I apologize if it what I wrote appeared to insinuate that you considered handicapped people less intelligent. I think I was trying to point out that what makes us human is very hard to define. When you wrote
I wonder, how well the super-computer did opening the door to leave the building.
I took it as in indication that this is one of the things that make humans better than computers. I agree completely that humans are better, but I wanted to indicate it can be difficult to define what it is about being human that makes us better. In this case I was pointing out opening a door isn't very important in being human.
As I mentioned previously, I had just read McCarthy's book. He makes similar arguments when someones says a computer can never have trait A because they don't do skill B or don't have a sense of C McCarthy points out that many people don't have skill B or sense C but we consider them to have trait A or that trait A turns out not to be an important part of what we consider human.
I'm sorry I came across mean. I meant to be playful. Okay, that's not quite right either. I meant to be mischievous (but in a nice way).
I don't know if strong AI is possible. I just think it would be good to define what it is that AI can't do with some concrete examples. For many people AI seems to be a moving target. As soon as a computer can do something that had been previously consider AI, it's now called "clever programming". (mischievous again, not mean (I hope))
Mike G, how do you define AI? Is it all about clever programming? Is there more to it?
Like I said before, I lean toward Penrose's view on A.I.
I once built a robot that could make its way out of a cluttered area. It turned out that the key component to was a faulty connection to a distance detector. Sometime a little randomness seems intelligent to the outside observer.
@Duane Degn, no problems... I was just messing with you too.
When I think of AI, I think about the movies. The context of this post "Fill the Big Brain" goes into some rather esoteric Big Brain consciousness and dreaming. Problems solving skills compared to the number of neuron connections possible with multiple propellers.
You mentioned a distinction between strong AI and AI. I think the subjects in these threads are of the strong variety.
Clever programming is not enough. A truly intelligent machine would have to transcend its programming to the point that the original programmer won't know how it does what it does. A case in point is the FPGA which was programmed via a genetic algorithm to perform a certain task. Although it learned the task quite well, the program that evolved made no sense from a logic standpoint, relying instead on analog quirks of the particular chip used in the experiment. (Ref: http://www.informatics.sussex.ac.uk/users/adrianth/ices96/paper.ps)
Clever programming is not enough. A truly intelligent machine would have to transcend its programming to the point that the original programmer won't know how it does what it does. A case in point is the FPGA which was programmed via a genetic algorithm to perform a certain task. Although it learned the task quite well, the program that evolved made no sense from a logic standpoint, relying instead on analog quirks of the particular chip used in the experiment. (Ref: http://www.informatics.sussex.ac.uk/users/adrianth/ices96/paper.ps)-Phil
Can a Machine Go Beyond Its Programming?
This is one of the most interesting topics to come along in all of our postings! - the ability of a machine to go beyond its programming in some transcendental aspect. If anyone has any experience with this, please join in..
There is no doubt in my mind that such a feat has occurred numerous times. One example is when I wrote AI programs for the manufacturing company which used assembly robots. A task was assigned to me but it soon became apparent that a solution would globally require over 20,000 unique programs and about 2 years time putting it all together.
We didn't have time or money in the budget. After considerable thought (a couple days worth), an idea was born. Another computer would be interfaced to the robot and it would be taught how to write its own computer programs using AI. It would learn the starting rules of programming, some goals, and it could carry on - on its own - without human intervention, or so I thought.
At the time, it was only my dream to have a machine do its own programming. With a small budget, the project was accepted. It was a big risk. If it failed, I would be fired and my work terminated.
Finally the machine was educated. It took only 20 pages of my compressed code to give the machine some intelligence and teach it all the rules. During the weekend I started it up and it was amazing to see the machine output these remarkable computer programs.
As I recall, the machine spit out programs for 2 weeks, then finished. I had defined its goals and results that it needed to achieve, but when I saw some the of output results with distance-path-curves that the machine had defined, I knew it had gone beyond my programming.
I immediately took this to the "boss" who was amazed as much as me. Thanks to the machine becoming smart and accomplishing its goals, I was able to keep my job and was promoted to company manager.
I kinda lean on Penrose's "The Emperor's New Mind". While it may be a little dated it made an impression during my college days. As I remember it still has valid points regarding the difficulties of AI in a deterministic system.
Mike G, which view is that? I don't have that book to refer to.
Basically, AI is a pipe dream.
Can a Machine Go Beyond Its Programming?
This is one of the most interesting topics to come along in all of our postings! - the ability of a machine to go beyond its programming in some transcendental aspect. If anyone has any experience with this, please join in..
There is no doubt in my mind that such a feat has occurred numerous times. One example is when I wrote AI programs for the manufacturing company which used assembly robots. A task was assigned to me but it soon became apparent that a solution would globally require over 20,000 unique programs and about 2 years time putting it all together.
We didn't have time or money in the budget. After considerable thought (a couple days worth), an idea was born. Another computer would be interfaced to the robot and it would be taught how to write its own computer programs using AI. It would learn the starting rules of programming, some goals, and it could carry on - on its own - without human intervention, or so I thought.
At the time, it was only my dream to have a machine do its own programming. With a small budget, the project was accepted. It was a big risk. If it failed, I would be fired and my work terminated.
Finally the machine was educated. It took only 20 pages of my compressed code to give the machine some intelligence and teach it all the rules. During the weekend I started it up and it was amazing to see the machine output these remarkable computer programs.
As I recall, the machine spit out programs for 2 weeks, then finished. I had defined its goals and results that it needed to achieve, but when I saw some the of output results with distance-path-curves that the machine had defined, I knew it had gone beyond my programming.
I immediately took this to the "boss" who was amazed as much as me. Thanks to the machine becoming smart and accomplishing its goals, I was able to keep my job and was promoted to company manager.
I've written, like many programmers, tons of code that writes code. I can't say I've ever written anything like what you just described. So why is it that when I ask for Big Brain code... there is none? Are you keeping it from the rest of us for some reason?
