|
Gary Gaulin
Guest
|
 |
« on: January 03, 2010, 07:53:12 AM » |
|
I could not resist explaining something I know about that would make the ultimate collaboration project for a fractal forum. Just in case anyone else here is up to this challenge. I have known about fractals and their basics but am relatively new to their details. From what I read here and elsewhere I appear to be working with one but never knew what to call it. Which brought me here looking for more information on them. It still appears to be a fractal, but other opinions on that are welcome. My primary occupation is in the graphic arts (printing) and to some degree science. With the economic crisis having devastated local industry and there being occasional money in software I write I went back to work on one of the projects that I have been experimenting with over the years that figures out how to "crosscut" paper to get the most cut pieces out of a given sheet possible.  After reading Paul N. Lee's excellent definition of fractal I better add that since it's only necessary to show the layouts with the greatest number out most of the fractal is never shown. It would be possible to have infinite layouts inside of layouts (and in the trims) that keep the same proportions of full to cut dimensions, but a printer just wants to know how to cut their stock down to size (not make fractal art) so that recursion is not included. The algorithm is now decribed here: http://crosscutfractal.blogspot.com/ If it is possible to conclude that it is not a fractal then I can zap the blog. It's not necessary for it to be one, it's just that a fractal is the only thing in math/science that I have ever found to explain it. And it very much relates to other recursive related phenomena. One of the most interesting comes from 1970's robotics, the work of David Heiserman who wrote books on building relatively simple electronically hardwired self-learning systems that model the behavior of biological systems that makes it to the protozoan or simple insect level. Since it's a lot cheaper and easier to model on a personal computer (that did not exist back then) than build a board for a robot to crash around the house I studied/simplified the algorithm in cyberspace. A tutorial on how it works with VB6 source code is the link below. http://intelligencegenerator.blogspot.com/ Self-similarity should be observable in a computer model of a cell where each of its molecules is its own intelligence loop learning to behave like its biochemical counterpart. The cell's behavior, something that fractaled out of that. Where you have many aggregating cells in the model able to change in form to adapt to survive you have stem cells that differentiate into a complex multicellular organism. Here's the loop from the above blog that through fractal recursion (and with computing power we only wish we had) could theoretically form societies of virtual plants and creatures: LoopStart:
1: Call RunMotors
2: If Stall=0 then Conf(Addr)=Conf(Addr)+1 Else Conf(Addr)=Conf(Addr)-1
3: Addr = LMF + (LMR*2) + (RMF*4) + (RMR*8) + (Stall*16)
4: If Conf(Addr)<1 Then Call RandomGuess: Conf(Addr)=1 It is important to notice it is "confidence driven" (like we are) and only needs to know how successful it was. And you are not modeling the exact physical shape of things, just their behavior which in turn can define form. Would here have to predict that a complex organism at our scale that could fractal out of these virtual atoms or molecules would also be confidence driven, without having to design that into their brain (which would likewise fractal out of the molecular behavior without us having to first design its brain ourselves). The "circuit" that it is simulating has two way communication like our brain to our motor muscles that have neuron to turn it on/off and another for feedback to see how well it's doing in response, which would predict two way muscle control will emerge on its own by it already being present in the behavior of the fractal it came from. A massively parallel computer to fully test the idea does not yet exist, but PC's are still powerful enough to get started. And instead of starting at molecules, the loop could model one cell body each but require software that simulates an even more complex system. The least complex would be where the emergent behavior of matter (hence living things) begins, that scientists would love to discover more about with the new CERN superconducting supercollider. Anyway, if you're looking for a challenging project then that is easily able to keep you busy for a lifetime too. Should be possible to go from what I have at this point to a new fractal theory to explain how to model the behavior of natural systems from quarks on up to stars. A theory of everything, that explains a fractal that fractals into everything. Or at least demonstrates how to get a part of the way there in a PC... |
|
|
|
« Last Edit: March 07, 2010, 01:54:46 PM by Nahee_Enterprises »
|
Logged
|
|
|
|
|
Nahee_Enterprises
|
|
« Reply #1 on: January 05, 2010, 11:03:18 AM » |
|
I could not resist explaining something I know about....
My primary occupation is in the graphic arts (printing)....
....I have been experimenting with over the years that figures out how to
"crosscut" paper to get the most cut pieces out of a given sheet possible. An interesting project, two interesting blogs, and some interesting ideas. I only had a few moments to glance over all of it, including the blogs, so will have to come back when I can spend a bit more time. Just wanted to say Greetings, and Welcome you to this particular Forum !!!!  Very kind of you to say so, but the first paragraph of that definition is mainly from a dictionary I was using way back when. The rest of it is my own writing though. |
|
|
|
« Last Edit: March 07, 2010, 01:56:54 PM by Nahee_Enterprises »
|
Logged
|
|
|
|
|
Gary Gaulin
Guest
|
|
« Reply #2 on: January 06, 2010, 03:59:12 PM » |
|
Hi Paul, thanks for responding! It was this later paragraph that was most useful. Wikipedia and other sources will cover it but it's hard to tell whether that is what people who experiment with fractals are normally working with, and the accuracy of information. The way you explained it here was all I needed to begin to make sense of all the formulas that I have been seeing: The standard Mandelbrot fractal equation takes the form z(n+1) = z(n)^2 + c, where c is the complex number x+iy corresponding to any point on the (x,y) coordinate plane. Fractal equations are iterative, in that the result of one calculation of the fractal equation becomes the z input to the next calculation. Over repeated evaluations of a fractal equation, values for each point in the (x,y) coordinate space either converge at single points, move toward the (0,0) origin point, or move toward infinity. The diverse colors in fractal plots reflect the rate of this movement for each point. Discussions of chaos theory frequently use fractals as examples, because slight variations in the fractal equation produce radically different results. I later found "Introduction to the Mandelbrot Set, A guide for people with little math experience. By David Dewey" which helped make sense of the imaginary numbers and complex number plane. I also looked at how fractal trees were calculated. But the method is so entirely different and in a way such a comparatively simple draw that along with what I read on Wiki I have to (with some humor) wonder whether plotting the points of a circle might also somehow qualify as drawing a fractal. In an intelligence algorithm, iteration is used to keep it going from frame to frame (time). The behavior of the pixels/entities produce the final image (as opposed to behavior of an equation that places the pixels) like in this experiment:  http://selflearningbots.blogspot.com/ In this case, to draw a Mandelbrot fractal the intelligence generator algorithm would use the Mandelbrot equation to determine whether the last step each entity/pixel took was successful in getting closer to forming one, or it failed. If successful then confidence is incremented, else decremented. In time they would learn to behave like the equation. Of course a "fractal of everything" should like in matter produce competing entities that increase in complexity to become competing entities at the next highest level, then the next. The iteration would be the same as you mentioned "Fractal equations are iterative, in that the result of one calculation of the fractal equation becomes the z input to the next calculation." but since an intelligence generator algorithm places/moves the particles it does not need an equation to form an image which at least simplifies the "complex number" part of the equation for those who have a hard time imagining an imaginary number!  |
|
|
|
|
Logged
|
|
|
|
|
kram1032
|
|
« Reply #3 on: January 06, 2010, 07:07:07 PM » |
|
nice stuff I wonder what OpenCL could do here |
|
|
|
|
Logged
|
|
|
|
|
Gary Gaulin
Guest
|
|
« Reply #4 on: January 07, 2010, 03:46:40 PM » |
|
nice stuff I wonder what OpenCL could do here Thanks for the compliment! And you must jest about OpenCL. High performance multi-core parallel computing with OpenGL 3D would be the ultimate! OpenCL (Open Computing Language) is a framework for writing programs that execute across heterogeneous platforms consisting of CPUs, GPUs, and other processors. OpenCL includes a language (based on C99) for writing kernels (functions that execute on OpenCL devices), plus APIs that are used to define and then control the platforms. OpenCL provides parallel computing using task-based and data-based parallelism.
OpenCL is analogous to the open industry standards OpenGL and OpenAL, for 3D graphics and computer audio, respectively. OpenCL extends the power of the GPU beyond graphics (GPGPU). OpenCL is managed by the non-profit technology consortium Khronos Group. http://en.wikipedia.org/wiki/OpenCLDo you program in OpenCL? If yes, then give the 3D "Particles In A Box Bots" idea a try!  http://selflearningbots.blogspot.com/I would love to see that recoded into a language worth migrating to from VB6 which still compiles into fast loops and all but the graphics slows it way down. It is far from optimized by setting single bits with "+ (YMR(N) * 4) +" and such that could be performed with an "Or" and use "And" also but I try to avoid making it harder for others to figure out the code. The main loop that does the thinking is the last subroutine. Private Sub WriteMainMemory(N As Long)
MainMem(Addr(N)) = XMR(N) + (XMF(N) * 2) + (YMR(N) * 4) + (YMF(N) * 8) + (ZMR(N) * 16) + (ZMF(N) * 32) + (Cnf(N) * 64)
End Sub
Private Sub ReadMainMemory(N As Long)
Dim Estr As String
Dim MM As Long
Dim M2 As Long
XM1 = XMR(N) + (XMF(N) * 2)
YM1 = YMR(N) + (YMF(N) * 2)
ZM1 = ZMR(N) + (ZMF(N) * 2)
If ShowMemCheck = 1 Then
If N < 16 Then
Estr = Format(XMR(N), "0") & Format(XMF(N), "0") & "-" & Format(YMR(N), "0") & Format(YMF(N), "0") & "-" & Format(ZMR(N), "0") & Format(ZMF(N), "0") & "-" & Format(At1, "000") & "-" & Format(At2, "000") & "-" & Greater2
End If
End If
MM = MainMem(Addr(N))
M2 = Fix(MM / 2)
XMR(N) = MM - (M2 * 2)
MM = M2
M2 = Fix(M2 / 2)
XMF(N) = MM - (M2 * 2)
MM = M2
M2 = Fix(M2 / 2)
YMR(N) = MM - (M2 * 2)
MM = M2
M2 = Fix(M2 / 2)
YMF(N) = MM - (M2 * 2)
MM = M2
M2 = Fix(M2 / 2)
ZMR(N) = MM - (M2 * 2)
MM = M2
M2 = Fix(M2 / 2)
ZMF(N) = MM - (M2 * 2)
MM = M2
M2 = Fix(M2 / 2)
Cnf(N) = MM - (M2 * 2)
MM = M2
M2 = Fix(M2 / 2)
Cnf(N) = ((MM - (M2 * 2)) * 2) + Cnf(N)
AvConf = AvConf + Cnf(N)
If ShowMemCheck = 1 Then
If N < 16 Then
BotList.Print Format(N, "00000") & " " & Estr & " " & Format(XMR(N), "0") & Format(XMF(N), "0") & "-" & Format(YMR(N), "0") & Format(YMF(N), "0") & "-" & Format(ZMR(N), "0") & Format(ZMF(N), "0") & "-" & Format(Cnf(N), "0")
End If
End If
XM2 = XMR(N) + (XMF(N) * 2)
YM2 = YMR(N) + (YMF(N) * 2)
ZM2 = ZMR(N) + (ZMF(N) * 2)
End Sub
Private Sub RandomAction(N As Long)
XMR(N) = Fix(Rnd * 2)
XMF(N) = Fix(Rnd * 2)
YMR(N) = Fix(Rnd * 2)
YMF(N) = Fix(Rnd * 2)
ZMR(N) = Fix(Rnd * 2)
ZMF(N) = Fix(Rnd * 2)
End Sub
Private Sub RunMotors(N As Long)
'Apply the motor direction bits to current bot location for a new position in 3D World.
