tag:blogger.com,1999:blog-6923947282926324208.post7402055550593914105..comments2018-01-12T05:05:20.999+01:00Comments on Entropic and Fractal Intelligence: Pathfinding problemSergio Hernandezhttps://plus.google.com/107797149522609875320noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-6923947282926324208.post-5809924288072078502016-07-27T17:04:45.348+02:002016-07-27T17:04:45.348+02:00Hi Pierre and thanks for your interest.
This post...Hi Pierre and thanks for your interest.<br /><br />This post is exactly about solving this path finding problem using "causal entropic forces", but Alexander's original formulas for entropic forces are not enough, as they only integrate on possible paths using the probavility of this path being taken by the system.<br /><br />It means the resulting behaviour of this AI is going to be focused on keeping as many different futures available to the system, ie. keeping the system up and running as long and steady as possible, but you can not define any other goal for the AI to follow, as in this case should be "try to be as near as the target position as you can".<br /><br />My implementation differs from Alezander's in that, I added a potential associated with a system's state, so I could redefine entropy to account for this potential and make the AI to follow my "orders".<br /><br />In this case, for any state S, I used:<br /><br />Potential(s) = 1/(1+Dist_From_Target_To(s))<br /><br />This is a positive funtion that is higher as you reach your goal, and the addition to it into the entropy formulas is equivalent to asking the AI to maximize this potential value over time, ie. find a way to reach to the target point.<br /><br />My first way to add this potential was using "feelings" that are higher as you get near of your goals, so the potential was something like a "emotional potential".<br /><br />You can read about it here: http://entropicai.blogspot.com.es/2014/10/introducing-emotional-intelligence.html<br /><br />As you can see, the potential definition is far from trivial, and the intelligence generated is limited in some aspects, so finally I changed from using entropy to usign fractals. Some how the fractal version decides with option has higer entropy without actually calculating it. This sohwed to be far more powerful that the previous version.<br /><br />Fractal version is not 100% finished, but in some moment next year I plan to release the basic form of fractal AI as a python pakage.<br /><br />Before that you can check the basic form of it by reading this post:<br /><br />http://entropicai.blogspot.com.es/2015/09/fractal-algorithm-basics.html<br /><br />So there is not a published version of the path finding problem using fractals, but the pre-fractal version of the algorithm is also capable of it (once you define a new potential or goal for the task) and you have code of it in the download link (pascal code written in delphi, no phyton version of it will be added as this a death branch of the developement).<br /><br />I hope I cleared most of your questions, feel free to re-ask, some parts of the algortihm are still considered "confidentiall" but the bulk is not.<br /><br />Sergio Hernandezhttps://www.blogger.com/profile/18108694861191833007noreply@blogger.comtag:blogger.com,1999:blog-6923947282926324208.post-29955297276072358522016-07-26T19:41:00.410+02:002016-07-26T19:41:00.410+02:00Hello, I am very interested in your research and e...Hello, I am very interested in your research and experiments. Have you tried to use the causal entropic force formalized by AD Wissner-Gross to find paths? I would like to see the algorithm if it is possible. Thank you very much. PierrePierre Rochahttps://www.blogger.com/profile/03785481568411511276noreply@blogger.com