I can't say I've ever written anything like what you just described.
I described a program that can write programs. In your quote above, you say you've written tons of code that writes code. You seem to contradict yourself.
So why is it that when I ask for Big Brain code... there is none? Are you keeping it from the rest of us for some reason?
I posted two working code versions, one for the Brain Stem communications, which is the nerve center of the Brain that handles motion control, and one for the Brain Base communications, which ensues multiple Propellers, and the schematic versions of each.
This is an open source project. Its development is open source, which means it's up to you to contribute. That means you may write code and contribute something positive. If I knew you had so much programming experience, I would have put you to work long ago!
Actually, yes, there is code kept under wraps, but only temporarily. It's a paper for presentation and publication, about the derivation of new computer algorithms and programs for the Brain and it's under NDA until published. The short wait will be worth it. Stay tuned..
A task was assigned to me but it soon became apparent that a solution would globally require over 20,000 unique programs and about 2 years time putting it all together.
What kind of task? How did you come up with the "20,000" figure?
Another computer would be interfaced to the robot and it would be taught how to write its own computer programs using AI.
Using AI? What kind of AI? What language did it write its programs in?
I had defined its goals and results that it needed to achieve, ...
>What kind of task? How did you come up with the "20,000" figure?
these were Manufacturing tasks, there was a program task for each number
>Using AI? What kind of AI? What language did it write its programs in?
I was studying AI, building robots and robot brains back then too.
Weak AI, at least that's what I thought, the kind where lots of rules & decisions are made and machine choices on how to derive geometrical mathematical pathways and control objects with different jobs requiring precision positioning and repeatability in 4D XYZT and some polar coordinates. I think there was some trig and calculus with that too and a mercator transform and rotational matrices. I was trying out stuff at that time, to see what worked best. I remember at one time the machine had learned how to write cursive English which was something impressive that came from the AI software.
It was 30 years ago...
It was in 6 programming languages, pinning 6502 machine language, entry Applesoft, Card Processor language (sorry, forgot the name as it was some number and alpha character), Japanese assembly line robots language PanaDAC 273 ML derivatives, Wat5, and CNC.
>What were those goals and results?
The goals I hoped for were the successful creation of one program for each task, which needed to handle motion control, hardware, all the controlling software, the automatic nature of the complete system in synchronization, the code for add-on manufacturing robots that I built and added to the system, the communications link (a Japanese thingy that I finally reverse engineered), the Japanese front end electronics interface, the controller card, motors and mechanics and industrial relays, timing, controls, the steppers and stepper boards, pick and place stuff, the solenoid poppers, programming bundles, and the commercial jobs for which it was designed, etc. The results - all the programs worked and tested perfect.
I think the point is we can write programs for intelligent machines and those programs can evolve to become more than their original composition. It might be a good idea, in the case of the Brain, to build up a repertoire of these more basic and fundamental routines for selective and concatenated use.
@Humanoido - it does concern me that my washing machine is smarter (and works harder) than some of my employees!
LOL
Paul
Paul, I hear you! My Korean washing machine also removes the wrinkles from the clothes and will dry them! It probably can do more work, but my Korean operating manual reading is a little rusty.
What really seems scarey is when the old less intelligent (non self balancing) washing machines are off balance and they crawl across the floor while making a loud noise! If your workers start doing that, you'd better take notice.
Mike G, which view is that? I don't have that book to refer to.
Penrose subscribes to Hameroff's theory that the brain is actually a quantum computing system, with quantum processing taking place in the microtubules in neurons. He argues that consciousness requires an actual brain, and can't be achieved with computer hardware. However, no one has actually shown that quantum computing takes place in the brain, and most people disagree with them, largely because the temperature is too high. Here is Hameroff's web site:
The Science Channel ran a program "Search for the Soul" which discussed microtubes and reincarnation. Could be reasonable, but still too early in the field. Maybe in 100 years?
I've been meaning to add this comment for a while in response to statements like "AI is a pipe dream", but I'm having difficulty expressing it correctly, but I'll give it a go...
One issue that this threat brings to mind is that the definition of "intelligence" is not established. The test moves from tasks a human can do, to being self aware, to transcending the original programmer's ability.
Concerning human comparisons, instinct or hardwired behaviors are usually not considered. Animals have pre-programed behaviors to breath, eat, escape destruction, reproduce etc. If these are not "built-in" to a system, it will never be very similar to a human intelligence. But those behaviors are the foundation of "self-aware" (don't let the predator eat ME). So far, most of the "measures" described for testing for true "intelligence" in AI are making apples to oranges comparisons.
IF we consider that ALL programs capture some small segment of the programmer's ability, and repeat those unerringly as-written, them we can say every program is an example of Artificial Intelligence already, and its just a matter of degree.
As in, "what subset of all human behaviors have to simulated so far"?
So self awareness is a function of what sensors and behaviors in support of instinct, tasks a human can do is demonstrated as do-able, and transcending the original programmers ability is already demonstrated, if we include "doing the same task over and over a million times with no deviation".
So the "lack of progress" on AI can be attributed to a bad set of definitions of what AI means.
I actually never heard of many programmers writing tons of code that writes code. Could you please show me tons of links since it's so common?
These days a lot of this tedious work is taken care of. Take a look at .NET's Entity framework, ASP's MVC, Web Matrix, Ruby on Rails, any Visual programming language wires up the GUI delegates and events.
I described a program that can write programs. In your quote above, you say you've written tons of code that writes code. You seem to contradict yourself.
I'm not sure how that's contradictory. I'm merely stating that what I've done is no where close to the complexity you described. A common thing to do about 10 years ago was to enumerate a database to create all your object stubs. It saves hours of work and tons of typing.
This is an open source project. Its development is open source, which means it's up to you to contribute. That means you may write code and contribute something positive. If I knew you had so much programming experience, I would have put you to work long ago!
