Object

Tangibility

Pattern

This is an abstract concept that generalizes the notion of pattern and randomness recognition. A Pattern often emerges from Comparisons. *** Notes to self *** What information would you have to code about a door? Would you have to tell the computer that the door was a portal (a hollow structure that allows passage from one space to another). But this would assume that the computer knows what a portal is. No, only physical properties need to be given to the computer. The computer should work on finding patterns in the physical properties of the objects When a bird chirps, a few things can be detected: the frequency, duration, the amplitude, etc. of the sound, and the direction from which the sound was emitted. A computer could be told to regard background anomolies as indications of an object. For example, if on a quiet day outside there was suddenly a bird chirp, the amplitude and frequency detected by the hearing system would suddely register large changes. This would tell the computer: "hey, there's an object nearby." From here, the computer would create a somewhat abstract template object out of the existing data that was present during the anomaly, and then compare this template to all of it's known objects (we will see that if properties are grouped properly, comparisons can be made *very* quickly.) It would find a set of potential matches, assigning a probability to each one (via some sort pre-determined algorithm). It could then tuck away the each object (i.e. data from an anomoly)... The computer treats an anomaly as an object. Not as an indication of an object, but as an object. At first, the computer would be recording *many* anomolies and creating vast numbers of objects (unnamed) that were identified solely by their properties. Now, the computer should have a built in pattern finder that goes and looks at each property (note: property, not object) and then goes to see what objects have that property. For example, the pattern matcher could look at the z-index value for all objects... When the computer looks to get the information on a newly detected object and find out what it probably is, it will have to compare the properties of the new object with the properties of all known objects. Thus, if it has some quick The computer must have a temporal pattern matching device. For example, if it detects lots of birds being present some times of the year, but hardly any at other times of the year, it shouldn't think about something like "average number of birds per year", but should just compare the fluctuation of the short-term populations. This is how to identify a new object; in this case seasons! Same with something like temperature. The computer shouldn't just continuously record temperature, and build up the notion of an average temperature of the object. Instead, it should be able to group all data on time intervals! For example, you could have the data be grouped by intervals. Like, data would be grouped by 4-second intevals, 16-second intervals, etc, and analysis would be done on the average values for the object over these intervals. This would get the "low-frequency" objects (i.e. those objects that cause a very slow change). /** Some data is sent to the computer **/ Object o = new Object(Properties p) { while (allProperties.hasNext()) { // all properties just be strings String apn = allProperties.next(); if (p.containsProperty(apn) { o.setProperty(apn,p.getProperty(apn).getValues()); } } } Here allProperties is a rather short list, because it contains only top-level property concepts, like geometry, spatiality, UV interaction, etc., ones that are highly grouped and able to cull vast amounts of information quickly. For example if the object is heard but not seen, geometry can be dropped, as well as UV interaction. Some spatiality data may be known; the property data can work it's way down the list like this, culling at each level. Do I have to code something like terrestrial location??? I shouldn't have to explicitly. That should be something deduced by the computer, an object that it constructs from noticing "low-frequency" pattern changes. Suppose you had a robot that had no way of detecting temperature while categorizing it's objects, but then a thermometer was attached and the robot moved it's way around the earth, or managed to find temperature differences simultaneously around the globe. It should be able to recognize patterns in the temperature, something that should look like a bunch of loosely outlined (because of the temperature *gradient*) shapes that are slowly morphing. It could then say, "hey, I have a bunch of new objects here. Lets see how these new objects compare to my existing objects to see if I can recognize a pattern" (the new objects would translate to human "temperature zones"). After recording temperature long enough, the computer would be able to notice a long term (yearly) cycle through temperatures. See, you couldn't immediately tell that birds could be categorized based on the temperature zone in which they lived. The problem is: the temperature zones would be defined by temperature and spatial location only. Looking at locations where you've found all the birds before, you might be able to pick out extremeties like tropical and arctic zones (a pattern would be noted like pelicans were 95% likely to be found in a tropical zone, or penguins were 100% likely to be found in an arctic zone). But for most