 |
||
 |
 |
 |
 |
 |
|
| Your failure to even attempt being informed is making me hemorrhage. by Senor Barborito 05/22/2003, 8:23pm PDT | ||||
 |
 |
|||
Entropy Stew wrote: How human intelligence functions isn't something we need to know or care about for this project. We just need to be able to filter out results humans will find aesthetically displeasing and to enforce the themes of the results they find maximally pleasing across the locale. Funny how I could end up confused, what with GAs having nothing to do with Cog or even AI and you using them as an example. GA isn't artificial intelligence, but it's from the same root principle as the bottom-up camp of AI - take dumb components (a single neural net) and combine them in interesting ways so that one might teach or arrive at intelligence to/within the superstructure. You know what's hilarious, though? I fucked up the quotes and responded to your previous post's argument here in this subthread, and I was able to literally cut and paste your current argument over it and not have to change my statement at all because you said the exact same thing twice. You don't get compelling design out of seeding a fractal world generator, dumbass. Even if this approach actually did result in a compelling world (through magic!), it would still be better accomplished by a GA. Current neural nets would just keep giving you back something that looks like the optimal solution. Fuck, your whole approach to neural nets is just ignorant. A neural net is just some hyperplane geometry (I think) running on a bunch of neurons. Neural nets have the same chance of creating intelligent behavior in their raw form as trinary math running on atoms. You keep trying to shoehorn worlduilding into the same difficulty level as detecting spam. "Use a Bayesian network! It'll learn - that's what they do!" 
When I went to school in 1998 neural networks were way, way beyond that point . There was a U of AZ (IIRC, this was almost 5 years ago) researcher back then who was working on a ton of stuff related to meta-data influencing the network, supervisor networks running layered over the network (the specific phrase 'seven-tiered neural network' sticks out), and adaptive networks that could grow whole new subnetworks to fit changing data, reorder where inputs for nodes went and where outputs came from (provided all data inputs were used) as necessary. Your information was out of date half a decade ago, dumbass, and this is just the bits I picked up/remembered from attending the project meetings for the AI side of the project - I was not and am not into AI all that much. Kindly spare me your fucking University of Google education on the topic. The worst part is you're not even trying to take what I say and see the obvious solution - you're attempting to shoot me down by any means you can so that you can appear more 'correct,' which is the kind of behavior I'd expect from a mealy-mouthed shit-eater like INC, not you. The obvious solution here is to take randomly generated values, use them to seed your topology/heuristics mappings 'strong AI' component, then use your 'weak AI' (neural network) to, yes, spam-filter the aesthetics and then propagate/recombinate the better points within the overall architectural theme of the town with the worse ones (or completely overwrite them). "AHA!" you shout, thinking you've caught me making a grievous error with the spam reference which you tried to use as a point against me, "Neural networks can't filter spam!" That's right, genius, they can't - because the context of the modern world requires far more specific knowledge and heuristics than can ever be reasonably encoded into a project like Cyc or a neural network (without tens of trillions of nodes sporting hundreds of interconnections like what the human brain has). You're quite right about it being AI-strong, and spam being so as well. What you're wrong about is medieval architecture (or NPCs for Turing Test purposes, but that's another fight) being within several orders of magnitude as complicated as either. I'm not even going to bother if you keep changing your goal mid-thread. Your entire argument seems to be based on trivilization of every aspect involved. The 100% originality you wanted is different from 'original' as I was using it in the quoted paragraph and you goddamned well know it. Don't play semantics games with me. --SB |
||||
 |
 |
|||
|