I'm just a hack.
@prof_braino, I did not mean to offend just state succinctly. I do agree that AI depends on your definition. To me AI is what you see in the movies.
Well, there you go. What you see in the movies is a cheap McGuffin used to move the plot along. This has nothing to do with computers.
If one can't state what one wants (and in a manner where it can be objectively determined if it been achieved or not) one will not get it.
Now, if one wanted to take what we see in the movies and make an actual functional definition of a system we want to construct, very quickly we run into contradictions.
Each contradiction must be resolved if one is to make a set of useable functional requirements. (Did that, didn't write it down, stopped after the following...)
Once the problem is decomposed into its constituent sub-problems (and so on), we find that a large part of the lowest level issues have been addressed. The issue becomes how to put them back together into a ginormous system that addresses all these low level issues, at the same time, in a coordinated fashion.
While I don't think that Humanoido will accidentally create Skynet (cool as that would be), I think he totally has his head around the basic issue. It doesn't matter if he or any one of us is individually successful, what matters is there is one more monkey at one more typewriter for an unspecified amount of time. Please pass the bananas...
Ummm... I think you just called me a dummy. You're probably right, I don't have the chops for all this AI stuff.
No, I wouldn't call you or anyone (but me) a dummy. Sorry if it came out that way. I do requirements and software process for $$; the problems I typically see are due to weak requirements at small outfits, and process deviations are large outfits. It all boils down to requirements and scope. "Human" does not automatically equal "intelligent" (take my boss, PLEASE). "Failures" in AI are usually attributable to failures to define the goals, requirements and scope. This is my input to the thread. I don't know Humanoido's requirements or goals (aside from "Fill the Big Brain") , or if he has even established any. He doesn't need to, this is only the way I would do. He will probably enjoy success before I do, he's DOING something, I am still "working up to it".
This thread discusses basic research, in my opinion. That totally rocks. In basic research, anything that give a result is a "success". I'm doing my work in a different order, maybe one of us will find something useful. The more possibilities explored, the greater the likehood of an interesting result.
"Human" does not automatically equal "intelligent"
And I would add, a machine that is intended to be intelligent, but fails, is not a failure of the machine but rather a failure of humans. Science fiction has long mirrored the machine's failures as a direct relation to man, most noted with HAL 9000. Yet, the flip side of the coin is a good one, with the most wondrous things that machines can do in supporting mankind and living/working within society.
"Failures" in AI are usually attributable to failures to define the goals, requirements and scope.
This is well seen in defining the fundamental compass of AI, as there are many variances and dissenting inferences and sometimes heated debates. Many are with preconceived ideas that won't budge when new material and knowledge becomes available. Some began with spectacular AI breakthroughs, yet succumbed to the wayside of life in a falling out.
I don't know Humanoido's requirements or goals (aside from "Fill the Big Brain"), or if he has even established any.
Believe it or not, there are goals. Filling the Brain is one. I could say this is a great journey, one in which if we had not taken, had not set forth on a voyage of uncharted Propeller space and machine intelligence and navigated the brain waters, could not lead us to not only new and great adventures, and contemplations of perhaps unique and fascinating discoveries beset within the new frontier, but truly result in one of the greatest learning experiences of all time.
Granted, anything new or different is bound to rattle a few cages, elicit a dispensing of moral renderings of "it can't be done," shouldn't be tried," was "never used like that before," and so on. I am only one person, trying to gain travel partners in the quest to overcome the darkness and reach the new and perhaps more intelligent island of light at the end of the tunnel.
It is within this process that we hope to create a new kind of machine life in brain fromat, befit for all robots who need a brain, enriched with an intelligence and a breakthrough of technology, filled with an enhancement never tried before, and with functions to make the brain a multi-diverse solution.
I believe the onset of this brain is an early model, a beginning, a toy leading to the dawning of a new realm, the beginning of a new work, the rebirth of the artist's canvas and how we paint will make all the difference in the world.. It is hoped it could pave the way for more advanced and better brains and brain technology. It is hoped, its interest will spread like a seed cast upon the international Parallax Forum trade winds. As always, the first steps we take as infants are slow, shaky, and encumbered. However, we rise, we fall, we get up again, and persevere.
This thread discusses basic research, in my opinion. That totally rocks. In basic research, anything that give a result is a "success". I'm doing my work in a different order, maybe one of us will find something useful. The more possibilities explored, the greater the likehood of an interesting result.
The thread attempts to develop a working brain, while including many aspects of related brain research and development, and the very basis, extensions, and tools of machine intelligence. In this quest, we research, design, try new things, try old things in new ways, and if it fits, we keep it, adapt it, mold it, and develop it.
The more possibilities explored, the greater the likehood of an interesting result.
In photography, as we take more and more photos, we increase the likelihood of obtaining that one perfect "breakthrough photo" and collecting many fine results. In the "Fill the Big Brain" thread, we will increase our likelihood of reaching our goals, gain spin-off technologies, and push towards the life of the machine.
I've had some projects that lasted 10 years each. So when setting upon the Humanoid Initiative, to build a full scale working humanoid, it was understood that required parts would be developed over a similar time period. The project design hit a brick wall early on with two things - battery technology was not up to par, and a small powerful brain was lacking.
A decision was made. The idea was to continue HI, and choose one of the two bricks and develop it. The Brain brick was chosen. The brain would be developed while the world could catch up in the development of better battery technology.
I immediately begin developing a model with multiple processors using Parallax Stamps. But the time 2002 rolled around, I traveled to Japan, met with SONY, HONDA and Fujitsu, and gained valuable insights into the requirements of their humanoids.
