amsler@mouton.arpa (Robert Amsler at mouton.ARPA) (02/07/86)
I don't know enough about copyright law to answer this, nor do I necessarily believe what we currently have as copyright law will survive another 5 years, but the following possibility grows ever nearer. If someone produces a parser (computer program) which they claim reads in English text and outputs a conceptual data representation of the ideas embodied in that text, and then another AI program takes either that data representation or a knowledge base containing that representation assimilated into its total knowledge and then generates another piece of text which is a paraphrase of the original text---have they violated the copyright of the original author? This assumes the first program read a machine-readable copy of the first text, obtained with permission for its use, but not for its redistribution. As I see it, the dispute would center on whether the conceptual data representation was indeed representative of the ideas in the text rather than the words of the text. Expert testimony from several quarters in the AI community could be summoned to support either side of the issue. Or have I missed some critical aspect of the issue? The answer to this may be critical to the information society in the 1990s when most of the textual information being produced in the world is available electronically and AI software is routinely performing content searches and reading text. A printed book will then become akin to a listing of a program in its computational utility. (The analogy for software law would be that of translation of a program from one language to another by reading the sourcecode).