I apologize for the stridency of my reply, but the statement that ANN or CNN can recognize objects better than 99.98% better than humans is an obvious over-generalization. CNN's have shown some pretty impressive results and I am not disputing that, but the overgeneralization is extremely misleading. Do a search for "cat" in google images, then do a search for "box". Spectacular results! Do a search for "cats jumping out of a box" and the results are disappointing. And yet this just requires a recognition of a combination of visual objects with one simple verb and one simple preposition. And prepositions are the low-hanging fruit of conceptual grammar. Of course the google programmers could react to my criticism and make sure that they train a system to recognize variations of the pseudo-root phrase, "cat jump out of box" but they haven't done so the last time I checked, maybe because they recognize that the shortcomings of the state of the art should be public enough for other programmers to discover. I am not sure how the search engines route different inputs of words, like "cat", or "box" but my guess is that there are no ANNs or CNNs involved in that stage of recognition. It is presumably a very old AI lexical look-up. Visual recognition is special because the imagery appears as a 2-dimension field. And yet the basic capability of animals to recognize 3 dimensional shapes from a pair of 2-D imagery sensors, and the ability of one eyed animals to infer 3-D spaces is beyond current AI. Why? Mostly just because it cannot be done fast enough for developers to swarm over it and for great algorithms to emerge. General conceptual processing requires more than basic recognition of kinds of objects from visual fields based on a pre-selection lexical processing of a search term. I apologize for making such strident statements, because I value the helpful remarks that people are making. The mathematics of conceptual relations cannot all rely on a metric of some sort. So how might a mathematics of discrete terms work? My idea is that it may be possible to create a new kind of mathematics that could be based on a profusion of mathematical computational methods. Then the problem is to make these variants on the computational abstraction rapidly available to be processed without having to be resolved by deep searches and trial and error methods that would have to insure that appropriate computational abstractions were selected for the input. My idea is that this system would be part of a preliminary indexing system that was designed to be flexible enough to work with late-binding or to correct for poor initial assumptions. But to simplify, I am trying to imagine if such a mathematics is possible. Is a mathematics of discrete objects that are not all arranged in a feasible metric-fields possible? Is it possible for a computer program to make rapid computations using a profuse abstraction set without going into deep searches and relevancy dilemmas where to get started the discrete system needs to know in advance which computational methods would be appropriate to use before they were used to understand the input? It may turn out that my efforts are too weak to make a difference, and that I would not even be able to understand the work of anyone who was able to create something similar to what I am thinking about. And maybe CNNs will solve the contemporary problems of AI in the very near future and that my ideas would be brittle even if I could come up with some interesting solutions. But I think that I might only need to understand one fundamental idea that I have yet to see in order to make some progress on this. Jim Bromer
On Tue, Jun 11, 2019 at 12:59 AM <rounce...@hotmail.com> wrote: > Doing things over time isnt mandatory, x and y on your graph could be any > metric under the sun, yes. > > Have you looked at expert systems before? They are a kind of symbolic > logic machine, where you break the problem up into constituent parts, > and presenting it with some information, and a goal - it will find the > "solution", using what u gave it. > *Artificial General Intelligence List <https://agi.topicbox.com/latest>* > / AGI / see discussions <https://agi.topicbox.com/groups/agi> + > participants <https://agi.topicbox.com/groups/agi/members> + delivery > options <https://agi.topicbox.com/groups/agi/subscription> Permalink > <https://agi.topicbox.com/groups/agi/T395236743964cb4b-Me69ee1298125ae41383770e7> > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T395236743964cb4b-M2f69b0a9f830ae4f4fe063d3 Delivery options: https://agi.topicbox.com/groups/agi/subscription