While running development of the big brain, small brains were needed for quick testing. This led to the small and medium brains. Small brain has a 5 propeller chip board and medium brain has 14 props on a board. The big brain is not yet maxed out with a current 23 boards. Another board is being developed which will add on to medium brain. The plan is to double medium brain in size to around 224 Cogs. Beyond that, medium brain could join big brain for a more massive number of cogs. A tiny brain is also planned, with boards holding one, two, or three props. Another medium brain is in the assembly area and it has sockets for chips with 96 cogs. The advantage of this mid-medium brain is a smaller size. Recently, a big-tiny brain was began with three sockets for 24 cogs.
The first little brain was a small distributed multi-processor brain working in a humanoid robot. It grew to about 5 processors using BS2 OEM kits, and was maxed out early on with motion control. Something more powerful was needed, like a larger array of processors. Other brand chips were available but discounted as they lacked readily available support and information to make relatively easy transitions to working models. I recognized that Parallax was the leader and multi-chip technology would be useful and therefore set upon developing these systems.
This resulted in several machines in the BASIC Stamp Supercomputer series. Each of these explored some new idea and feature leading up to a more powerful brain. The BSS was all about lots of mixed version processors and seeing they could work together. The MSS was the minimal config unit. The TSS that appeared in Robot Magazine was all about minimizing size and creating portability. The TS was a two Stamp version that explored interfacing..
The TCS was a TriCore that exercised perhaps the most tiny rendition of "Tiny AI" and was used for testing SEED. M.O.M. was the Master Offloader Machine that could upgrade any "supercomputer" instantly by adding 10 more processors. AM was an algorithm machine that tested new interface designs.
Finally, the SEED machine was all about the beginning of machine intelligence. With 10 processors in parallel, each processor was a life form - it was self enumerating, deterministic, could listen, talk, converse, remember, recall, had a sense of time, could self evolve a personality, do work (some ended up lazy), nap and dream a little dream on the Debug screen.
SEED had ten life forms inside the machine. This is output from
one life form's 256 Bytes machine code.
Somewhere between then and now, Phil got me started on the Propeller chip. I started learning Spin and building larger machines that could do more. This led to another 20 machines of progressive size, that were torn down and recycled from one to the next, finally culminating in the largest UltraSpark series, going from 2 props on up to 40 props The data from the US40 led to the first 320 cog Brain experiment on Boe-Bot. This went so well (IMO), I immediately began hatching plans for a newer, leaner, more powerful brain.
That in a nutshell brings the story up to date. A lot is left out, but this accountancy is already long enough. So now we have one massive brain development and two smaller ones ensuing development that were built primarily for tests. Small, medium and big brains..
Main Points Only One Program
SEED is unlike previous stamp supercomputer software that had a different program load into each computer. SEED is only one program - a life form. The same program loads into all computers in the supercomputer,
Life Form
The program is born and then it evolves into a life form! The collective supercomputer has ten life forms that are evolving at the same time.
Unique
After evolving, no two life forms are alike.
Life Span
Each life form has a life span of about one minute of activity.
Birth
It is first born based on its predetermined DNA, in this case, declarations and announcement.
Self Enumerating
Next, it looks at itself and determines its own unique enumerating identity. It does this by looking at pin 1 which contains a unique rc circuit.
Self Deterministic
Then SEED self determines its own computer number. It is now self aware and knows its own unique identity and its position in the world.
Memory & Recall
It then exercises its memory by remembering its own characteristics and demonstrates its recall ability.
Napping & Random Dreaming
All of this leads to taking a series of naps followed by pseudo random dreams. The dream are in Vers Libre, a kind of simple word-verse poetry. The nap time is accumulative, with similar rest effects that humans have. In this case, the power is reduced from 1 ma to 25 ua. The pseudo random dream generator is seeded by the unique identity of the lifeform.
Dream Watching
A Dream View Port puts the dreams up on the debug screen for viewing. It includes words, the random binary bit, and the pseudo random byte decimal number.
Sleep & Fuzzy Clock
The life form will now sleep using its fuzzy alarm clock. The approximate sleep time is unique and unlike the other lifeforms because it is based on its own unique characteristic of its computer identification number. The higher the number, the longer it sleeps. Therefore each life form has its own sleeping habits.
Talking to Neighbors
It wants to learn more about its world, so it sends out a message telling who he is and what his identity is to anyone listening.
Listening to Neighbors
Next, he listens for a reply. If he is life form number 1, he will get 9 additional replies. If he is number 5, he will get 5 replies. Finally, life form 10 will think he is alone. It's because each computer sleeps a different length of time. If a life form is sleeping, it will not hear the others talking.
Getting to know Neighbors
Now he gets to know his neighbors on a personal name basis. He memorizes the personal information of each neighbor that is heard. Next, all the neighbors are remembered.
Remembering Neighbors
He memorizes the personal information of each neighbor that is heard. Next, all the neighbors are remembered and recalled.
Thinking & Working
Some thinking/working takes place to recall the personal identity, determine the number of seconds in sleep and how many life forms were heard, plus how many life forms were not heard, and how many dreams were remembered.
End of Life
Finally, the life ends but is not entirely terminated, as it goes into stasis (I am not comfortable about creating life and then causing it to die).
Personalities
Each life form has its own evolved personality. There are several unique features. The personal identity, computer number, its living location, the random dreams it has, and the length of sleep habit. Its sleeping habit determines how friendly or how much of a hermit it will become. It also determines how much information about neighbors it will remember. Some life forms remember a lot of information while others remember very little. Thinking outcome is unique too.
Restrictions
Due to memory restrictions, PLAN Piezo Language is not included with this version, even though PLAN is a spinoff of the SEED software. The entire communications view port was built upon the Debug screen.
Becoming Sick
Because these are ten individual life forms, sickness can result. Below is a synopsis.
' PERSONIFIED MALADIES
' --------------------
' When you download a program into the BASIC Stamp 1, it is stored in the
' EEPROM starting at the highest address (255) and working towards the
' lowest address. Most programs don't use the entire EEPROM, so the lower
' portion is available for other uses. This portion is used for long term
' memories. As a result, like a real human person, this artificial life
' form is capable of succumbing to sickness and malady.