Xm = Xbot(N) + YMF(N) - YMR(N)
Ym = Ybot(N) + XMF(N) - XMR(N)
Zm = Zbot(N) + ZMF(N) - ZMR(N)
'Check to see if it is trying to pass through the Wall.
Bump(N) = 0
Wall(N) = 0
If Xm < 0 Then Wall(N) = 1: Xm = 0
If Ym < 0 Then Wall(N) = 1: Ym = 0
If Zm < 0 Then Wall(N) = 1: Zm = 0
Sz = WorldSize
If Xm > Sz Then Wall(N) = 1: Xm = Sz: Xbot(N) = Sz
If Ym > Sz Then Wall(N) = 1: Ym = Sz: Ybot(N) = Sz
If Zm > Sz Then Wall(N) = 1: Zm = Sz: Zbot(N) = Sz
'Look for Bump into stationary bot it cannot occupy space of.
Colr = BotColor
If Xm = Xbot(N) And Ym = Ybot(N) And Zm = Zbot(N) Then
Else
If Bot3D(Xm + 3, Ym + 3, Zm + 3) > 0 Then
Colr = BumpColor
Bump(N) = 1
Xm = Xbot(N)
Ym = Ybot(N)
Zm = Zbot(N)
Exit Sub
End If
End If
'Exit the subroutine with bot unmoved if Wall was detected.
If Wall(N) = 1 Then
Colr = HitWallColor
Xm = Xbot(N)
Ym = Ybot(N)
Zm = Zbot(N)
Exit Sub
End If
'Making it here, means it can move there, so physically change its X,Y location in 3D World.
Bot3D(Xbot(N) + 3, Ybot(N) + 3, Zbot(N) + 3) = 0 'Clear Bots last position in World array.
Bot3D(Xm + 3, Ym + 3, Zm + 3) = 1 'Save Bots new postion, by setting to 1.
Xbot(N) = Xm 'Update the X,Y location in bot array.
Ybot(N) = Ym
Zbot(N) = Zm
End Sub
Private Sub FormMemoryAddress(N As Long)
Addr(N) = XMR(N) + (XMF(N) * 2) + (YMR(N) * 4) + (YMF(N) * 8) + (ZMR(N) * 16) + (ZMF(N) * 32) + (Wall(N) * 64) + (At1 * 128) + (At2 * 32768) + (Greater2 * 8388608)
End Sub
'>>>>>>>>>>>>>>>>>>>>>>>> Main Intelligence Generation Loop <<<<<<<<<<<<<<<<<<<<<<<
Private Sub MainLoop()
Dim N As Long
Dim AngNum As Long
Dim Angle As Long
'This subroutine keeps jumping back up to here until EndMainLoop=True
StartMainLoop:
Moves = Moves + 1 'Count how many times bots were Moved for screen.
'Draw the header where memory each bot is using is shown on screen.
If ShowMemCheck = 1 Then
BotList.Cls
BotList.Print " ADDRESS DATA"
BotList.Print "---------------------------------------"
BotList.Print "Num. Xm Ym Zm 1 2 3+ Xm Ym Zm Cf"
BotList.Print "---------------------------------------"
End If
'Clear variables to track particle progress shown on screen.
AvConf = 0
AvNeighbors = 0
AvBumps = 0
AvWallHits = 0
'For each of the bots in the simulation (starting with number 1)
For N = 1 To Bots
'Get the Position (and Distance not used) of other bots around it.
X = Xbot(N) + 3
Y = Ybot(N) + 3
Z = Zbot(N) + 3
Call LookAround
Neighbors2 = BotNeighbors(N)
BotNeighbors(N) = Neighbors
'If not Training then the motor Data environment addresses, is the output Action.
If TrainCheck = 0 Then
Call FormMemoryAddress(N)
Call ReadMainMemory(N)
Call RunMotors(N)
GoTo BotIsDone 'Skip the rest of the loop, right back to StartMainLoop above.
End If
'INSTINCTS are described by the following If..Then.. statement.
If Neighbors = 1 Or Neighbors = 2 Then CnfChange = 1
If Neighbors2 = 0 And Neighbors = 1 Then CnfChange = 3
If Neighbors2 > 2 And Neighbors = 2 Then CnfChange = 3
If Abs(XMR(N) - XMF(N)) + Abs(YMR(N) - YMF(N)) + Abs(ZMR(N) - ZMF(N)) = 0 Then CnfChange = -1
If Wall(N) = 1 Or Bump(N) = 1 Or Greater2 = 1 Then CnfChange = -1
'Keep particle moving in one direction by not allowing sudden direction change.
If InhibitReversalsCheck = 1 Then
If XM1 = 1 And XM2 = 2 Then CnfChange = -1
If YM1 = 1 And YM2 = 2 Then CnfChange = -1
If ZM1 = 1 And ZM2 = 2 Then CnfChange = -1
If XM1 = 2 And XM2 = 1 Then CnfChange = -1
If YM1 = 2 And YM2 = 1 Then CnfChange = -1
If ZM1 = 2 And ZM2 = 1 Then CnfChange = -1
End If
'Adjust confidence in Memory, up or down.
Cnf(N) = Cnf(N) + CnfChange
'Make sure it stays between 0 to 3.
If Cnf(N) < 0 Then Cnf(N) = 0
If Cnf(N) > 3 Then Cnf(N) = 3
'Save Confidence in the action it tried, in Main Memory.
Call WriteMainMemory(N)
'Put together motor settings, wall bit, 2 angles, whether > 2, into one large number.
Call FormMemoryAddress(N)
'Set bot motors and confidence level to what is found at that Addr.
Call ReadMainMemory(N)
'Confidence = 0 causes the Random Guess motor response to happen.
CheckConfidence:
If Cnf(N) = 0 Then
Cnf(N) = 1
Guesses = 0 'For timeout, or endless loop when trapped.
GuessWhatToDoNext:
Guesses = Guesses + 1 'Add one to timeout count.
If Guesses > 50 Then GoTo BotIsDone
Call RandomAction(N) 'Set bots Motor with random and moving.
Call RunMotors(N) 'Run virtual motors (change X,Y in 3D array if possible)
If Bump(N) = 1 Or Wall(N) = 1 Then 'If Bump or Wall then try another guess.
AvBumps = AvBumps + Bump(N) 'First count it for screen display of how it's doing.
AvWallHits = AvWallHits + Wall(N)
GoTo GuessWhatToDoNext 'Loop back up again, to take another guess.
End If
'Count number of new memories there are, to show on screen.
If MainMem(Addr(N)) = 0 Then 'Only a never used memory Addr is zero.
Colr = NewMemColor 'Change the color bot will be drawn using.
MemCount = MemCount + 1 'They are counted to show on screen.
MemCountLabel = MemCount
End If
'Save new or updated memory.
Call WriteMainMemory(N) 'It got somewhere, so remember for next time happens.
Else
'Confidence > 0 causes the NonRandom from Memory response to be used instead of Random Guess.
Call RunMotors(N)
End If
'Do next bot Number to be done.
BotIsDone:
AvNeighbors = AvNeighbors + Neighbors
Cbot(N) = Colr
Next N
'Learned intelligence is detected by keeping track of how well they are doing, esp. confidence.
FrameNumber = FrameNumber + 1
AvConf = AvConf / Bots
AvConf2 = (AvConf2 * (1 - AvSm1)) + (AvConf * AvSm1)
AvConfLabel = Format(AvConf, "0.0#####")
AvConfLabel2 = Format(AvConf2, "0.0######")
AvBumps = AvBumps / Bots
AvBumps2 = (AvBumps2 * (1 - AvSm1)) + (AvBumps * AvSm1)
AvWallHits = AvWallHits / Bots
AvWallHits2 = (AvWallHits2 * (1 - AvSm1)) + (AvWallHits * AvSm1)
AvNeighbors = AvNeighbors / Bots
AvNeighbors2 = (AvNeighbors2 * (1 - AvSm1)) + (AvNeighbors * AvSm1)
AvBumpsLabel = Format(AvBumps, "##0.0####")
AvBumpsLabel2 = Format(AvBumps2, "##0.0######")
AvWallHitsLabel = Format(AvWallHits, "##0.0####")
AvWallHitsLabel2 = Format(AvWallHits2, "##0.0######")
AvNeighborsLabel = Format(AvNeighbors, "##0.0####")
AvNeighborsLabel2 = Format(AvNeighbors2, "##0.0######")
'If bot achieves perfect confidence 3 with repetitive cycle force them to stop being boring.
