from prior efforts with best practices but often we don't have the vision or time or temperament to do this. That's a good explanation for the points by Alex and Dan. In summary, most people who need to know something about Ai technology (users and funding agencies, for example) will not know or remember the detail of a formal definition, Even if they read the definition, it will be easier to understand and remember if the words are used in ways that are consistent with common usage -- as codified in common dictionaries. An example of a bad choice is the term 'fundamental model'. Both words are commonly used, but that combination does not give any hint of what the term means. But the terms 'functional pattern' and 'structural pattern' use common words that give an approximate idea of the meaning. That makes them easier to learn, easier to remember and easier to use by everybody -- programmers, managers, funding agencies, and intelligent outsiders who want to know what is happening. John