AI Language
A few years ago I contemplated a language without nouns. I felt that the main difficulty with creating a intuitive interface with a machine was the lack of a mutual language. To that end, I felt that a middle-language would have to be created without nouns.
The idea came in part from the book "Lila" (By the author of "Zen and the Art of Motorcycle Maintenance"); the main character attempts to organize everything in the world into categories-in which the categories kept increasing in complexity and in number.
The difficulty faced by character in Lila is similar to that of attempting to use quantum mechanics to predict weather patterns. Let us say we wanted to plug all the data of weather patterns for as long as there have been adequate records, allow the computer to interconnect the data into a series of cause and effect-and when finished make a weather prediction as how current conditions compare with conditions in the past. The difficulty is that the data will grow exponentially as more and more connections are found and will never be completed enough to make a good prediction. The more data that comes in, the more data that comes out. It never finalizes into a single solution that will output an answer.
Perhaps a better example can be borrowed from Asimovs Foundation and Empire series. All of history is plugged into the tools of psychohistory, every connection between cause and effect is plotted-finally you can make predictions on what new trends in history will occur. Again the same problem, when the data is fed into the program-the program generates more data from that data set rather than less. It will evaluate say all the times a hit to the local economy follows an earthquake and label this even A. It will then evaluate all the times a hit to the economy is followed by political turmoil; this event is labeled B. The system will keep going creating more data. Every time event A is followed by event B will now be labeled event C. The data grows as the computer finds more and more connections- the goal is to minimize data to get a few answers not produce more data. It is similar to a child having a question answered; in turn the child has two more questions to ask. Each answer provokes more questions.
I think there are similarities between the above problems and a real language that a computer can actually understand. Nouns themselves are the variables, the interconnections between them grow exponentially. If you say chair alone, the potential for what ‘chair’ means/implies is near infinite. Without nouns however, the word ‘chair’ would be replaced with exactly what it means being situational and temporally specific.
I had a lot more on the subject written in lost journals. I even attempted to break down the fewest possible grammar techniques I would need. It’s an interesting puzzle however to just contemplate a language without nouns. A noun would be the equivalent of a memory location/variable for a program. Rather than plugging in all the data for each possible word and then trying to train a computer to recognize all possible interconnections-a midpoint language would have to have all the data necessary self contained in the commands/languages themselves.
Hope to write more later on the subject.