If AvConfLabel2 = "3.0" Then
If PlaceAutoCheck = 1 Then
Call DrawAllBots
Call SaveBitMapFile
Call PlaceBotsCommand_Click 'Space them in the corner again, start from scratch.
Call ResetConfidenceToShowInChart
AvConf2 = 0
End If
If PlaceRandomAutoCheck = 1 Then
Call DrawAllBots
Call SaveBitMapFile
Call PlaceRandomCommand_Click 'Or kick all over the place, for drastic change.
Call ResetConfidenceToShowInChart
AvConf2 = 0
End If
End If
'Display the averages on screen.
Call DrawGraphSmall
'Display when selected by CheckBox.
If DrawCheck = 1 Then
Call DrawAllBots
End If
Call CascadeTrail
'Let the Windows operating system handle new Events, or changes just made and more is ignored.
DoEvents
'Check to see whether new events included setting EndMainLoop = true for ending Main Loop.
If EndMainLoop = True Then
MainLoopEnded = True
Exit Sub
End If
'Now jump back up to the very start of this subroutine, instead of leaving it with "End Sub".
GoTo StartMainLoop
End Sub Ideally the "MainLoop" should be reduced to four lines so that the four "requirements" that the system has stay sorted out and simple to visualize and correlate with how this relates to biology. And please don't mind my pushing the limits of theory by the ending of this one but where scientifically possible, I can't resist: From nonrandom behavior of matter comes a progression of self-assembling emergent behaviors where at the molecular, cellular and multicellular levels each is an increasingly complex fully autonomous self-learning associative memory confidence driven intelligence system that in turn produces fractal-similar emergence at the next intelligent level on up to us. Computer models of this common intelligence system that is present at each level shows its mechanism reduces to four necessary requirements; Something for intelligence to control (motors, muscles, metabolic cycle), sensory addressable memory to store motor actions in response, feedback to gauge failure or success in actions taken, and a guess mechanism that can try a new action which may include good guesses such as in crossover exchange recombination that makes offspring a little different from each other (not clones) and recombining of small conserved domains that are the small nuts and bolts and motors of complex molecular machinery that all together keep living things alive.
Designs that successfully reproduce remain in memory in the population (gene pool) to keep going the billions year old learning process that is the cycle of life where through continual reproduction of previous state of genetic memory one replication at a time builds upon previous designs in memory. Thus a cladogram of resultant lineage shows a progression of adapting designs evidenced by the fossil record where never once was there not a predecessor of like design present in memory for the descendant design to have come from.
It is this progression of intelligent causality from nonrandom subatomic behavior in matter that makes possible the complexity of cells, speciation, Cambrian Explosion and all existing biodiversity. Without this intelligent cause, living things that we now see would not exist. This in as few words as possible explains how to detect the confidence driven behavior that is in the system, which is relatively easy to do at the program level in computer models because the variables influenced by it (like battery/energy level, hungry, overall confidence) only need to be monitored: Intelligence Detection
To determine whether a system is "intelligent" we look for the previously explained "four requirements" that must be met. This will rule out (as being intelligent) a simple heating thermostat, toilet float and water molecules. Where a system does not include all requirements for intelligence we have a system with a "behavior" that may possibly appear to be intelligent but would not qualify as an "intelligence system" or "intelligence". The four requirements will be listed as:
REQUIREMENT #1 of 4 - SOMETHING TO CONTROL
REQUIREMENT #2 of 4 - ADDRESSABLE MEMORY
REQUIREMENT #3 of 4 - FEEDBACK TO GAUGE FAILURE AND SUCCESS
REQUIREMENT #4 of 4 - ABILITY TO TAKE A GUESS
Where all four processes required to produce intelligence are properly functioning together there should also be detectable synchronized cycles indicative of proper functioning and good health. Where these cycles are no longer present the intelligence is then nonfunctional. The above will be needed after the particles are behaving in a way that an emergent level above it begins to form, which is not expected to be as easy to detect but we still have its whole world reduced to variables. Just need to know what to look for in the data that needs to be put on the screen that will rule out say an image of a panda that looks alive when the model was more like forming a cloudlike substance with no memory of any kind to tap into (thus ruled out as intelligent) that just happened to take the shape of one. In biology Self-Replicating RNA's are a very simple nucleotide coded memory system, with an analogy that might form in the virtual environment where there is no doubt they are coded entities competing with other and getting better at controlling each other and their environment as time goes on. Prions are proteins that compete with each other and their environment much the same way but not enough seems to be known about them to rule one way or the other. Hopefully that helped explain missing details you or anyone else needs to know to program it. And if a new more fractal based theory is possible I'll leave that up to consensus because the title of the one I already have is overwhelming enough. To make a long story short the releasing of the Ben Stein movie that made it seem like the Discovery Institute had a useful intelligence theory stirred me up real good. After weeks of brainstorming an idea to counter it, all the science that was scattered around blogs and ftp came together in my mind then I wrote it down before forgetting how the logic connected together to become the theory they asked for. Only problem is it's a theory that's supposed to be impossible to write and illegal in the school district of Dover, PA so I made sure the teachers there were the first to know through their local newspaper forum so that reporters and others will know what it is and not worry about it. I was also published for a self-assembly experiment that developed in another forum (and was inspired by a few Creationists) that was published by the National Science Teacher Association (#1 science education journal) to help introduce the self-assembly concept to US schools. Since it demonstrates an entirely "nonrandom" process the dwelling on the origin of life having been from "random" processes met its match with just a little egg yolk in oil in water and shaken into phospholipid membranes as makes our cell membranes. All cellular organelles replicate without genetic code just self-assembly that kinda goes "Poof!" and it's there, can't evolve. So even though it's just science and I cannot see myself as a Creationist (just follow the science where it leads) it just sometimes goes in very surprising directions. But those who genuinely want the controversy to end with everyone having learned something can see who is making progress and who is not. And it is vitally important that a lot of honest people (who are not looking to destroy science ) who put a lot of faith in a theory of ID understand why Dover went so badly for the Discovery Institute's theory. It was a "premise" for a theory and arguments against another when a theory has to stand on it's own scientific merit. Must explain how the stated phenomena works in a way that others can experiment with it. In time that becomes a part of science without needing to influence school boards or anything messy like that. Power the Discovery Institute could only wish it had, which only belongs to those who are genuinely into science entirely for science's sake. When you are and you know how to evidence what supposedly 99% of scientists agree was impossible it's impossible not to try to at least see what does develop. It's one way of solving a complex scientific problem where you start off with the goal to evidence something you only have an abstract generalization for that describes it in this case "Intelligent Cause" then do the best you can to reach that goal. Where you fall short you still end up further than anyone has ever been along that path of discovery, so you write up what you have minus the far reaching goal that would be out of place where it's obvious to others that it was not really evidenced. But in this case "Intelligent Cause" is probably the best two word phrase to describe the type of causation being experimented with and theories change how words and phrases are defined, not the other way around. Since you might find (or already found) the http://theoryofid.blogspot.com/ site I had to explain what that is for. It currently has the info on previously mentioned RNA's and other things pertaining to biology and other sciences plus origin of life experiments. A new theory or great paper that could be made in this forum would be worded in with as few sentences as possible then linked to in its References like they normally do in science papers. Gives credit where due and don't have to cover the same ground all over again. And whatever may be collaborated is yours as separate without being the other, so that none of us have the strings that are attached to the other theory attached to us when just working with fractal math/science none have good reason to get all shook up over. |
|
|
|
« Last Edit: January 07, 2010, 04:15:48 PM by Gary Gaulin »
|
Logged
|
|
|
|
|
kram1032
|
|
« Reply #5 on: January 07, 2010, 09:34:40 PM » |
|
I fear not lol My current graphics card doesn't even support openCL  However a combination of openCL and openGL could not only increase the performance by, like, an infinite times, but at the same time could give an environment which is graphically more meaningful to a, uh.... average person I just saw some demonstrations of OpenCL and this kind of calculation must be about perfect for the advantages. Those particles by now must have learned quite a bit as they don't hit the walls that often anymore and start to form groups here Right now I'm using 7 Particles (none stationary) and a world size of 11. Current Memory-count is just over 4000 but I think it didn't save the memory count from yesterday. I just let it ran without drawing during surfing and dining and stuff and just checked draw from time to time to see if something changed. Pure randomness by now is definitely gone Maybe with some optimization you could do massiv simulations with a million particles or so - the theoretical mechanics behind it don't look too complex. There where more complicated particle systems made with reasonablish render times. Some youtube-vids with interactive particles on Cuda or OpenCL can be found A rather simple thing like this, not even requiring complex formulas could hit amazing performance that way. - However I dunno which Graphics Cards they used on those YouTube vids... Maybe two cards at once or all the possible craziness... (Too bad I can't use either, using a 2-year-old ATI^^) What would be interesting too would be some kind of learning BOIDS with, say, 5 species with different pretador/prey configurations but each species at dirst has to learn how to move nicely like that single robot has to to find his food-sources (the charging platforms) One could be omnivorous, one could be a plant-eater (forgot the nice name xD) and the other three differently sized carnivores "eating" each other with different conditions, for instance the bigger can only be eaten if there are more of the smaller... Plants would basically spawn randomly. However that could easily be way more complex as you'd also have to consider dynamic particle numbers when an individuum dies or gets born and when they get born. Basically learning BOIDS with the abillity to reproduce. At first, both for simplicity and actually also closer correlation to the real world, just do that on a toric or if somehow possible spherical 2D world (Talking about kinds of worlds: I once looked into how a world on a klein-bottle would look like, rather than a torus. It would be like on a torus, you go out of the world on one side and come back in on the other but the side you'd walk would be mirrored. So if you go out of the world at the left side of the top edge, you would come back in at the right side of the bottom edge. Such a world could also be interesting ) I'm not sure if I entirely followed you with all your descriptions but if I got it right, you'd like to kind of copy the DNA's coding and just let it run so intelligent-ish structures start to emerge. Very interesting but looking at how long just such a simple system as the Particles in a Box takes to learn, something like the whole evolution will take for ever even with openCL except if you have a huge computation farm or something However you could and probably should do some computational optimizations behind the structure of computer-data: In DNA, information is stored in triplets basically and each combination has some double meanings. Those triplets form amino acids which then in one way or an other allow communication in the whole body. DNA by that doesn't really follow a numberic approach but rather a boolean-ish just with a more diffuse system. The double-meanings for instance help to at one hand allow variations which at the other hand don't change anything at all. For computers, such a unit wouldn't be a triplet but an octet - the byte. Interesting would be if you could even find some rules to let it order itself. Find some basic rules which lead to self organizing rules. Let it define on itself, how many codes need to be given twice or even trice and what each code actually means. Seen like that, defining two things (00 and 11) as stop is actually a natural thing as that's just what nature does to ensure greater stabillity. All that is very interesting but the sheer complexity has an attractor at infinty |
|
|
|
|
Logged
|
|
|
|
|
Gary Gaulin
Guest
|
|
« Reply #6 on: January 08, 2010, 07:39:33 PM » |
|
I fear not lol My current graphics card doesn't even support openCL I have a Celeron CPU so I'm early Pentium. The program speeds things up with the "Draw" checkbox that skips all the graphics to go 20 times or so faster. Even though graphics are slower it's only needed to peek in every once in a while and where it goes much faster needs slowing down to see what the particles are doing. Works good enough for what we now have for hardware, and software. Those particles by now must have learned quite a bit as they don't hit the walls that often anymore and start to form groups here Right now I'm using 7 Particles (none stationary) and a world size of 11. Current Memory-count is just over 4000 but I think it didn't save the memory count from yesterday. I just let it ran without drawing during surfing and dining and stuff and just checked draw from time to time to see if something changed. Pure randomness by now is definitely gone I left mine on overnight. Used 100 particles with the Inhibit Reversals unchecked, then when I got back to it they were going in tight circles back to start and jittering one step forward then back. They are all perfect solutions to the problem of avoiding walls but keep moving while staying together, achieved 100% success. Maybe with some optimization you could do massiv simulations with a million particles or so - the theoretical mechanics behind it don't look too complex. What needs working on most now is the way the confidence is adjusted, to better approximate the properties of an atom. Otherwise not have anything worth trying a million particles for. What would be interesting too would be some kind of learning BOIDS with, say, 5 species with different pretador/prey configurations but each species at dirst has to learn how to move nicely like that single robot has to to find his food-sources (the charging platforms)
One could be omnivorous, one could be a plant-eater (forgot the nice name xD) and the other three differently sized carnivores "eating" each other with different conditions, for instance the bigger can only be eaten if there are more of the smaller...
