/ commented on getting computers to identify common e.g. strawman fallacies /
Couple of things to remember when you're aiming to build a logical model of meat people:
1. humans are illogical until they learn to be logical ... natural language is not designed to be leak-proof, which is why we don't bother to try and fix natural language, instead we just start working from the other end and invent formal languages which are leak-proof. Opposingly, arising as a direct consequence from the formal language project, computers are logical until they learn how to be illogical. You may literally program random errors into systems to help them pass Turing Tests.
2. Specific to the issue of logical fallacies in natural language ... trying to correct someone's logic, if they are not trying to be logical, may be a waste of time. The reason is that the subsystems motivating most conversations are based on lower inputs, conveniently labeled the four Fs, feeding, fight, flight, and foobar. So achieving logical proofs is the wrong strategy if the system you are trying to hack is invulnerable to logic. To get a hungry person to agree with you, you pay them with food, not proofs.
3. Finally on the naive answer: other people here have already addressed it ... the project of delineating natural language (fundamentally of illogical origin) requires a lot of data ... for complete understanding, your system needs to have a physical and aesthetic (sense data structured) model of each word and sentence it is parsing, and a cultural model of the statistical variance between semantic referents and words (example, system needs to know what balls are, what things look like balls, what things feel like balls, where balls are manufactured, the history of balls, when a ball is just a ball, and when a ball refers to a dance, and when baller means good, and when baller means a person who balls, and when balls just means bullshit, etc.) because this is what a meat person computes in their head when they see "ball". To (user)'s point ... it doesn't need a huge storage network, if you have the right data structure. It fits instead the ball thing on top of our torsos ...
Good luck ... :)
No comments :
Post a Comment