birds it would not be so clear cut. You may have detected sparrows up North before, but if it's cold in the north now, you will be in error if you say "you are 98% likely to find sparrows in Maine on any given day", because this is not true over the course of a year. It may have been true last summer, but if it's -15 degrees there now there aren't going to be any sparrows. That's why the computer needs to have a very temporarly-aware pattern-matching system. Should I worry about the syntactic differences between "the sound that came from the sparrow", or "the sound that came from vibrations of the sparrow's vocal cords"? No, this is pointless. Even if you *could* get a precisely agreed upon definition for every word (a practical impossibility for humans) the latter does not convey any extra information because all of that information is *implied* by the first statement. You can have this be the case if there are just default places that sound can come from on a sparrow. For instance, if you hear the wings of a sparrow beating, as well as the song of a sparrow, there is no confusion when you say "the sound of wings beating that came from the sparrow", because the sound of wings beating is easily distinguished from that of the song, thus making it implicit that the sound came from the wings and not the mouth. Thus every object should have a sound property associated with it if a sound is heard coming from any part of the object. Afterwards you may more thoroughly investigate the object and end up with a bunch of new objects, and then recognize You know, you can make a computer that can look just like the human brain as far as categorizing objects and keeping a mental map of the objects in one's universe, but the computer doesn't really come to life without *desires*. Desires are trivial though. Ahh. Desires aren't even known. They can just be impulses. Impulses like "Fire: retreat", or "Coconut: eat", or big object making noise coming at me "create odd feeling in me (just a property, heh) and create impulse: run". The pattern recognition would recognize a pattern in the way the "odd feeling", the desire to run, and the large object always were showing up together. It would recognize a pattern in it's own behavior that told it to eat. If it observed other creatures it would notice a similar pattern in their behavior. If it then saw an animal that did not eat, it would notice an anomoly in this animal (it doesn't eat), and consequently if it was seen to die, a flag should raise in the computer's head that makes it investigate a connection between eating and death. Now, if eventually the connection were made that not eating for x days led to a y% chance of death, then the computer would be able to be "threatened" by saying "you won't get any food", because the computer would recognize that it would die if not given food. Now, how would it come to want to avoid death in the first place? Well, it doesn't necessarily have to. Have there been humans who did not fear death? The fear of death is something that is built into us. Patterns are engraved into our brains that help us recognize the threat of death. The foundation is there from birth. Like the chickens who see a hawk shape and go into a frenzy. The human does one better. The threat of death can be totally abstracted by the human brain! The human has the capacity to *imagine* himself in death scenes, and consequently *feel the terror* as if it were really happening. This can happen through a story or just as an independent thought. (Quite similar to dreaming, now that I think about it). And the ones that feel the terror at a movie or during a story are the ones who most fully imagine themselves to be in it. Is it flying to the warm, or from the cold that motivates a bird to migrate? Or do both play roles at different times; and sometimes is it a reaction to anything but instinct (like, the bird doesn't feel pain or pleasure, it just suddenly, for no reason it can explain, decides to fly south). ------------------------------------------- **Test Program** Repetition (indicator of pattern) is something that occurs over space and time. A bird chirping is going to register as a single event. A bird flying by is a continuous event. How are we going to store these events? Let's say a bird comes into view. A new object is created that contains the detectable properties of the bird. Now, one thread might start working on finding out whether or not this object fits any known scheme, but another thread needs to be tracking the bird's position. An array of points, perhaps collected at a rate of 100 per second (actual amount should be based on velocity--the higher the angular velocity, the faster snapshots should be taken) should specify the 3D path of the object for as long as it remains visible. From this a spacecurve can be generated, as well as derivitives and so on. Say two birds chirp, one right after the other, but the chirps are different. If these two occur close enough together... Things that occur close together in time should be easier to identify a cause-effect relationship. If the birds chirp 1/2 second apart consistently, I'll probably pick up the relationship very quickly. If they're 20 seconds apart, it'll probably take longer. How can I simulate this on a computer? Well, once I have two instances of this object (inasmuch as I can assume the two instances can be grouped), I can start looking at objects that popped up around