Plants would basically spawn randomly.
However that could easily be way more complex as you'd also have to consider dynamic particle numbers when an individuum dies or gets born and when they get born.
Basically learning BOIDS with the abillity to reproduce. You have the right idea. And the ability to reproduce would be essential. It's such a part of what keeps our kind of intelligence going I just finished having to write about (PG13) where males/female sexuality comes from which is important to know to be able to model it properly. Can say that in the beginning, were hermaphrodites. And they're still here. So if they're not in the model then it's not a realistic approximation of the real thing, for when that is trying to be achieved. Once the particles behave like atoms they can be assembled into proteins (made by hydrogen-bonding amino acids together) and ribonucleotides to make prions and self-replicating RNA flocking around certain things without worrying about having to add BOIDS behavior. It might be possible to derive a Confidence +1/-1 number from valence charts and whatever else is on the internet. I also have a Lennard-Jones model that makes them attract and repel properly. Forms the predictable hexagonal close-packing pattern. I'll post the code for it here, in case anyone needs it. Plenty on the internet on it too. The formula is: V = 4 * Epsilon * ((Sigma / R) ^ 12 - (Sigma / R) ^ 6) V is the velocity, same thing as how many pixels to move one way or another. Others adjust size of particle and attraction strength between them. I put it on the screen in 2D with this: Const XYpixels = 512
Const SC = 8
Dim XmYm(XYpixels, XYpixels, 1) As Byte
Dim Mols As Long
Dim M As Long
Dim Xm(10000) As Double
Dim Ym(10000) As Double
Dim Zm(10000) As Double
Dim X, Y, Z As Double
Dim R As Double
Dim Epsilon As Double
Dim Sigma As Double
Dim V As Double
Dim N As Long
Dim N1 As Long
Dim N2 As Long
Dim DX, DY, DZ As Double
Dim Xv, Yv, Zv As Double
Private Sub Form_Load()
Epsilon = 1
Sigma = 1
Call PlotCurve
Pic1.DrawWidth = SC / 2
Pic2.DrawWidth = SC / 2
Call PlaceMolecules
Timer1.Enabled = True
End Sub
Private Sub PlotCurve()
N = Picture1.Width / 3
For R = 0.8 To 3 Step 0.0005
V = 4 * Epsilon * ((Sigma / R) ^ 12 - (Sigma / R) ^ 6)
Picture1.PSet (R * N, (V * 100) + (Picture1.Height / 2)), 0
Next R
End Sub
Private Sub PlaceMolecules()
Mols = 0
For X = 1 To 9
For Y = 1 To 9
Mols = Mols + 1
Xm(Mols) = X
Ym(Mols) = Y
Next Y
Next X
M = Mols / 2
End Sub
Private Sub Timer1_Timer()
For N1 = 1 To Mols
Xv = 0
Yv = 0
For N2 = 1 To Mols
If N1 = N2 Then GoTo Same1
DX = Xm(N2) - Xm(N1)
DY = Ym(N2) - Ym(N1)
R = Sqr(DX * DX + DY * DY)
V = 4 * Epsilon * ((Sigma / R) ^ 12 - (Sigma / R) ^ 6)
V = V / 80
Xv = Xv + (V * DX)
Yv = Yv + (V * DY)
Same1:
Next N2
Xm(N1) = Xm(N1) - Xv
Ym(N1) = Ym(N1) - Yv
Next N1
'Heat Entropy adds a small random amount of displacement to X,Y position.
' Call Entropy
Pic1.Cls
For N = 1 To Mols
Pic1.PSet (Xm(N) * SC, Ym(N) * SC), 0
Next N
Pic2.Refresh
End Sub
Private Sub Entropy()
For N = 1 To Mols
Xm(N) = Xm(N) + (Fix((Rnd * 3) - 1.5) / 8)
Ym(N) = Ym(N) + (Fix((Rnd * 3) - 1.5) / 8)
Next N
End Sub
(Talking about kinds of worlds: I once looked into how a world on a klein-bottle would look like, rather than a torus. It would be like on a torus, you go out of the world on one side and come back in on the other but the side you'd walk would be mirrored. So if you go out of the world at the left side of the top edge, you would come back in at the right side of the bottom edge. Such a world could also be interesting ) A physics mind too! I kinda expected that. I'm not sure if I entirely followed you with all your descriptions but if I got it right, you'd like to kind of copy the DNA's coding and just let it run so intelligent-ish structures start to emerge. Very interesting but looking at how long just such a simple system as the Particles in a Box takes to learn, something like the whole evolution will take for ever even with openCL except if you have a huge computation farm or something Only need to get something like Self-Replicating DNA on the screen and even Jerald Joyce would be here checking that one out, and I'm not kidding either. But to make that we need to make a single ribonucleotide which is not that many atoms/particles. So I keep it down to something that we can do right now with whatever we already code in, by only needing to only make small molecules. However you could and probably should do some computational optimizations behind the structure of computer-data:
In DNA, information is stored in triplets basically and each combination has some double meanings.
