"OK. But you did express that you thought the distinction (between paper math and computation) isn't meaningful (at least not in perpetuity). Yet you admit that (in perpetuity) we should preserve the distinction at least for the sake of efficiency/performance. You have to admit that can seem paradoxical."
I don't understand why you connect special purpose devices with paper math vs. computation. I claim the problem with paper math is that 1) the former does not carry or enforce correctness checks, 2) it is not put in context -- things are pulled out of thin air as "the reader should know this", and 3) there isn't a formal mapping or harness to a universal computer. So, going back to the "Ask Dad" approach to computing things, one could imagine a detailed model of the sorts of calculations Dad could do very well, but nonetheless leave the actual calculation to him. I could share this model with other people and we could agree it was a "Certified Dad compliant" interface. Regarding 2), ideally a paper's citations and bibliography will provide nodes on the semantic graph to start pulling, but it isn't required or consistently enforced by publishers and it certainly isn't machine readable. The audience of today's technical literature is assumed to be other human domain experts, not, say, a Watson. "Re: _technical_ papers being literate computer programs ... I agree. But a recurring theme in this forum is the poor job journalists do communicating scientific efforts. Analogously, we can predict that when/if all technical papers are literate programs, we'll have a similar problem. This same conversation will continue to occur when the Nicks of the world ask the you-folks of the world what some program means. So the distinction will persist as long as there are general intelligences (Nicks) attempting to parse domain-specific artifacts." If all domain-specific artifacts were built up with machine readable ontologies, then the general intelligent agents will have threads to pull to start putting the artifacts in context. Perhaps some kinds of agents, like humans, would benefit from additional `analogy modules' to assist with mapping large semantic graphs into similar pre-existing ones. That would be an accelerator for learning, not a question of having a sufficient semantic representation. Marcus ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College to unsubscribe http://redfish.com/mailman/listinfo/friam_redfish.com