' PSYCHOTIC BREAKDOWN
' -------------------
' AI memory begins at memory location 0 and works upward. If
' the memories become too much, they will fall upon code and a psychotic
' action will take place, resulting in a breakdown as the main program
' will be overwritten.
' FITFUL SLEEP
' ------------
' We don't know how effective sleep really is for the AI supercomputer
' as its sleep is constantly interrupted every 2.3 seconds.
' The Basic Stamp 1s output pins will toggle briefly when using SLEEP,
' NAP or END. Inside the BASIC Stamp's interpreter chip is a watchdog
' timer whose main purpose is to reset the interpreter chip if, for some
' reason, it should stop functioning properly. The SLEEP and NAP commands
' also utilize the watchdog timer to periodically, every 2.3 seconds
' "wake-up" the BASIC Stamp from its low-power mode. Upon reset, the
' I/O pins are set to inputs for approximately 18 ms before returning
' to their previous directions and states. If you have an output pin set
' to a logical 1 state (+5V) and you use the SLEEP command, every 2.3 seconds
' during sleep mode that I/O pin will switch to an input for 18 ms causing
' a momentary signal loss. This "power glitch" is easily viewable with an
' LED and a 470 ohm resister tied to an I/O pin and switched on just before
' entering sleep mode. In many cases this problem can be remedied by tying
' a pull-up or pull-down resistor to the I/O pin in question to provide
' a constant source of power should the I/O pin change directions. Allowing
' a PBASIC program to end naturally, or using the END command, will exhibit
' the same "power glitch" behavior because the interpreter chip enters
' a low-power state.
'
' SUDDEN DEATH, BRAIN AMNESIA, NONRESUSCIATING SITUATION
' ------------------------------------------------------
' Life Terminates after power off, or by running another program. If a
' 2nd program is run (or the same program run a 2nd time), it will put all
' 0s into the EEPROM, thus destroying any memories and the original identity.
' It will be useless trying to resuscitate and bring back the original
' memories.
'
' ALTERNATE IDENTITY SYNDROME
' ---------------------------
' The stamp is reset causing the program to run again, and the birthing
' process takes place and a different identity emerges.
'
' STROKE
' ------
' The stamp is reset or repowered. It tries to communicate with other
' stamps but it is not in sync and the serial command hangs.
' It is unable to reply or do anything at this point.
'
' STUTTERING
' ----------
' Battery low causing short term repeating system resets
Beginning wiring - At the base of the Big Brain is a second brain built on a solderless breadboard with the same number of cogs.
The second brain is designed for testing, investigating the
transmission of neural substance, and testing DIP Propellers.
______________________________________________________
Many hours are now invested in the creation of a second brain which duplicates 160 cogs of the first. It is important to test the neural matter injectors with the largest number of cogs, at full capacity. Since expansion was in the schedule, it was decided to created an expansive unit first. All add-on 20 Propellers are working and provide a form of injecting neural substance.
This Brain Dupe will handle over 20,000 neural elements though more tests are required to see if the propagation can use existing interfacing. After the Propellers are booted, all of its pins become available for use and recycle.
Unexpectedly, there was a 70% failure of props to handle the rc loading. Using a technique of select, test and replace, the full quota of 20 was established, however at the end of the phase, 14 non-usable props were counted. This phenomena is being investigated.
Connecting the two twin brains, if that happens, will bring up the neural quota to over 40,000. The challenge will be managing neural matter.
Comments
LOL
Paul
I apologize if it what I wrote appeared to insinuate that you considered handicapped people less intelligent. I think I was trying to point out that what makes us human is very hard to define. When you wrote
I took it as in indication that this is one of the things that make humans better than computers. I agree completely that humans are better, but I wanted to indicate it can be difficult to define what it is about being human that makes us better. In this case I was pointing out opening a door isn't very important in being human.
As I mentioned previously, I had just read McCarthy's book. He makes similar arguments when someones says a computer can never have trait A because they don't do skill B or don't have a sense of C McCarthy points out that many people don't have skill B or sense C but we consider them to have trait A or that trait A turns out not to be an important part of what we consider human.
I'm sorry I came across mean. I meant to be playful. Okay, that's not quite right either. I meant to be mischievous (but in a nice way).
I don't know if strong AI is possible. I just think it would be good to define what it is that AI can't do with some concrete examples. For many people AI seems to be a moving target. As soon as a computer can do something that had been previously consider AI, it's now called "clever programming". (mischievous again, not mean (I hope))
Like I said before, I lean toward Penrose's view on A.I.
I once built a robot that could make its way out of a cluttered area. It turned out that the key component to was a faulty connection to a distance detector. Sometime a little randomness seems intelligent to the outside observer.
When I think of AI, I think about the movies. The context of this post "Fill the Big Brain" goes into some rather esoteric Big Brain consciousness and dreaming. Problems solving skills compared to the number of neuron connections possible with multiple propellers.
You mentioned a distinction between strong AI and AI. I think the subjects in these threads are of the strong variety.
-Phil
Can a Machine Go Beyond Its Programming?
This is one of the most interesting topics to come along in all of our postings! - the ability of a machine to go beyond its programming in some transcendental aspect. If anyone has any experience with this, please join in..
There is no doubt in my mind that such a feat has occurred numerous times. One example is when I wrote AI programs for the manufacturing company which used assembly robots. A task was assigned to me but it soon became apparent that a solution would globally require over 20,000 unique programs and about 2 years time putting it all together.
We didn't have time or money in the budget. After considerable thought (a couple days worth), an idea was born. Another computer would be interfaced to the robot and it would be taught how to write its own computer programs using AI. It would learn the starting rules of programming, some goals, and it could carry on - on its own - without human intervention, or so I thought.
At the time, it was only my dream to have a machine do its own programming. With a small budget, the project was accepted. It was a big risk. If it failed, I would be fired and my work terminated.