Those triplets form amino acids which then in one way or an other allow communication in the whole body. All that would be due to arrangement of atoms in molecule. Would happen on its own without needing to be put in, just by having the behavior of atoms in the particles well enough approximated. All that is very interesting but the sheer complexity has an attractor at infinty I think I know what you are saying about the attractor, maybe not. But I look at it as just needing to get a single +1/-1 number to adjust confidence by from a subroutine that should be possible by compiling valence and other chemical data. Not have to care when it is solid liquid or gas just the basic dynamics built into the bot so it gets the Lennard-Jones attraction (throttle of its X,Y,Z Motors) get set in the right directions, does not have to be precise value just right direction. Otherwise have the problems you mentioned where there is so much calculating needing to be done it's beyond what we could model on our PC. Here, once the memory has been trained the bots only read that to find out what to do in response (direction to set motors) so can go fast enough to see something happen. Even small molecules behaving like the real thing would be amazing. Anything having to do with heat and pressure is easy. Should be able to set the temperature scroll by finding what level water molecules become ice. And heat vibration is like the old Rock Em' Sock Em Robot's that moved on vibrating table, need that but its easy. I know that we do have to account for Valence. But that's easy too, here's the numbers: 'Number Element Valence
1 Hydrogen (-1), +1
2 Helium 0
3 Lithium +1
4 Beryllium +2
5 Boron -3, +3
6 Carbon (+2), +4
7 Nitrogen -3, -2, -1, (+1), +2, +3, +4, +5
8 Oxygen -2
9 Fluorine -1, (+1)
10 Neon 0
11 Sodium +1
12 Magnesium +2
13 Aluminum +3
14 Silicon -4, (+2), +4
15 Phosphorus -3, +1, +3, +5
16 Sulfur -2, +2, +4, +6
17 Chlorine -1, +1, (+2), +3, (+4), +5, +7
18 Argon 0
19 Potassium +1
20 Calcium +2
21 Scandium +3
22 Titanium +2, +3, +4
23 Vanadium +2, +3, +4, +5
24 Chromium +2, +3, +6
25 Manganese +2, (+3), +4, (+6), +7
26 Iron +2, +3, (+4), (+6)
27 Cobalt +2, +3, (+4)
28 Nickel (+1), +2, (+3), (+4)
29 Copper +1, +2, (+3)
30 Zinc +2
31 Gallium (+2). +3
32 Germanium -4, +2, +4
33 Arsenic -3, (+2), +3, +5
34 Selenium -2, (+2), +4, +6
35 Bromine -1, +1, (+3), (+4), +5
36 Krypton 0
37 Rubidium +1
38 Strontium +2
39 Yttrium +3
40 Zirconium (+2), (+3), +4
41 Niobium (+2), +3, (+4), +5
42 Molybdenum (+2), +3, (+4), (+5), +6
43 Technetium +6
44 Rubidium (+2), +3, +4, (+6), (+7), +8
45 Rhodium (+2), (+3), +4, (+6)
46 Palladium +2, +4, (+6)
47 Silver +1, (+2), (+3)
48 Cadmium (+1), +2
49 Indium (+1), (+2), +3
50 Tin +2, +4
51 Antimony -3, +3, (+4), +5
52 Tellurium -2, (+2), +4, +6
53 Iodine -1, +1, (+3), (+4), +5, +7
54 Xenon 0
55 Cesium +1
56 Barium +2
57 Lanthanum +3
58 Cerium +3, +4
59 Praseodymium +3
60 Neodymium +3, +4
61 Promethium +3
62 Samarium (+2), +3
63 Europium (+2), +3
64 Gadolinium +3
65 Terbium +3, +4
66 Dysprosium +3
67 Holmium +3
68 Erbium +3
69 Thulium (+2), +3
70 Ytterbium (+2), +3
71 Lutetium +3
72 Hafnium +4
73 Tantalum (+3), (+4), +5
74 Tungsten (+2), (+3), (+4), (+5), +6
75 Rhenium (-1), (+1), +2, (+3), +4, (+5), +6, +7
76 Osmium (+2), +3, +4, +6, +8
77 Iridium (+1), (+2), +3, +4, +6
78 Platinum (+1), +2, (+3), +4, +6
79 Gold +1, (+2), +3
80 Mercury +1, +2
81 Thallium +1, (+2), +3
82 Lead +2, +4
83 Bismuth (-3), (+2), +3, (+4), (+5)
84 Polonium (-2), +2, +4, (+6)
85 Astatine ?
86 Radon 0
87 Francium ?
88 Radium +2
89 Actinium +3
90 Thorium +4
91 Protactinium +5
92 Uranium (+2), +3, +4, (+5), +6
Radii too: 'From: http://environmentalchemistry.com/yogi/periodic/atomicradius.html
'Periodic Table of Elements
'Sorted by Atomic Radius
' Name Sym #
0.49 Å Helium He 2
0.51 Å Neon Ne 10
0.57 Å Fluorine F 9
0.65 Å Oxygen O 8
0.75 Å Nitrogen N 7
0.79 Å Hydrogen H 1
0.88 Å Argon Ar 18
0.91 Å Carbon C 6
0.97 Å Chlorine Cl 17
1.03 Å Krypton Kr 36
1.09 Å Sulfur S 16
1.12 Å Bromine Br 35
1.17 Å Boron B 5
1.22 Å Selenium Se 34
1.23 Å Phosphorus P 15
1.24 Å Xenon Xe 54
1.32 Å Iodine I 53
1.33 Å Arsenic As 33
1.34 Å Radon Rn 86
1.4 Å Beryllium Be 4
1.42 Å Tellurium Te 52
1.43 Å Astatine At 85
1.46 Å Silicon Si 14
1.52 Å Germanium Ge 32
1.53 Å Zinc Zn 30
1.53 Å Antimony Sb 51
1.53 Å Polonium Po 84
1.57 Å Copper Cu 29
1.62 Å Nickel Ni 28
1.63 Å Bismuth Bi 83
1.67 Å Cobalt Co 27
1.71 Å Cadmium Cd 48
1.72 Å Iron Fe 26
1.72 Å Tin Sn 50
1.72 Å Magnesium Mg 12
1.75 Å Silver Ag 47
1.76 Å Mercury Hg 80
1.79 Å Gold Au 79
1.79 Å Manganese Mn 25
1.79 Å Palladium Pd 46
1.81 Å Gallium Ga 31
1.81 Å Lead Pb 82
1.82 Å Aluminum Al 13
1.83 Å Platinum Pt 78
1.83 Å Rhodium Rh 45
1.85 Å Chromium Cr 24
1.87 Å Iridium Ir 77
1.89 Å Ruthenium Ru 44
1.92 Å Vanadium V 23
1.92 Å Osmium Os 76
1.95 Å Technetium Tc 43
1.97 Å Rhenium Re 75
2.0 Å Titanium Ti 22
2.0 Å Indium In 49
2.01 Å Molybdenum Mo 42
2.02 Å Tungsten W 74
2.05 Å Lithium Li 3
2.08 Å Niobium Nb 41
2.08 Å Thallium Tl 81
2.09 Å Scandium Sc 21
2.09 Å Tantalum Ta 73
2.16 Å Hafnium Hf 72
2.16 Å Zirconium Zr 40
2.23 Å Calcium Ca 20
2.23 Å Sodium Na 11
2.25 Å Lutetium Lu 71
2.27 Å Yttrium Y 39
2.4 Å Ytterbium Yb 70
2.42 Å Thulium Tm 69
2.45 Å Erbium Er 68
2.45 Å Strontium Sr 38
2.47 Å Holmium Ho 67
2.49 Å Dysprosium Dy 66
2.51 Å Terbium Tb 65
2.54 Å Gadolinium Gd 64
2.56 Å Europium Eu 63
2.59 Å Samarium Sm 62
2.62 Å Promethium Pm 61
2.64 Å Neodymium Nd 60
2.67 Å Praseodymium Pr 59
2.7 Å Cerium Ce 58
2.74 Å Lanthanum La 57
2.77 Å Potassium K 19
2.78 Å Barium Ba 56
2.98 Å Rubidium Rb 37
3.34 Å Cesium Cs 55
And Electron Configuration:
Atomic electron configuration table
This is a table of electron configurations of atoms.
No. Element 1 2 3 4 5 6 7
1 Hydrogen 1
2 Helium 2
3 Lithium 2 1
4 Beryllium 2 2
5 Boron 2 3
6 Carbon 2 4
7 Nitrogen 2 5
8 Oxygen 2 6
9 Fluorine 2 7
10 Neon 2 8
11 Sodium 2 8 1
12 Magnesium 2 8 2
13 Aluminium 2 8 3
14 Silicon 2 8 4
15 Phosphorus 2 8 5
16 Sulfur 2 8 6
17 Chlorine 2 8 7
18 Argon 2 8 8
19 Potassium 2 8 8 1
20 Calcium 2 8 8 2
21 Scandium 2 8 9 2
22 Titanium 2 8 10 2
23 Vanadium 2 8 11 2
24 Chromium 2 8 13 1
25 Manganese 2 8 13 2
26 Iron 2 8 14 2
27 Cobalt 2 8 15 2
28 Nickel 2 8 16 2
29 Copper 2 8 18 1
30 Zinc 2 8 18 2
31 Gallium 2 8 18 3
32 Germanium 2 8 18 4
33 Arsenic 2 8 18 5
34 Selenium 2 8 18 6
35 Bromine 2 8 18 7
36 Krypton 2 8 18 8
37 Rubidium 2 8 18 8 1
38 Strontium 2 8 18 8 2
39 Yttrium 2 8 18 9 2
40 Zirconium 2 8 18 10 2
41 Niobium 2 8 18 12 1
42 Molybdenum 2 8 18 13 1
43 Technetium 2 8 18 14 1
44 Ruthenium 2 8 18 15 1
45 Rhodium 2 8 18 16 1
46 Palladium 2 8 18 18 0
47 Silver 2 8 18 18 1
48 Cadmium 2 8 18 18 2
49 Indium 2 8 18 18 3
50 Tin 2 8 18 18 4
51 Antimony 2 8 18 18 5
52 Tellurium 2 8 18 18 6
53 Iodine 2 8 18 18 7
54 Xenon 2 8 18 18 8
55 Caesium 2 8 18 18 8 1
56 Barium 2 8 18 18 8 2
57 Lanthanum 2 8 18 18 9 2
58 Cerium 2 8 18 19 9 2
59 Praseodymium 2 8 18 21 8 2
60 Neodymium 2 8 18 22 8 2
61 Promethium 2 8 18 23 8 2
62 Samarium 2 8 18 24 8 2
63 Europium 2 8 18 25 8 2
64 Gadolinium 2 8 18 25 9 2
65 Terbium 2 8 18 27 8 2
66 Dysprosium 2 8 18 28 8 2
67 Holmium 2 8 18 29 8 2
68 Erbium 2 8 18 30 8 2
69 Thulium 2 8 18 31 8 2
70 Ytterbium 2 8 18 32 8 2
71 Lutetium 2 8 18 32 9 2
72 Hafnium 2 8 18 32 10 2
73 Tantalum 2 8 18 32 11 2
74 Tungsten 2 8 18 32 12 2
75 Rhenium 2 8 18 32 13 2
76 Osmium 2 8 18 32 14 2
77 Iridium 2 8 18 32 15 2
78 Platinum 2 8 18 32 17 1
79 Gold 2 8 18 32 18 1
80 Mercury 2 8 18 32 18 2
81 Thallium 2 8 18 32 18 3
82 Lead 2 8 18 32 18 4
83 Bismuth 2 8 18 32 18 5
84 Polonium 2 8 18 32 18 6
85 Astatine 2 8 18 32 18 7
86 Radon 2 8 18 32 18 8
87 Francium 2 8 18 32 18 8 1
88 Radium 2 8 18 32 18 8 2
89 Actinium 2 8 18 32 18 9 2
90 Thorium 2 8 18 32 18 10 2
91 Protactinium 2 8 18 32 20 9 2
92 Uranium 2 8 18 32 21 9 2
93 Neptunium 2 8 18 32 22 9 2
94 Plutonium 2 8 18 32 24 8 2
95 Americium 2 8 18 32 25 8 2
96 Curium 2 8 18 32 25 9 2
97 Berkelium 2 8 18 32 27 8 2
98 Californium 2 8 18 32 28 8 2
99 Einsteinium 2 8 18 32 29 8 2
100 Fermium 2 8 18 32 30 8 2
101 Mendelevium 2 8 18 32 31 8 2
102 Nobelium 2 8 18 32 32 8 2
103 Lawrencium 2 8 18 32 32 9 2
104 Rutherfordium 2 8 18 32 32 10 2
105 Dubnium 2 8 18 32 32 11 2
106 Seaborgium 2 8 18 32 32 12 2
107 Bohrium 2 8 18 32 32 13 2
108 Hassium 2 8 18 32 32 14 2
109 Meitnerium 2 8 18 32 32 15 2
110 Darmstadtium 2 8 18 32 32 16 2
111 Roentgenium 2 8 18 32 32 17 2
112 Ununbium 2 8 18 32 32 18 2
113 Ununtrium 2 8 18 32 32 18 3
114 Ununquadium 2 8 18 32 32 18 4
115 Ununpentium 2 8 18 32 32 18 5
116 Ununhexium 2 8 18 32 32 18 6
117 Ununseptium 2 8 18 32 32 18 7
118 Ununoctium 2 8 18 32 32 18 8
Names are good to have too: 1 Hydrogen H
2 Helium He
3 Lithium Li
4 Beryllium Be
5 Boron B
6 Carbon C
7 Nitrogen N
8 Oxygen O
9 Fluorine F
10 Neon Ne
11 Sodium N
12 Magnesium Mg
13 Aluminum Al
14 Silicon Si
15 Phosphorus P
16 Sulfur S
17 Chlorine Cl
18 Argon Ar
19 Potassium K
20 Calcium Ca
21 Scandium Sc
22 Titanium Ti
23 Vanadium V
24 Chromium Cr
25 Manganese Mn
26 Iron Fe
27 Cobalt Co
28 Nickel Ni
29 Copper Cu
30 Zinc Zn
31 Gallium Ga
32 Germanium Ge
33 Arsenic As
34 Selenium Se
35 Bromine Br
36 Krypton Kr
37 Rubidium Rb
38 Strontium Sr
39 Yttrium Y
40 Zirconium Zr
41 Niobium Nb
42 Molybdenum Mo
43 Technetium Tc
44 Ruthenium Ru
45 Rhodium Rh
46 Palladium Pd
47 Silver Ag
48 Cadmium Cd
49 Indium In
50 Tin Sn
51 Antimony Sb
52 Tellurium Te
53 Iodine I
54 Xenon Xe
55 Cesium Cs
56 Barium Ba
57 Lanthanum La
58 Cerium Ce
59 Praseodymium Pr
60 Neodymium Nd
61 Promethium Pm
62 Samarium Sm
63 Europium Eu
64 Gadolinium Gd
65 Terbium Tb
66 Dysprosium Dy
67 Holmium Ho
68 Erbium Er
69 Thulium Tm
70 Ytterbium Yb
71 Lutetium Lu
72 Hafnium Hf
73 Tantalum Ta
74 Tungsten W
75 Rhenium Re
76 Osmium Os
77 Iridium Ir
78 Platinum Pt
79 Gold Au
80 Mercury Hg
81 Thallium Tl
82 Lead Pb
83 Bismuth Bi
84 Polonium Po
85 Astatine At
86 Radon Rn
