Elsewhere on the internet...

The League of Reason has some social media accounts! You can find us on Facebook or on Twitter for some interesting links and things.

Artificial Intelligence

Post new topic Reply to topic  Page 3 of 3
 [ 47 posts ] 
Artificial Intelligence
Author Message
bemanosUser avatarPosts: 22Joined: Wed Jan 13, 2010 7:23 pmLocation: Greece,Athens Gender: Cake

Post Re: Artificial Intelligence

http://www.youtube.com/watch?v=D21TF1WeNfM
interesting :geek:
Sat Jul 31, 2010 10:31 pm
UrbanMasqueUser avatarPosts: 67Joined: Mon May 24, 2010 7:01 am Gender: Male

Post Re: Artificial Intelligence

bemanos wrote:
http://www.youtube.com/watch?v=D21TF1WeNfM
interesting :geek:


Very Interesting. I want that program he was using towards the end of the video.
Sun Aug 01, 2010 1:31 am
DaealisUser avatarPosts: 235Joined: Sun Feb 22, 2009 4:48 pmLocation: Tampere

Post Re: Artificial Intelligence

I think the evolutionary models are the key. Start with a simple learning model that can expand into applying the learned facts to a given problem.

What I've seen these models do are swarm-problem solving, defining a rudementary language between units and the different types of optimizations in learning to fly, walk or survive in the "simulated battle for the cube". All of the tasks were given as "flap the wings to gain maximal efficiency and uplift", or "walk fast", and the programs themselves figured out the details. This type of learner is definately more approachable than just hard-coding a fully functional consciousness.
"Two things are infinite: the universe and human stupidity; and I'm not sure about the universe."
- Albert Einstein
Thu Sep 23, 2010 1:26 pm
borrofburiModeratorPosts: 3527Joined: Sun Jul 26, 2009 5:27 pm Gender: Male

Post Re: Artificial Intelligence

Daealis wrote:I think the evolutionary models are the key. Start with a simple learning model that can expand into applying the learned facts to a given problem.

What I've seen these models do are swarm-problem solving, defining a rudementary language between units and the different types of optimizations in learning to fly, walk or survive in the "simulated battle for the cube". All of the tasks were given as "flap the wings to gain maximal efficiency and uplift", or "walk fast", and the programs themselves figured out the details. This type of learner is definately more approachable than just hard-coding a fully functional consciousness.

This is actually a machine learning model, not an evolutionary one (well, the learning could be done via evolutionary). The problem with machine learning is it's hard, you have to know what parameters you're optimizing, or the order in which you're optimizing it... If you give a robot dual cameras and just tell it to learn how to get from point A to point B around obstacles, it has to learn on as much as 14 million variables; if instead you teach a robot to translate those 14 million variables into 6 to 12 bins (vertical columns) and THEN teach it to make decisions off those bins, the robot learns MUCH faster.

Machine learning is hard.
Thu Sep 23, 2010 2:05 pm
DaealisUser avatarPosts: 235Joined: Sun Feb 22, 2009 4:48 pmLocation: Tampere

Post Re: Artificial Intelligence

borrofburi wrote:If you give a robot dual cameras and just tell it to learn how to get from point A to point B around obstacles, it has to learn on as much as 14 million variables; if instead you teach a robot to translate those 14 million variables into 6 to 12 bins (vertical columns) and THEN teach it to make decisions off those bins, the robot learns MUCH faster.


I'm aware of this problem that it has, with robots not being able to "guesstimate" but doing it in strict manner that leads to said billion variable "efficiency". But, I still feel that this is a more viable approach than the solid sort of hard-coding the whole thing. The problem being of course that we'd have to give the machine a set of goals, one of which includes a process of optimizing themselves.

One is the said "guesstimate" thing, getting a machine to throw a working solution at first and simply tuning it just "enough", instead of calculating the ultimate value to the point of 500 decimals. I think I've seen an article where they've tried to do this kind of estimating processes. Not sure if it was just an idea never pursued or a real study done.
"Two things are infinite: the universe and human stupidity; and I'm not sure about the universe."
- Albert Einstein
Thu Sep 23, 2010 3:09 pm
borrofburiModeratorPosts: 3527Joined: Sun Jul 26, 2009 5:27 pm Gender: Male

Post Re: Artificial Intelligence

bemanos wrote:
http://www.youtube.com/watch?v=D21TF1WeNfM
interesting :geek:

God damnit, it's kurzweill... again... he doesn't know much about biology or computers... definitely not biology.
Thu Sep 23, 2010 4:36 pm
EnDSchultzPosts: 12Joined: Sat Feb 28, 2009 12:53 amLocation: Washington State, USA

Post Re: Artificial Intelligence

JustBusiness17 wrote:I want my robot to be a Lion :arrow: Then I'll ride it whenever I need to run errands 8-)

[snip]


http://www.youtube.com/watch?v=cNZPRsrwumQ


BigDog all the way, man.
Wed Oct 27, 2010 8:53 pm
Previous
Post new topic Reply to topic  Page 3 of 3
 [ 47 posts ] 
Return to Technology/Engineering

Who is online

Users browsing this forum: No registered users and 4 guests
cron