From: kymhorsell@gmail.com   
      
   Working with AI's. It can be a bitch. They can see things you can't. They talk   
   down to you like you're a baby.   
   They do things you told them to do but not how you expected. (Good   
   practice for aliens). It should all be banned and I'm glad Elon has written a   
   letter about it. ;)   
      
   Before I get to the movie, one funny story from the past couple of   
   weeks. I was working on a program to model astro-navigation inside the   
   solar system. Some of the data suggests "someone" may be tooling   
   around between the planets using something better than chemical   
   rockets. As we've seen in a prev post, the UFO sightings daily ups and   
   downs seem to match up pretty well with flights from certain planets   
   or moons of those planets assuming trips are pretty much straight line   
   and constant speed.   
      
   The program works out the "most likely" parameters (given the match is   
   statistically significant in the 95% not-just-chance range on both my   
   fave statistical tests) is straight line motion around 1 AU per   
   day. Meaning when certain planets are close to Earth you see a UFO   
   reporting "wave", and when they move further away again the reports   
   tend to decline precipitously. The match is maybe not as good as other   
   models might find, hence I'm trying to handle actual astro navigation   
   including possible high-speed flight plans and maybe multi-planet   
   trips between "home" and here.   
      
   Anyway, as part of the software development I've written a simple   
   gravity simulator that can take the parameters of an orbit --   
   basically a position inside the solar system relative to the sun plus   
   a starting speed in the X Y and Z directions -- and model where an   
   object at that position and speed would end up over time.   
      
   One of the possible "trips" we might be interested in is from planet   
   Mongo to a stable orbit near the earth. So I asked an AI s/w to find   
   what X,Y,Z,VX,VY,VZ parameters would quickly take an object into an   
   orbit near earth from some place far out in the solar system. It   
   easily found a relevant set of coordinates that worked -- it said.   
   But it turned out -- like the infamous djin -- the AI had solved the   
   problem, but not to my liking.  It found a set of coordinates -- a   
   position and a velocity -- that led to a spacecraft quickly moving to   
   a point 1 AU from the sun. I had specified a test to carry out to see   
   if the path actually ended up near the earth. 1 AU from the sun. And   
   all points after that had to also be 1 AU from the sun. A stable   
   "earth type" orbit. I thought I had asked for.   
      
   The AI said "yes, master, I have done exactly what you asked". But   
   when I looked at a plot of the answer it turned out to be a nose-dive   
   into the sun.   
      
   The AI had determined that the last point it could solve for was 1 AU   
   from the sun -- as requested.  And "all points after that were also at   
   1 AU from the sun". Except there were no other points. The simulated   
   rocket was travelling so fast at that point that the next time-step of   
   the track was inside the sun. And because the program the AI was using   
   the do the orbital calculation "cuts out" when a simulated rocket or   
   asteroid either dives into the sun or hits the sun's escape velocity   
   -- meaning it will not be staying around the sun much longer so wave   
   goodbye now -- the AI didn't see any output after the 1 AU   
   point. Hence no point was NOT 1 AU from the sun and all parts of the   
   request were met exactly.   
      
   It's an infamous logical problem -- induction over an empty set.   
   E.g. (in my case) all my older sisters are blondes is true.   
   Because I have no elder sisters and none of them are not blonde QED!   
      
   But to a more interesting update for normal people. ;)   
      
   I've written a couple things to hunt through the now mountains of   
   space telescope output to see what can be seen. Always the AI and   
   stats programs say that the brightness of many stars and even what   
   appears to be dust goes up and down "exactly" in accord with UFO   
   sightings reported some number of hours or days later. They are so   
   insistent and have so much evidence this is true it's hard to refute now.   
   Unless we can find some problem with all the space telescopes that   
   otherwise explains it. (And, yes, they do have some glitches and   
   problems but so far none very well explains the statistics I'm seeing).   
      
   The latest program does a "grid search" across the images from   
   different parts of the sky to find something that is "visible"   
   evidence of something going on. It divides the sky into 4 quadrates,   
   find the "best quadrant" to search by comparing its average brightness   
   with UFO report numbers over following days, then zooms in. And keeps   
   going until the area is small enough to see something "obvious".   
      
   It's all very well for the program to point at some series of images   
   with vague changing shades of grey (usually some tiny points nr some   
   star) and it says these shades of grey get lighter and darker exactly   
   in accord with later UFO report numbers.  But we want to SEE the damn things!   
      
   I originally had posted some movies of lines of stars going a little   
   darker one by one. That might indicate something is crossing in front   
   of them.  But -- of course -- that something might be an asteroid or space   
   junk.   
      
   But the latest movies shows something a little more "widespread". The   
   AIs have found at least one part of the sky where things seem to be   
   going on "all the time".  While I can't be totally sure at this point   
   there is not some stoopid bug in one of the tools that put it all together   
   (including the imaging part as well as the statistics part and the   
   hunting-for-something-interesting section), the movie looks pretty cool.   
      
   It's hunted up a section of some telescope images where the instrument   
   was pointed in the same place continuously for many hours, with images   
   taken every few mins.  Without pushing things to extremes -- and some   
   of the image processing I'm doing magnifies things and image enhances   
   them so they look pretty extreme -- it seems to show "ants" moving   
   around between various stars. But since (in this case) the moving ants   
   predict UFO sightings across N American in a matter of hours later   
   they must be much closer than the stars.   
      
   The latest movie is at .   
      
   This particular set of images supposedly predicts 60% of UFO sightings   
   2 hrs later.  The stats model even manages to predict a local max in   
   reports -- more or less -- saying some hour in the future would see 6   
   UFO reports. The actual number seen in that hour turned out to be 12   
      
   [continued in next message]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)