Now that all the data files that I found are here, you can see what there is to design a bot from. Let me know where you know of something missing that needs to be added. If there is then it doesn't seem like there's much more needing to be accounted for. Most everything else like density seems calculable from that. Might be easiest to start by designing is by the way memory Addressing would have to look. Data stored at each Address location is just the X,Y,Z motor settings so the rest after that point in the circuit is easy. Going the other way from Addressing is Sensory to neighbors in a 3D array system that does not need to calculate far away particles. Later can tile interactions, like to make a vesicle one tile is repeated around the sphere that from there holds itself roughly in that shape depending on heat and what it is attracted to or repelled from in the various directions. There can sum up atom on up behavior of phospholipids and such in one bot. But first need the atom bots that make summing up molecule bot behavior easy. I have also connected one memory system so that Sensory addresses Data, that is Sensory for the next where it's in a sense summed up so that only necessary detail be sent on down the line to where eventually Data goes straight (with added entropy amount) to X,Y,Z motors (move pixel that amount on screen). I like to keep things as simple as possible, so that what I propose is not expensive or hard to code. What makes it scientifically novel is the way it works, the confidence/behavior/intelligence algorithm and theory it is part of. Hard part is explaining what makes it all relatively straightforward. And explaining my sordid scientific past that includes still entertaining that taboo theory but I cannot help myself, it's too useful for helping to make predictions that lead to science that is not written anywhere yet. Besides, even university level intelligence science took a hammering by just having the "I" word in it. But in that area evolutionary theory that now exists can really only say that things change over time and along with it intelligence evolves and that's honestly about it, and I know from having been one that tried to explain origin of intelligence with existing theory with 0 success convincing someone of that. They wanted to know where love and emotions plus consciousness came from and are for, not be told "natural selection did it" for the 250'th time. ET simply cannot go there. Needed a theory that can predict a starting state of hermaphrodite which is somewhat contrary to what the prevailing wisdom saw as the starting point for reproduction on up to our level which means their models are boring and incomplete. And my wife was laughing to tears reading where males and females come from in a link above, so it obviously helps make science fun which is important due to getting people to read something scientific is not easy either. I'm glad there are exceptions here. In edit (for anyone trying this) I have to add that later on energy levels will need to be added. Increasing the energy level of oxygen should shift the virtual orbits of electrons to bond to form O3 ozone, in the unenergized O2 environment. |
|
|
|
« Last Edit: January 11, 2010, 07:29:12 AM by Gary Gaulin »
|
Logged
|
|
|
|
|
kram1032
|
|
« Reply #7 on: January 11, 2010, 10:50:10 PM » |
|
hmmm.... Do you think the "atoms" would "learn" faster if you split the things to be learned into different memory files, running different simmulations for each with different things being considered, rather than throw the "full problem" on them? It would be cool if you could get rid of the grid so the movements could become "smooth" |
|
|
|
|
Logged
|
|
|
|
|
Gary Gaulin
Guest
|
|
« Reply #8 on: January 12, 2010, 01:23:14 PM » |
|
hmmm.... Do you think the "atoms" would "learn" faster if you split the things to be learned into different memory files, running different simmulations for each with different things being considered, rather than throw the "full problem" on them? Good thinking! I thought about that too. From what I have seen from their behavior it might take a little longer to train them one problem at a time (but never confirmed it). An analogy for why, would be learning to jump across to the other side of something by going full speed at it. It would not take long to learn how to do that. Then we could add the next problem where if they go too fast then they bash into the wall just on the other side. Where each has its own memory (not needed in this model since identical particles have identical behaviors) they would all crash into the wall from being so used to there not being one there. Same thing as pulling a chair from under someone just before they sit down which even causes injury it's such a "surprise". They are not used to that happening because chairs do not normally move. Where they did they would be more careful and "keep an eye on it" while sitting. As a result all the time training them to be very successful at one task would make them immediately terrible at another that has changing parameters, in which case they become like lemmings following each other off a cliff. Might as well try to pull the chair from under them while first learning to use one so that they know it can happen when a prankster is behind them, so are prepared for it then do the right thing every time. Dividing the problem is though very important to how memory is constructed. For example there is the ring of neurons that gives awareness of where something went that it cannot see, which adds a very simple memory circuit to the one that does the actual thinking which greatly increases navigation success. In that case the full problem only needed to be divided at the bot design level. It would be cool if you could get rid of the grid so the movements could become "smooth" I agree. And it's not that hard to do. Just have to use a different world-array system where their velocity is a variable derived from motor acceleration and direction stored as precise x,y,z points in space (as opposed to one of the possible squares or cubes in the world to step to). This is great because the world can then be as large as you want and motion would be smooth, but the drawbacks are slower speed caused by having to calculate all movements and positions in physical space instead of an instant step in one of the possible directions. Also, the properties of atoms include integer value "harmonics" that double like *2, *4, *8 or triple *1, *3, which suggests that a grid might be able to do a better job than it appears. And you might have noticed that when the particles become unstable they fly off in one or more directions with a repeating wave with a frequency that is here in the shape of triangular, square, or combination that produces a "wavelength" and the particle is acting as both a particle and a wave (which as you may know is still a big mystery of science to fully understand). Even though the model has low granularity, that might be all it needs. So instead of starting with something that has been complicated by an algorithm that adds its properties to those of the algorithm that produces intelligent behavior I started with the least complex method. From what I can see it accounts for mass and other properties that could also be calculated into it but doing that properly would require knowing what is already being expressed by the particles. Guess you can say that in this case it's best to learn to walk before running or else all that would be expected are stumbling bots. |
|
|
|
|
Logged
|
|
|
|
|
kram1032
|
|
« Reply #9 on: January 12, 2010, 02:52:03 PM » |
|
I see I have yet an other idea: Right now, your particles totally rely on the confidence-system as some kind of fixed (hardcoded) base of intelligence. But what if you would add the confidence-system inside the learning recursion? Do some kind of general version where the "optimal confidence behaviour" would be learned on the fly during trying to learn the "rules of the game" So, not only learn to behave intelligent but also learn to actually learn If you look at all the confidence systems you tried so far, do you see some kind of generic pattern which could be generalized to a learnable method? Confidence-driven confidence behaviour? |
|
|
|
|
Logged
|
|
|
|
|
LesPaul
Guest
|
|
« Reply #10 on: January 12, 2010, 10:47:03 PM » |
|
Gary, your initial post sounds rather similar to concepts seen frequently in "genetic algorithms" and "genetic programming." These might be really interesting to you if you're looking for further research. As a very high-level introduction, genetic algorithms are computer programs that find their solution by random guessing, but refined by a "fitness function." This sounds very much like what you're describing. There are many techniques in genetic algorithms that I think you would find very useful -- things like "crossover" and "mutation" and "selectivity" play very important roles in helping the population of random guesses approach the best solution. To take a step further, you come to "genetic programming." In a genetic algorithm, the actual program is fixed (written by a human) and it uses the method described above to find the solution to a problem, for example, the best shape of an antenna. In genetic programming, the program itself is not fixed and is actually evolved over time. So, little bits of code like a for-loop or an if-else statement make up the population and they are randomly smashed together. Here, the "fitness function" is simply the measure of how close the program comes to solving the problem you're trying to solve. You might think that just random pieces of code would take billions of years to actually do anything useful, but it is surprising how effective this technique can be. For example, John Koza ( http://www.genetic-programming.org) has created a sort of home-brew supercomputer that runs genetic programs all day long, and it has produced results substantial enough that his machine has been granted a patent! I don't mean that the machine itself is patented, I mean that the machine actually invented something that was worthy of a patent! It was the first non-human to ever be granted a patent. [edit: fixed "John Koza" typo] |
|
|
|
« Last Edit: January 12, 2010, 11:00:47 PM by LesPaul »
|
Logged
|
|
|
|
|
Gary Gaulin
Guest
|
|
« Reply #11 on: January 13, 2010, 09:22:41 AM » |
|
