Ai Dreams Forum

Artificial Intelligence => General AI Discussion => Topic started by: LOCKSUIT on May 24, 2019, 02:11:50 am

Title: Search-Space Optimization
Post by: LOCKSUIT on May 24, 2019, 02:11:50 am
Search-Space Optimization

Say you have AGI, and you want it to recognize there is a cat in the 'image' that it takes a look at.
How do you know which output to output when you have such a large search-space?

Say you have AGI, and you want it to achieve goals in real life.
How do you know which output to output when you have such a large search-space?

Say you have AGI, and you want it to segment the image correctly.
How do you know which output to output when you have such a large search-space?
Title: Re: Search-Space Optimization
Post by: AndyGoode on May 24, 2019, 09:35:53 pm
What do you mean by "output to output"? An image is usually considered the input to a recognition system, and the recognition or identification is considered the output.
Title: Re: Search-Space Optimization
Post by: LOCKSUIT on May 24, 2019, 09:43:22 pm
An output is like an action, it may say "I SEE A CAT". So outputting this is returning it to the user, i.e. "to output" an output, is literally to give a gift to the user. To return an answer. To output an output.

I feel SSO is interesting because we want an understanding of what/why/how AI works, and a way to implement it on our computers.

I have good answers to the OP but I want to hear others inputs on my output :)

My questions are litterally asking you how does the AI find the answer when there is 99999999999 zillion possible answers!?
Title: Re: Search-Space Optimization
Post by: HS on May 24, 2019, 09:48:16 pm
It's got to know enough about cats to predict their potential forms. When the AI sees something unknown it starts fiddling around with it's mental model of a cat to see if it can make it match to what it sees.
Title: Re: Search-Space Optimization
Post by: LOCKSUIT on May 24, 2019, 09:49:46 pm
Also, as you may have noticed in my last post - AGI=heuristics, because it must eliminate many unlikely answers - exactly a way for us to implement it on our current computers !
Title: Re: Search-Space Optimization
Post by: LOCKSUIT on May 24, 2019, 09:50:40 pm
That's a start HS! Yes it has to be fed a lot of data, and start clustering it.
Title: Re: Search-Space Optimization
Post by: LOCKSUIT on May 24, 2019, 10:03:53 pm
You need hints, where to look in your memory. Votes to weigh in.
Title: Re: Search-Space Optimization
Post by: HS on May 24, 2019, 10:06:31 pm
Oh! If the AI has a body with joints or already knows about some other animals with similar anatomy to cats, then it can still predict the cat shapes without needing as much info specifically on the cat. If it encounters an unknown creature it can use broad knowledge to "triangulate" the creatures potential properties!
Title: Re: Search-Space Optimization
Post by: LOCKSUIT on May 24, 2019, 10:08:34 pm
quantinization!
Title: Re: Search-Space Optimization
Post by: HS on May 24, 2019, 10:08:44 pm
I think you only need a moderate amount of good quality knowledge/memory to deduce most things.
Title: Re: Search-Space Optimization
Post by: HS on May 24, 2019, 10:10:37 pm
Like mixing paints! Primary colors -----> Primary knowledge!  Hhehehehe I be genious!
Title: Re: Search-Space Optimization
Post by: HS on May 24, 2019, 10:21:01 pm
You'd think this primary knowledge would be physics, because you can deduce everything with it. But thats too processor intensive if you attempt to predict behaviors of anything more complex than a virus. This would be so much easier if the AI had goals. Then it could self select useful knowledge.
Title: Re: Search-Space Optimization
Post by: goaty on May 25, 2019, 04:29:47 pm
Think of it like chess.

If you test every single possibility, one must result in the kings death, the computer with enough search power is a perfect player, amazing and true. 
But theres an exponential amount of positions to check,  so the search space is too large.

But its actually not that bad...  if you think about it,  a lot of the board states will be invalid moves,  only 1 piece is allowed to move at a time,  and if you can look less than only 10 moves in front, its already better than the greatest human player just about.

Budgetting it,  is a bit like budgeting the lotto,  you have to work out the odds, of ending up at every state.

Binary probability trees are a good thing to study.

The thing that's exciting about this,  is if you had a quantum computer,  you would have no limits to the amount of states you need to check.  even a google states it could handle...  and could be a secret of how to get AGI to happen.

But if there is an optimization here,  its a tricky and top secret one...    like P=NP.
Title: Re: Search-Space Optimization
Post by: LOCKSUIT on May 25, 2019, 10:44:08 pm
Remember i said talk to 100 girls a day and 1 will hook with you? Everything works in #s. Teams. Spears. Etc.

If you send out 55 look scouting ants in your chess search tree, one will find a better path! Skip search space!? Just an idea, may not even work. Seems right though.
Title: Re: Search-Space Optimization
Post by: HS on May 25, 2019, 11:05:09 pm
Sounds like a job for parallel computing.
Title: Re: Search-Space Optimization
Post by: AndyGoode on May 26, 2019, 03:48:21 am
I think you only need a moderate amount of good quality knowledge/memory to deduce most things.