Finally the machine was educated. It took only 20 pages of my compressed code to give the machine some intelligence and teach it all the rules. During the weekend I started it up and it was amazing to see the machine output these remarkable computer programs.
As I recall, the machine spit out programs for 2 weeks, then finished. I had defined its goals and results that it needed to achieve, but when I saw some the of output results with distance-path-curves that the machine had defined, I knew it had gone beyond my programming.
I immediately took this to the "boss" who was amazed as much as me. Thanks to the machine becoming smart and accomplishing its goals, I was able to keep my job and was promoted to company manager.
Mike G, which view is that? I don't have that book to refer to.
Basically, AI is a pipe dream.
I've written, like many programmers, tons of code that writes code. I can't say I've ever written anything like what you just described. So why is it that when I ask for Big Brain code... there is none? Are you keeping it from the rest of us for some reason?
-Phil
I described a program that can write programs. In your quote above, you say you've written tons of code that writes code. You seem to contradict yourself.
I posted two working code versions, one for the Brain Stem communications, which is the nerve center of the Brain that handles motion control, and one for the Brain Base communications, which ensues multiple Propellers, and the schematic versions of each.
This is an open source project. Its development is open source, which means it's up to you to contribute. That means you may write code and contribute something positive. If I knew you had so much programming experience, I would have put you to work long ago!
Actually, yes, there is code kept under wraps, but only temporarily. It's a paper for presentation and publication, about the derivation of new computer algorithms and programs for the Brain and it's under NDA until published. The short wait will be worth it. Stay tuned..
What kind of task? How did you come up with the "20,000" figure?
Using AI? What kind of AI? What language did it write its programs in?
What were those goals and results?
-Phil
these were Manufacturing tasks, there was a program task for each number
>Using AI? What kind of AI? What language did it write its programs in?
I was studying AI, building robots and robot brains back then too.
Weak AI, at least that's what I thought, the kind where lots of rules & decisions are made and machine choices on how to derive geometrical mathematical pathways and control objects with different jobs requiring precision positioning and repeatability in 4D XYZT and some polar coordinates. I think there was some trig and calculus with that too and a mercator transform and rotational matrices. I was trying out stuff at that time, to see what worked best. I remember at one time the machine had learned how to write cursive English which was something impressive that came from the AI software.
It was 30 years ago...
It was in 6 programming languages, pinning 6502 machine language, entry Applesoft, Card Processor language (sorry, forgot the name as it was some number and alpha character), Japanese assembly line robots language PanaDAC 273 ML derivatives, Wat5, and CNC.
>What were those goals and results?
The goals I hoped for were the successful creation of one program for each task, which needed to handle motion control, hardware, all the controlling software, the automatic nature of the complete system in synchronization, the code for add-on manufacturing robots that I built and added to the system, the communications link (a Japanese thingy that I finally reverse engineered), the Japanese front end electronics interface, the controller card, motors and mechanics and industrial relays, timing, controls, the steppers and stepper boards, pick and place stuff, the solenoid poppers, programming bundles, and the commercial jobs for which it was designed, etc. The results - all the programs worked and tested perfect.
That's all I remember..
I think the point is we can write programs for intelligent machines and those programs can evolve to become more than their original composition. It might be a good idea, in the case of the Brain, to build up a repertoire of these more basic and fundamental routines for selective and concatenated use.
What really seems scarey is when the old less intelligent (non self balancing) washing machines are off balance and they crawl across the floor while making a loud noise! If your workers start doing that, you'd better take notice.
Penrose subscribes to Hameroff's theory that the brain is actually a quantum computing system, with quantum processing taking place in the microtubules in neurons. He argues that consciousness requires an actual brain, and can't be achieved with computer hardware. However, no one has actually shown that quantum computing takes place in the brain, and most people disagree with them, largely because the temperature is too high. Here is Hameroff's web site:
http://www.quantumconsciousness.org/
Here are some links to his work with Penrose:
http://www.quantumconsciousness.org/publications.html#penrose
He seems to be getting nuttier as he gets older.
I resemble that remark!
The Science Channel ran a program "Search for the Soul" which discussed microtubes and reincarnation. Could be reasonable, but still too early in the field. Maybe in 100 years?
I've been meaning to add this comment for a while in response to statements like "AI is a pipe dream", but I'm having difficulty expressing it correctly, but I'll give it a go...
One issue that this threat brings to mind is that the definition of "intelligence" is not established. The test moves from tasks a human can do, to being self aware, to transcending the original programmer's ability.
Concerning human comparisons, instinct or hardwired behaviors are usually not considered. Animals have pre-programed behaviors to breath, eat, escape destruction, reproduce etc. If these are not "built-in" to a system, it will never be very similar to a human intelligence. But those behaviors are the foundation of "self-aware" (don't let the predator eat ME). So far, most of the "measures" described for testing for true "intelligence" in AI are making apples to oranges comparisons.
IF we consider that ALL programs capture some small segment of the programmer's ability, and repeat those unerringly as-written, them we can say every program is an example of Artificial Intelligence already, and its just a matter of degree.
As in, "what subset of all human behaviors have to simulated so far"?
So self awareness is a function of what sensors and behaviors in support of instinct, tasks a human can do is demonstrated as do-able, and transcending the original programmers ability is already demonstrated, if we include "doing the same task over and over a million times with no deviation".
So the "lack of progress" on AI can be attributed to a bad set of definitions of what AI means.
I'm not sure how that's contradictory. I'm merely stating that what I've done is no where close to the complexity you described. A common thing to do about 10 years ago was to enumerate a database to create all your object stubs. It saves hours of work and tons of typing.
I'm just a hack.
@prof_braino, I did not mean to offend just state succinctly. I do agree that AI depends on your definition. To me AI is what you see in the movies.
Well, there you go. What you see in the movies is a cheap McGuffin used to move the plot along. This has nothing to do with computers.
