How to store knowledge in a computer?

Computers are by nature pocket calculators. They have emerged from mechanical calculators and the first working computers were equipped only with 1 kb of ram and small processing speed. The basic function of all the electronic pocket calculator is to add and subtract numbers. The human enters some numbers into the machine and gets the result back on the LED display.

Its not a surprise, that computer science has its root in mathematics. The first questions solved by computers were mathematical problems like finding prime numbers and determine decimal places of Pi. Some more advanced pocket calculator and later computer software is able to multiple matrices and find solution for equations.

Technically a modern 64bit CPU has much in common with early 8bit CPU used to power pocket calculators. Both have an assembly language instruction set and both can add and subtract numbers very fast. The same computer program which was compiled for a 64bit CPU can be compiled for an 8bit CPU, the only difference is that it will run slower.

This introduction about the nature of computing is important to understand the limitation of a computer. Every problem which can't be formulated as mathematical number crunching problem can't be solved with a computer. If the computer should control a robot arm or act as a chatbot for a human, then its difficult to program a computer this way.

What computer scientists can solve very good are mathematical problems. If the task is to add 100 numbers given in the memory or to sort 100 numbers, then a computer will master the challenge easily. Programming in the core sense means to calculate numbers in a highly efficient way.

The technical problem which needs to be adressed is how to describe AI related problems as mathematical problems. It took until the year 2013 before the word2vec algorithm was invented. Word2vec and the simpler one hot encoding variant are mapping words to a matmehatmical vector. This mapping allows a computer to process text. In case of human to machine interaction there is a similar approach available. A numerical communiation code helps to teleoperate a robot. For example the command "move to waypoint B" gets converted into [1="move to",3="waypoint",2="B"] or short [1,3,2]. Such kind of numerical array can be stored and processed by computers, it can also feed into a neural network.