I agree with this, and so does Jeff Hawkins. Most of what the cortex does is pattern matching (especially associative) and recall, not huge amounts of parallel processing that tackle each new problem from scratch. I keep quoting this book as answers to so many questions on this forum that it sounds to me that people here should read it as good foundation knowledge:

(p. 67)
   So how can a brain perform difficult tasks in one hundred
steps that the largest parallel computer imaginable can't solve in
(p. 68)
a million or a billion steps? The answer is the brain doesn't
"compute" the answers to problems; it retrieves the answers
from memory. In essence, the answers were stored in memory a
long time ago. It only takes a few steps to retrieve something
from memory. Slow neurons are not only fast enough to do this,
but they constitute the memory themselves. The entire cortex is
a memory system. It isn't a computer at all.

(p. 68)
Let me show, through an example, the difference between com-
puting a solution to a problem and using memory to solve the
same problem. Consider the task of catching a ball. Someone
throws a ball to you, you see it traveling toward you, and in less
than a second you snatch it out of the air. This doesn't seem too
difficult--until you try to program a robot arm to do the same.
As many a graduate student has found out the hard way, it seems
nearly impossible. When engineers or computer scientists
tackle this problem, they first try to calculate the flight of the ball
to determine where it will be when it reaches the arm. This cal-
culation requires solving a set of equations of the type you learn
in high school physics. Next, all the joints of a robotic arm have
to be adjusted in concert to move the hand into the proper posi-
tion. This involves solving another set of mathematical equations
more difficult than the first. Finally, this whole operation has to
be repeated multiple times, for as the ball approaches, the robot
gets better information about the ball's location and trajectory. If
the robot waits to start moving until it knows exactly where the
ball will arrive it will be too late to catch it. It has to start moving
to catch the ball when it has only a poor sense of location and
it continually adjusts as the ball gets closer. A computer requires
millions of steps to solve the numerous mathematical equations
to catch the ball. And although a computer might be programmed
to successfully solve this problem, the one-hundred step rule
tells us that a brain solves it in a different way. It uses memory.
(p. 69)
   How do you catch the ball using memory? Your brain has a
stored memory of the muscle commands required to catch a ball
(along with many other learned behaviors). When a ball is
thrown, three things happen. First, the appropriate memory
automatically recalled by the sight of the ball. Second, the mem-
ory actually recalls a temporal sequence of muscle commands.
And third, the retrieved memory is adjusted as it is recalled to
accommodate the particulars of the moment, such as the ball's
actual path and the position of your body. The memory of how
to catch a ball was not programmed into your brain; it was
learned over years of repetitive practice, and it is stored, not cal-
culated, in your neurons.
   You might be thinking, "Wait a minute. Each catch is
slightly different. You just said the recalled memory gets contin-
ually adjusted to accommodate the variations of where the ball
is on any particular throw . . . Doesn't that require solving the
same equations we were trying to avoid?" It may seem so, but
nature solved the problem of variation in a different and very
clever way. As we'll see later in this chapter, the cortex creates
what are called invariant representations, which handle varia-
tions in the world automatically. A helpful analogy might be to
imagine what happens when you sit down on a water bed: the
pillows and any other people on the bed are all spontaneously
pushed into a new configuration. The bed doesn't compute
how high each object should be elevated; the physical proper-
ties of the water and the mattress's plastic skin take care of the
adjustment automatically. As we'll see in the next chapter, the
design of the six-layered cortex does something similar, loosely
speaking, with the information that flows through it.

Hawkins, Jeff. 2004. On Intelligence. New York: Times Books.
Title: Re: Search-Space Optimization
Post by: LOCKSUIT on May 26, 2019, 04:32:00 am
I always said RNNs would fade out. Recurrence is not right. Yes its sequence but it doesn't loop like that. Transformers were born. Highly parallel. You can watch the slide below if you don't believe me:
https://www.slideshare.net/xavigiro/attention-is-all-you-need-upc-reading-group-2018-by-santi-pascual
Title: Re: Search-Space Optimization
Post by: goaty on May 26, 2019, 04:43:51 am
Saying that a memory system is all you need is good until you have to do something for the first time,  then uve got nothing to remember. :)
Title: Re: Search-Space Optimization
Post by: HS on May 26, 2019, 05:28:28 am
But a few seconds later you do!  ;D
Title: Re: Search-Space Optimization
Post by: LOCKSUIT on May 26, 2019, 05:37:30 am
Don't worry, don't worry, as soon as I understand GPT-2, I'm gonna combine it with my AGI brain design, and it will be able to recall etc.
Title: Re: Search-Space Optimization
Post by: goaty on May 26, 2019, 05:44:03 am
But a few seconds later you do!  ;D

Like after its walked off a cliff? :)

.
Title: Re: Search-Space Optimization
Post by: HS on May 26, 2019, 05:54:08 am
If its a very tall cliff lol.