If one can't state what one wants (and in a manner where it can be objectively determined if it been achieved or not) one will not get it.
Now, if one wanted to take what we see in the movies and make an actual functional definition of a system we want to construct, very quickly we run into contradictions.
Each contradiction must be resolved if one is to make a set of useable functional requirements. (Did that, didn't write it down, stopped after the following...)
Once the problem is decomposed into its constituent sub-problems (and so on), we find that a large part of the lowest level issues have been addressed. The issue becomes how to put them back together into a ginormous system that addresses all these low level issues, at the same time, in a coordinated fashion.
While I don't think that Humanoido will accidentally create Skynet (cool as that would be), I think he totally has his head around the basic issue. It doesn't matter if he or any one of us is individually successful, what matters is there is one more monkey at one more typewriter for an unspecified amount of time. Please pass the bananas...
No, I wouldn't call you or anyone (but me) a dummy. Sorry if it came out that way. I do requirements and software process for $$; the problems I typically see are due to weak requirements at small outfits, and process deviations are large outfits. It all boils down to requirements and scope. "Human" does not automatically equal "intelligent" (take my boss, PLEASE). "Failures" in AI are usually attributable to failures to define the goals, requirements and scope. This is my input to the thread. I don't know Humanoido's requirements or goals (aside from "Fill the Big Brain") , or if he has even established any. He doesn't need to, this is only the way I would do. He will probably enjoy success before I do, he's DOING something, I am still "working up to it".
This thread discusses basic research, in my opinion. That totally rocks. In basic research, anything that give a result is a "success". I'm doing my work in a different order, maybe one of us will find something useful. The more possibilities explored, the greater the likehood of an interesting result.
The uncharted space of brain technology
And I would add, a machine that is intended to be intelligent, but fails, is not a failure of the machine but rather a failure of humans. Science fiction has long mirrored the machine's failures as a direct relation to man, most noted with HAL 9000. Yet, the flip side of the coin is a good one, with the most wondrous things that machines can do in supporting mankind and living/working within society.
This is well seen in defining the fundamental compass of AI, as there are many variances and dissenting inferences and sometimes heated debates. Many are with preconceived ideas that won't budge when new material and knowledge becomes available. Some began with spectacular AI breakthroughs, yet succumbed to the wayside of life in a falling out.
Believe it or not, there are goals. Filling the Brain is one. I could say this is a great journey, one in which if we had not taken, had not set forth on a voyage of uncharted Propeller space and machine intelligence and navigated the brain waters, could not lead us to not only new and great adventures, and contemplations of perhaps unique and fascinating discoveries beset within the new frontier, but truly result in one of the greatest learning experiences of all time.
Granted, anything new or different is bound to rattle a few cages, elicit a dispensing of moral renderings of "it can't be done," shouldn't be tried," was "never used like that before," and so on. I am only one person, trying to gain travel partners in the quest to overcome the darkness and reach the new and perhaps more intelligent island of light at the end of the tunnel.
It is within this process that we hope to create a new kind of machine life in brain fromat, befit for all robots who need a brain, enriched with an intelligence and a breakthrough of technology, filled with an enhancement never tried before, and with functions to make the brain a multi-diverse solution.
I believe the onset of this brain is an early model, a beginning, a toy leading to the dawning of a new realm, the beginning of a new work, the rebirth of the artist's canvas and how we paint will make all the difference in the world.. It is hoped it could pave the way for more advanced and better brains and brain technology. It is hoped, its interest will spread like a seed cast upon the international Parallax Forum trade winds. As always, the first steps we take as infants are slow, shaky, and encumbered. However, we rise, we fall, we get up again, and persevere.
The thread attempts to develop a working brain, while including many aspects of related brain research and development, and the very basis, extensions, and tools of machine intelligence. In this quest, we research, design, try new things, try old things in new ways, and if it fits, we keep it, adapt it, mold it, and develop it.
In photography, as we take more and more photos, we increase the likelihood of obtaining that one perfect "breakthrough photo" and collecting many fine results. In the "Fill the Big Brain" thread, we will increase our likelihood of reaching our goals, gain spin-off technologies, and push towards the life of the machine.
Part 1
I've had some projects that lasted 10 years each. So when setting upon the Humanoid Initiative, to build a full scale working humanoid, it was understood that required parts would be developed over a similar time period. The project design hit a brick wall early on with two things - battery technology was not up to par, and a small powerful brain was lacking.
A decision was made. The idea was to continue HI, and choose one of the two bricks and develop it. The Brain brick was chosen. The brain would be developed while the world could catch up in the development of better battery technology.
I immediately begin developing a model with multiple processors using Parallax Stamps. But the time 2002 rolled around, I traveled to Japan, met with SONY, HONDA and Fujitsu, and gained valuable insights into the requirements of their humanoids.
Small, medium and big
While running development of the big brain, small brains were needed for quick testing. This led to the small and medium brains. Small brain has a 5 propeller chip board and medium brain has 14 props on a board. The big brain is not yet maxed out with a current 23 boards. Another board is being developed which will add on to medium brain. The plan is to double medium brain in size to around 224 Cogs. Beyond that, medium brain could join big brain for a more massive number of cogs. A tiny brain is also planned, with boards holding one, two, or three props. Another medium brain is in the assembly area and it has sockets for chips with 96 cogs. The advantage of this mid-medium brain is a smaller size. Recently, a big-tiny brain was began with three sockets for 24 cogs.
Part 2
The first little brain was a small distributed multi-processor brain working in a humanoid robot. It grew to about 5 processors using BS2 OEM kits, and was maxed out early on with motion control. Something more powerful was needed, like a larger array of processors. Other brand chips were available but discounted as they lacked readily available support and information to make relatively easy transitions to working models. I recognized that Parallax was the leader and multi-chip technology would be useful and therefore set upon developing these systems.