the same time. If around both instances of the object I find that another object popped up, then I might assume some relationship between the two objects; like if one is present, I might be able to compute the probability that the other is within some given range of the first. Ah, you don't have to record everything, just the events/objects that stand out; the important thing is to time-stamp them. Will instincts coupled with a supreme ability to recognize patterns produce consciousness? Suppose a robot had the instinct to detect objects and look for patterns among them, and furthermore had the desire to investigate them, doing things like squeezing them, etc. All you need is internal feeling, like pain or pleasure, to be the foundation for consciousness. The first time a robot detects itself, whether through the observation of an arm motion... A pattern: rock hits robot, sensation of pain. rock hits robot, sensation of pain. rock hits other robot, no sensation. The connection: only when a rock hits *this* robot does a sensation of pain also occur. *This* robot, then, is different than every other robot in some fundamental way. *This* robot becomes a very important concept. Looked at from a different angle: Robot arm touches hot metal, sensation of pain, arm recoils, yelp emitted. Other robot arm touches hot metal, no sensation of pain, other robot arm recoils, other robot yelps. The latter will be true for every robot observed except one. That one will be form a separate object: "I". Later, suppose the robot sees a piece of hot metal. The robot knows that touching the metal will result in a sensation of pain. Touching hot metal is associated locally with pain. Yelps are associated locally with pain. Arm recoil is associated locally with pain. Yelps are associated globally with arm recoil. Touching hot metal is associated globally with yelps and arm recoil. So what do you need to have the "I" pop out of this equation? Pattern recognition. All robots behave similarly when touching hot metal. Only one robot feels pain when touching hot metal. That robot I will call "me". "me" resists touching hot metal because it results in the sensation of pain. Suppose I see one of the "others" resisting touching hot metal. What do I attribute the cause to? My (as the robot) line of reasoning: touching hot metal results in pain-- pain results in a yelp and a pulling away of the arm. Pain is only something sensed by "me". --------------------- new day... --------------- Heh heh, ok imaging this refrigerator moving toward the robot. The robot doesn't know to move. The robot detects the objects properties: position, velocity, geomotry. Then the fridge hits the robot, sending this awful feeling. The robot is programmed to react strongly to any "bad feeling" flag, instantly devoting a vast amount of processor time to figuring out what could have been the cause of the "bad feeling". It would start looking at all events that surrounded the "bad feeling" in time. One event would be the refrigerator coming at the robot (represented internally as just an object with a particular velocity relative to the robot's internal reference frame (which would move with the robot's "eyes")). Other events would be someone yelling "look out", and the collision of two objects (the one at location 0,0,0, and the refrigerator). Other unrelated events might also be picked up. A car going by might be registered, but it wouldn't have had anything to do with the pain. The robot, however, just makes a note of *all* of these events, and puts varying degrees of red flag alerts on them (!!! Based on a "probability of being the cause" algorithm--and this algorithm can just be as simple as "how soon before did it occur", because the algorithm can be modifiable through experience!!!). Now, the next time a robot sees any object, it first polls all of it's "red flags" to see if any match the criteria. If the new object doesn't match any "red flag" criteria, then robot processing can go on as normal. But if a red-flag match is made, then the robot needs to be hyper-alert to the situation, because it's looking (in this case) for the cause of the painful feeling. What's happening internally is this: a new object is seen, say a large box. The large box shares some properties with a refrigerator. You have to have the red flags propogate upward on the structure tree. They'll be, like dim at the top if only a few of the subobjects are lit, and bright if a lot are lit. Take spatiality: There will be the [Man, we can really start basic with this thing--I'm envisioning just having this really simple algorithm; no notion of coordinate systems; well, we're talking evolution here... What good are the eyes until you learn how to use the EM information that they are conveying. When you see an object that is coming at you, how do you *know* it is coming at you? I mean, even if you knew that you had to avoid objects coming at you, and you had done it for years by *hearing* the objects around you, how would you make the connection that an object that is gradually taking up more of your field of view and is becoming less fuzzy represents an *object coming at you*. Well, you wouldn't. It would basically have to be hard-coded into the brain: "an object gradually taking up more of your field of view... represents an object coming at you". "Object coming at you" is definitely the top-level concept here. You would only need to be able to react to "object