Hi LesPaul, thanks for mentioning all that! Gary, your initial post sounds rather similar to concepts seen frequently in "genetic algorithms" and "genetic programming." These might be really interesting to you if you're looking for further research. Yes, it is similar to a genetic algorithm. As a very high-level introduction, genetic algorithms are computer programs that find their solution by random guessing, but refined by a "fitness function." This sounds very much like what you're describing. There are many techniques in genetic algorithms that I think you would find very useful -- things like "crossover" and "mutation" and "selectivity" play very important roles in helping the population of random guesses approach the best solution. In this model "random guess" is the most rudimentary guess mechanism, with it being better (but not always necessary) to include "good guess" that can be added a number of ways. Same idea though. Finding the best solution would be necessary for a GA too so in that sense they are the same but it's "confidence driven" which is entirely internal as opposed to "natural selection" or "fitness" that is external. It is either able to survive (possibly reproduce) in the environment or it is not. Modeling a reproducing animal could be accomplished by each gene being a small bot responding to local control molecules and conditions, possibly many genes combined into one memory that acts as one unit of exchange during crossover. Where it needs one, another memory would be it's brain. Since there is university level evidence that the centrosome is also one of these mechanisms it would be used to add complex IR sensing (and what else they sense) guided migration behavior to the cell(s). After you have critters wandering around on the screen you can then be the natural selector by zapping out the slow boring ones. I doubt they would need that to speciate. We're here essentially reducing variation in the population needed for new species to emerge. Zapping out the slow ones would eliminate all hope of there ever being snails, starfish, and other slow creatures in the environment. To take a step further, you come to "genetic programming." In a genetic algorithm, the actual program is fixed (written by a human) and it uses the method described above to find the solution to a problem, for example, the best shape of an antenna. In genetic programming, the program itself is not fixed and is actually evolved over time. So, little bits of code like a for-loop or an if-else statement make up the population and they are randomly smashed together. Here, the "fitness function" is simply the measure of how close the program comes to solving the problem you're trying to solve. In this algorithm there are no loops being tried in new combinations, it is always just a simple Input/Output memory like a RAM chip, neural network, genome. You might think that just random pieces of code would take billions of years to actually do anything useful, but it is surprising how effective this technique can be. For example, John Koza ( http://www.genetic-programming.org) has created a sort of home-brew supercomputer that runs genetic programs all day long, and it has produced results substantial enough that his machine has been granted a patent! I don't mean that the machine itself is patented, I mean that the machine actually invented something that was worthy of a patent! It was the first non-human to ever be granted a patent. I have to agree that GA's are excellent at designing new inventions. Better fan blades on up. Problem though is they are not inherently confidence driven which is useful in modeling "emergent" biological processes that work the same at each level, which is a new area of science. It is easy enough to adapt a GA to work the same way but then it's being used for a new purpose that has no natural selection or fitness it is how confident the entity in in its actions. On the other side of the equation I don't think that kind of model would be very useful in industry where they want a fan-blade, not a virtual intelligent amoeba sliming around on the screen. Each has its application it's best suited for and required vocabulary. It's great you noticed the similarities. Most completely miss that, even where they claim to be experts in how they work. To also help answer Krams questions this section of the theory helps boil things down to the confidence driven mechanism that is in us that we can feel. Theory predicts that's exactly how our intelligence works too. Same thing but at a greater complexity level. Impossible to make a prediction like this with GA theory. Multicellular Intelligence
Multicellular intelligence as produced by a brain operates with clock cycles that can be detected from the outside by tuning to waves with an electroencephalograph (EEG) machine to observe brain-waves. In humans, lifetime learning is shown by academic test scores and personal accomplishments.
REQUIREMENT #1 of 4 - SOMETHING TO CONTROL
Human intelligence controls a body with muscles. Each muscle has a connection from brain to muscle to control when it will flex, and a second connection back to brain from muscle to sense that it is indeed flexing as commanded. This is seen in the circuit of the computer model by the two bit (opposing muscles) motor on/off Action data (on right) connecting to motors, while on the other side (on left) are two bits added to Sensors to sense that motors are not stalled against a wall or stuck.
External to the human body are feedback circuits such as a thermostat to control room temperature. The intelligent cause that created the intelligent design is the human intelligence that ultimately controls the room temperature, not the thermostat that only makes it easier for its human intelligent designer to control it with an electrical feedback system of that design. When it fails to perform its mechanical function it gets mercilessly ripped out of the wall to be replaced by a new one. That right there shows where the intelligence controlling the room temperature actually is (and it is not the mechanical thermostat), and where the thermostat that is a product of human intelligent design came from.
Academics control using a reward system by "degrees" and "credentials" which often prevents employment to those who did not "make the grade" even where there are self-learners who have more knowledge and experience from learning and experience while growing up. Human intelligence is here again controlling something for it's own benefit. In this case learning, and knowledge itself.
REQUIREMENT #2 of 4 - ADDRESSABLE MEMORY
Brain neurons store memories with synapic junctions (synapse) between their many neural fibers that grow from them to other neurons. Addressing is through tree-like networks that route signals to proper places. At first a synapse has a small space between them forming a non-permanent connection between neurons which makes them a "short term memory". With repetition/practice the synapse is permanently fused together by glue-like molecules to become part of a "long term memory".
REQUIREMENT #3 of 4 - FEEDBACK TO GAUGE FAILURE AND SUCCESS
Confidence changing feedback which gauges failure and success is something we can consciously "feel". At our level we are consciously rewarded by "success" and feel punished by "failure". For that reason games and sports are very popular to achieve the euphoria that accompanies success.
By being able to "feel for others" we can share in the success or failure of another intelligence simply by watching them. We therefore have heroes who succeed and villains who fail us. In human culture this is well expressed as winning over the 1970's "pinball machine" that preceeded home video games and personal computers. The pinball machine had to be fed quarters which in turn kept many teenagers out of spending money, which in turn helped make the impossible dream of being able to control the game for endless replays the ultimate success for many of that generation.
In the musical movie by the Who named "Tommy" is the song/scene "Tommy and the Pinball Machine" where a "deaf dumb and blind kid" that can only do one thing at all (get endless free games) first discovers a pinball machine that was luring him to wander off. After finding his "calling" the world "lights up" around him by that superhuman success of beating a machine that always eventually won having been achieved.
In reality the ability to fully control a pinball machine would not make it function without being plugged into an electrical outlet or can light up the air around them but our human intelligence abstracts things this way because of how the intelligence mechanism inherently seeks to as much as possible control other things and how increased confidence in reward for being able to do so feels good. And after spending much money attempting to control a pinball machine, human intelligence would then spend more money to see a confidence building movie that feels good by showing what that ultimate superhuman control over a controlling machine would look like when abstracted through art. Our intelligence here understands a reality by relating to something that in reality could not possibly happen. What is in human culture is here useful for explaining what is producing human intelligence by these expressions through the images that exist in its art.
Low confidence of repeated failure or being held down by others attempting to control us produces an imprisoning "bad feeling" that we will work very hard to "get free" from. This is expressed by the Tommy movie song "I'm Free" where he breaks through to the other side of the mirror he once endlessly stared into. The running through the world inviting all to join him then through scenes that resemble going back in time to a primordial planet is here an expression of confidence level suddenly greatly increasing.
Humans have such a need to fill memory with knowledge many feel incomplete especially when it comes to the "big questions" like where we came from and in time will go. Some may seek knowledge from history and/or religion. Scientists may try to answer that by searching for new knowledge scientifically, driven to keep taking their intellect and science to new levels. In fact, that powerful need for knowledge is why this theory exists. In human culture the search for knowledge is often expressed as climbing a mountain for the light of knowledge and wisdom as in the movie Tommy (The Who) - See Me, Feel Me - Listening to You (1975) where after following Tommy because of all wanting what he has tragedy forces aside the controlling of a game what once seemed important and Tommy must now run through the flames then on that long journey to the knowledge that all their lives they suffered to discover.