This resulted in several machines in the BASIC Stamp Supercomputer series. Each of these explored some new idea and feature leading up to a more powerful brain. The BSS was all about lots of mixed version processors and seeing they could work together. The MSS was the minimal config unit. The TSS that appeared in Robot Magazine was all about minimizing size and creating portability. The TS was a two Stamp version that explored interfacing..
Part 3
The TCS was a TriCore that exercised perhaps the most tiny rendition of "Tiny AI" and was used for testing SEED. M.O.M. was the Master Offloader Machine that could upgrade any "supercomputer" instantly by adding 10 more processors. AM was an algorithm machine that tested new interface designs.
Finally, the SEED machine was all about the beginning of machine intelligence. With 10 processors in parallel, each processor was a life form - it was self enumerating, deterministic, could listen, talk, converse, remember, recall, had a sense of time, could self evolve a personality, do work (some ended up lazy), nap and dream a little dream on the Debug screen.
SEED had ten life forms inside the machine. This is output from
one life form's 256 Bytes machine code.
http://forums.parallax.com/showthread.php?113829-BASIC-STAMP-SEED-Supercomputer&p=817126
Part 4
Somewhere between then and now, Phil got me started on the Propeller chip. I started learning Spin and building larger machines that could do more. This led to another 20 machines of progressive size, that were torn down and recycled from one to the next, finally culminating in the largest UltraSpark series, going from 2 props on up to 40 props The data from the US40 led to the first 320 cog Brain experiment on Boe-Bot. This went so well (IMO), I immediately began hatching plans for a newer, leaner, more powerful brain.
That in a nutshell brings the story up to date. A lot is left out, but this accountancy is already long enough. So now we have one massive brain development and two smaller ones ensuing development that were built primarily for tests. Small, medium and big brains..
Main Points
Only One Program
SEED is unlike previous stamp supercomputer software that had a different program load into each computer. SEED is only one program - a life form. The same program loads into all computers in the supercomputer,
Life Form
The program is born and then it evolves into a life form! The collective supercomputer has ten life forms that are evolving at the same time.
Unique
After evolving, no two life forms are alike.
Life Span
Each life form has a life span of about one minute of activity.
Birth
It is first born based on its predetermined DNA, in this case, declarations and announcement.
Self Enumerating
Next, it looks at itself and determines its own unique enumerating identity. It does this by looking at pin 1 which contains a unique rc circuit.
Self Deterministic
Then SEED self determines its own computer number. It is now self aware and knows its own unique identity and its position in the world.
Memory & Recall
It then exercises its memory by remembering its own characteristics and demonstrates its recall ability.
Napping & Random Dreaming
All of this leads to taking a series of naps followed by pseudo random dreams. The dream are in Vers Libre, a kind of simple word-verse poetry. The nap time is accumulative, with similar rest effects that humans have. In this case, the power is reduced from 1 ma to 25 ua. The pseudo random dream generator is seeded by the unique identity of the lifeform.
Dream Watching
A Dream View Port puts the dreams up on the debug screen for viewing. It includes words, the random binary bit, and the pseudo random byte decimal number.
Sleep & Fuzzy Clock
The life form will now sleep using its fuzzy alarm clock. The approximate sleep time is unique and unlike the other lifeforms because it is based on its own unique characteristic of its computer identification number. The higher the number, the longer it sleeps. Therefore each life form has its own sleeping habits.
Talking to Neighbors
It wants to learn more about its world, so it sends out a message telling who he is and what his identity is to anyone listening.
Listening to Neighbors
Next, he listens for a reply. If he is life form number 1, he will get 9 additional replies. If he is number 5, he will get 5 replies. Finally, life form 10 will think he is alone. It's because each computer sleeps a different length of time. If a life form is sleeping, it will not hear the others talking.
Getting to know Neighbors
Now he gets to know his neighbors on a personal name basis. He memorizes the personal information of each neighbor that is heard. Next, all the neighbors are remembered.
Remembering Neighbors
He memorizes the personal information of each neighbor that is heard. Next, all the neighbors are remembered and recalled.
Thinking & Working
Some thinking/working takes place to recall the personal identity, determine the number of seconds in sleep and how many life forms were heard, plus how many life forms were not heard, and how many dreams were remembered.
End of Life
Finally, the life ends but is not entirely terminated, as it goes into stasis (I am not comfortable about creating life and then causing it to die).
Personalities
Each life form has its own evolved personality. There are several unique features. The personal identity, computer number, its living location, the random dreams it has, and the length of sleep habit. Its sleeping habit determines how friendly or how much of a hermit it will become. It also determines how much information about neighbors it will remember. Some life forms remember a lot of information while others remember very little. Thinking outcome is unique too.
Restrictions
Due to memory restrictions, PLAN Piezo Language is not included with this version, even though PLAN is a spinoff of the SEED software. The entire communications view port was built upon the Debug screen.
Becoming Sick
Because these are ten individual life forms, sickness can result. Below is a synopsis.
2nd brain twins the first
Beginning wiring - At the base of the Big Brain is a second brain built on a solderless breadboard with the same number of cogs.
The second brain is designed for testing, investigating the
transmission of neural substance, and testing DIP Propellers.
______________________________________________________
Many hours are now invested in the creation of a second brain which duplicates 160 cogs of the first. It is important to test the neural matter injectors with the largest number of cogs, at full capacity. Since expansion was in the schedule, it was decided to created an expansive unit first. All add-on 20 Propellers are working and provide a form of injecting neural substance.
This Brain Dupe will handle over 20,000 neural elements though more tests are required to see if the propagation can use existing interfacing. After the Propellers are booted, all of its pins become available for use and recycle.
Unexpectedly, there was a 70% failure of props to handle the rc loading. Using a technique of select, test and replace, the full quota of 20 was established, however at the end of the phase, 14 non-usable props were counted. This phenomena is being investigated.
Connecting the two twin brains, if that happens, will bring up the neural quota to over 40,000. The challenge will be managing neural matter.