coming at you", and all of the detection methods (eyes, ears, smell, etc) could use the same code when deciding what to do. So, "what to do when something !!! Properties form objects !!! Objects are created out of patterns in properties. If a bunch of objects were detected that had the properties of "living" and "flying", you would form a new object. Heck, "flying" would be consist of fundamental properties itself, primarily "z-index". ------- sub thought ------- If you've ever caught yourself about to rest your hand on a stovetop, and then at the last moment noticed that one of the burners was on, you'll know what it means to react without thinking. Probably you quickly jerked your hand away without any conscious consideration. Now, what exactly went on inside your head? You weren't recoiling from a burner; after all, if you hadn't have noticed that it was on, you wouldn't have recoiled at all. Instead, you were recoiling from something very hot. And "very hot" has a very strong mental connection to "potentially harmful". And this is all that needs to be hardcoded into a brain: If the possibility of harm is forseen, take evasive action. Couple this with "If the possibility of pleasure is forseen, try to achieve it". With these two commands, you can do away with all of the hard-coding of what to do in certain instances. It's just a matter of telling a robot what brings it pleasure and what brings it pain. [Heck, that's all that *really* goes on inside of us--whoa, something like yelling because of pain--what does that serve, why did it originate? It's possible that in the case of avoiding pain, the organism reacts in some general way. Due to genetic variation, different orgamism might be coded to react slightly differently in situations of sudden pain. If one of those organisms had the genes to "vibrate vocal cords", and it served to save the life of that organism, then that organism would pass on the "yell when attacked" genes.] The robot can then make all of the connections between objects, pleasure, and pain. public class Eye { // this file just contains the java code for // transferring UV info to the robot's brain public void run() { sendImageToBrain(); sleep(10); } } So, basically the Eye is just something that sends a picture to the brain 100 times a second. The brain then needs to be able to analyize this picture and detect patterns. Do I need to hard-code object creation. Suppose you turn this thing on for the first time, and the only "senses" it has is a microphone and a touch-sensitive case: The robot hates being touched. Now, suppose a person walks into the room. The microphone is going to record a series of anomalies in the background (like, footsteps or voices). Now suppose it just goes on recording stuff like this for a while. People walking by, people talking, people getting closer (increasing amplitude), people moving away (decreasing amplitude). Every anomaly just gets jotted down and marked with a time stamp. (Hmm, now how does the computer identify an "anomaly". For instance, if a lamp came on and started making a steady humming sound, the computer would need to record some of this, but know to recognize when there hasn't been enough change to warrant further investigation, and quit recording the data. All data should be *processed*, but only anomalous data need be *recorded*. The trick is deciding how closely to analyze the incoming information. For instance, how little time variance in amplitude or frequency does a sound have to have to be regarded as uninteresting. How "steady" does that steady 60Hz actually have to be to warrant only writing one event: that which signaled the sound being heard. Of course you can always increase or decrease your tolerance, the trick is knowing when and how much. If a robot knew to be "alert", it would more carefully scrutinize the incoming data. Like if that 60Hz grew in frequency to 63Hz periodically, and this went unrecorded (or unnoticed, if the hardware tolerance or input data can be varied in precision) in the less-alert mode, then this time around it would be recorded: Something would be recorded like "the frequency of the lamp at time t1 was 60Hz, and at time t2 it was 63Hz. The rate of change of frequency was roughly constant." (How would it determine "constant"? Well, it would initially be examining the frequency very carfully, and getting a time-based plot of the frequency, to which it could be told to try to plot various graphs and determine the best-fitting curve. Ah, but all this comes *far* later, after it has derived oscillations and mathematics... maybe there is a feeling inside you that says "". This is for creatures that can learn. Creatures that can't learn... the reaction must be hard-coded into the brain for every different stimulus, or at least based on generalizations. No, the "don't do it feeling" can be there even witho --------- end sub --------- Radial coordinates are easy to visualize because we have dealt with them implicityly for millions of years. Radial coordinates are the natural coordinate system for objects with radial perception. Cartesian coordinates can be imagined because we are able to see the world outside of ourselves. *** End Notes ***

Properties

Geometry

Spatiality

Position

Velocity

Acceleration

Composition

UV interaction

Sound

Chemical analysis

Thermal properties

Pliability

Firmness

Grouping

Containment

Release

Exchange

Creation

Properties

Comparison

Example