REQUIREMENT #4 of 4 - ABILITY TO TAKE A GUESS
Externally human intelligence can take a random guess aided by flipping a coin or draw from deck of cards. A good guess is based on success and failure of actions in memory for similar environmental conditions being sensed by their "senses". |
|
|
|
|
Logged
|
|
|
|
|
kram1032
|
|
« Reply #12 on: January 13, 2010, 04:36:45 PM » |
|
I wonder, what a nice combination of confidence driven and genetic programming could do... For a full model, you might need both... Btw: I don't see the problem with your statement that the industry wants blades and not amoebiae: After all, you want the 2nd and thus, genetic programming would be valid for that too. Though, if I understood your answer correctly, you say "no" to confidence driven confidence |
|
|
|
|
Logged
|
|
|
|
|
Gary Gaulin
Guest
|
|
« Reply #13 on: January 14, 2010, 02:33:53 AM » |
|
I wonder, what a nice combination of confidence driven and genetic programming could do... For a full model, you might need both... That might be an excellent way to show the similarities and the differences in the techniques. Help answer questions that you have that are hard to explain without such a model. I know I would have to read every word of that one too! In this case need someone into GA's like I'm into the Intelligence Generator loop/circuit, which would abbreviate to IG which for some reason sounds so funny it has me laughing! My first thought was use the GA to provide a variety of critters like the Intelligence Generator and Detector that has the six sided ring of neurons for a sense of awareness of what is around itself:  http://intelligencegenerator.blogspot.com/It is definitely worth trying to hybridize the two. Good number of educators would want to know about that. The next step after that would be to eliminate the GA from the model, by directly modeling the 4500 gene genome of E.coli K-12 along with all known epigenetic controls. Here's more stats: http://www.ecocyc.org/ECOLI/organism-summary?object=ECOLIHere's excellent info on its "cell sensing" circuit: http://regulondb.ccg.unam.mx/CellSensing.jspEach gene would be a bot. The gene's regulators would be it's environmental inputs. Instead of producing movement of motors/muscles action data would produce RNA that would sent on a given pathway like to the golgi to be used to replicate protein molecule bots. Not have to right away include all cellular processes, can start with just some of the interesting sensing circuits that are known. But without some form of rudimentary intelligence the best I would expect from adding flagella is random motion. Else be stuck in one direction always on or off. http://www.cellsalive.com/animabug.htmThere is excellent info at the start of this paper describing their "swarming" behavior on an agar surface. It also goes into detail about how some of it might work. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1797309/http://www.rowland.harvard.edu/labs/bacteria/movies_swarmecoli.htmlCan start simple with a single memory circuit that has its confidence level set by what E.coli would do when it reaches the end of the raft and in between, and in response to chemical environment swimming through where here each possible condition like 0=dry, 1=agar/gel, 2=slippery, 3=liquid (two bits) is added to the memory addressing, so it knows what it is in. From there it can be made to replicate. After that we better know what is needed to produce variation, but that's here more of a molecular reaction from the behavior of the gene-bots having changed. Are then ready to look again at what is here a "learning curve" of the combined bots in the cell and cell types that have emerged at that point in time: Cambrian Explosion
The computer model displays detailed graphs of electronic intelligence that are available on the screen. Molecular intelligence leaves behind fossils and for survivors of time their genetic memories can be phylogenetically read. It's therefore possible to compare the two datasets, with it here predicted that the near-exponential growth curve of the electronic intelligence also exists in fossil and phylogenetic data.
Trilobites (extinct arthropods) are known to have had well developed crystal lens compound eyes of modern insects. This level of complexity was reached approximately half way through the sudden proliferation of multicellular organisms some 530 million years ago. Trilobite morphology alone requires us to account for a genomic mechanism capable of this rate of information increase. We will here graph the information increase of beta class intelligence then look for correlation with fossil and phylogenetic evidence to determine whether this could be the cause.
The intelligence will be kept busy going from feeder to feeder while we monitor memory (information) increase with time. With its relatively static yet challenging environment this would represent typical information increase in an unexploted niche.
A blue line shows number of highest confidence 3 memories. This is a measure of the most successful responses to environment, a measure of overall "fitness".
The middle two lines show number of lesser confidence 1 and 2 memories, a measure of relative uncertainty.
The black line shows total memory, a measure of relative genome size.
There is at first a very rapid information increase. The number of confidence 3 memories in comarison to total can suggest a relatively large genome size in comparison to overall usefulness.
We can now predict at least two major events. The first when molecular intelligence emerged which rapidly proliferated cells. Then a second when eukaryote cellular intelligence emerged to rapidly proliferate multicellular organisms. This two event possibility is shown with dotted line in the phylogenetic data shown below.
The very left of the graph shows the first event where the simple prokaryotes appeared virtually at the same time. This is the first DNA molecular intelligence. It can exist as a rudimentary cell or in combination as an organelle for cells that fully achieve the next emergent level of cellular intelligence.
http://www.biomedcentral.com/1471-2148/4/2
The middle event would be expected where molecular intelligence has passed the threshold to become cellular intelligence where the cell itself meets all 4 requirements of any intelligence. There is then social/migration behavior of amoeba (also stem cells) and excellent hunting skills of unicellular "single celled animals" the protozoans.
Catastrophic environmental events would alter the static environment causing extinctions followed by quick recover that retakes control of lost niches. The fossil evidence therefore shows the relative information increase below.
http://www.clas.ufl.edu/users/jmeert/timeline2.jpg
We can conclude that rapid proliferation of multicellular biodiversity is possible where a genome crosses the threshold to meet all four requirements of the next possible emergent level of intelligence. This is by no means the only explanation for the Cambrian Explosion. There would first have to be an oxygen source for animals to exploit, before their emergence would have been possible. Here we show how information increase typical of at least beta class intelligence reasonably accounts for such a rapid proliferation of biodiversity. This again shows how there is a property of intelligence that makes it the kind of thing where once it gets started "Watch out!" because it relatively quickly takes control of everything it can figure out how to take control of, as here likely also happened in the Cambrian Explosion.Anyone want to try throwing all together over the weekend? Or at least part of it? |
|
|
|
« Last Edit: January 14, 2010, 03:16:05 AM by Gary Gaulin »
|
Logged
|
|
|
|
|
kram1032
|
|
« Reply #14 on: January 14, 2010, 08:14:41 PM » |
|
Too bad I'm not good at coding at all You know what? In nature, C-type intelligence developed to B-type intelligence which got to the next type of intelligence (A) by evolutionary processes. It would be interesting to write a confidence driven GA type thing which's actual goal is to get more and more intelligent for aribitary systems so it basically progressively becomes A-intelligent from maybe a C-intelligence base, even. Set the starting conditions for C-intelligence but give a tiny bit of information so it goes into the direction of getting more intelligent. Maybe, a generic memory system would be enough for that without any actual confidence system. In the end, the ideal confidence system would start to form on its own.... - it has to solve the following problems: -finding a way to use memories which at first get used at random. Just a fitness function leads to actual first success. - extending the found solution (which still willl continue to be approximated better and better) to allow generic intelligence - developing new kinds of collective intelligence systems (interspecies communication where species could be a single type of cell) as well as specialized split evolution for parts of cell-groups (shaping bigger bodies which are made from single cells of certain kind, like muscle-fibres as oppsed to nerve-cells) - allow for even manipulating the fittness-function in some nice way which means, there must be a basic fittness-function which tries to extend intelligence and the abillity to solve problems and sub-fittness functions which can be shaped at the species' "own will", so to say, so new problems can be found and solved. (This might actually be a requirement for the one before) It might be hard to give the whole system a direction in a world that starts off as nothing. So I suggest that the system has to learn "turing completeness" and some way of during-runtime-new-problem-input which would be comparable with a catastrophy or new ecological niches to be found in order to give the evolution a new thing to solve. That is not supposed to stop the previous quests but rather just to add a new situation so on one hand the algorithm could become even more intelligent (due to the main fittness-curve) and on the other hand, new species could be derived from already existing ones which approximate problems nicely. So a few things you could throw in to start with would be your atomic simulation or, to start simple, "find an algorithm as simple as possible for addition" and as soon as that's solved, go further to "find one for multiplication" and so on. That way you could rediscover basic maths and soon (those basic cases probably will be solved in a satisfying way rather quickly) more complex problems like, say, an algorithm for the gamma function or for rotating a given point in a vectorfield around another point in the same field (or do so with the whole field). Also an interesting application would be to somehow generalize the dimensions the cells could act in. So, at first they only have a single motor and thus can only move back and forth. At the same time, they can "feel" in the 1D-directions and that's all for now. They first have to figure out how to come to dimension 2 by evolutionary adding motors and organs which provide both information and movement along the new directions. Just let it learn about the existence of higher dimensions and maybe solve the problem of the best Close-packing of spheres in 4D and higher looks like more or less on the fly. One important thing would be that already concluded things can be applied to new topics or combined to problems yet to be found. In case of unicellular organisms, that usually works even over the borders of a species which means that two not really related species sometimes share genomes and lead to a new kind of organism which might (or might not) be way "fitter" than any earlier organism. In case of multicellular organisms, that's a bit harder as the single cells are often highly specialized. Usually, additional information in that case would lead to illnesses or, if the corresponding parts are deactivated (by splicing for instance), nothing but immunity agains the corresponding type of illness. Anyways, to actually allow to share information even between two kinds of "results of certain problems" so, under circumstances, to either find the solution of a totally unrelated problem - maybe unknowingly a problem which actually (in reallity) only is discovered in two or so years from now - or to merge two found subproblems to the solution of the actual problem will be important as well. Just got an idea how confidence could fit into that in a different way. The site posted above mentioned that to find a solution for problems with contrary goals, one often has to set priorities to the fittness function. Rather than desciding this beforehand, just do it confidence drivenly. The factors will slowly converge against the best values which lead to the best/fastest intelligence increase. Also never totally kill any solutions, just maybe lower their confidence so recursion-priorities can be set. "Better" solutions get more attention in order to allow for faster progress but "not so good ones" might just hold that tiny bit that by the above mentioned exchange of information could be the key to not only a better but possibly even the BEST solution (However, in that case, confidence from 1-3 probably isn't enough as the system most likely will try to solve more problems over time...) Puh... I hope, all that isn't too random |
|
|
|
|
Logged
|
|
|
|
|
October 19, 2006, 05:22:58 PM
Dilber MC Theme by HarzeM