Home | Videos | Contact Us   




                         << content                                      Chapter 5

Searching for data in the robot’s brain

The search function for the brain is very basic and universal.  The data in memory is stored in terms of association, however, each data can be same, similar or different.  An entire network of data can be created for one object called an object composite.  The object composite structures data in the network in a hierarchical manner so that important data is prominent and minor data is less prominent.  The importance of data is determined by many factors and said data can be in any four data types:  5 sense objects, hidden objects, activated element objects and pattern objects.  Usually, words and sentences are prominent data and long pathways and detailed images are minor data.  The search function simply identifies object composites in memory (also called floaters) and extract information hierarchically, whereby prominent data is searched first before searching the minor data. 

Intelligent pathways in memory have the necessary instructions to search in memory for data.  The instructions in the search pathways are not learned by teachers, but they are based on patterns found from lessons taught by teachers.  The teachers teach the robot a lesson and the robot’s brain found patterns between the lesson and the retrieval and extraction of data in memory.  



FIG. 16 is a simple example illustrating how patterns are found between a lesson taught by teachers and the search instructions.  The teacher asks the robot what it did 2 hours ago.  The robot searches its long-term memory for data related to actions it took.  The robot gives the answer:  “I was drawing a picture”.  If you compare this example to many similar examples you will notice that the time of where in the long-term memory to extract information is same or similar.  Within pathway56, the search instruction will search the long-term memory exactly 2 hours ago.  It does not have to search the robot’s memory, only long-term memory.  By comparing many similar examples, the search pattern found out that it needs to extract data that happened approximately 2 hours ago. 

If the teacher asked another question such as:  what did you do 1 day ago or 4 days ago or 1 year ago, the universal pathway will be able to understand where in the long-term memory to search to find information.  Instead of searching all data in long-term memory or all data in the brain, the search pathways have patterns to search for data in a heuristic way. 

The search pathways are learned by lessons taught in school and through pattern recognitions. 

Questions such as:  what was I doing earlier?, would require the robot to quickly scan the long-term memory closest to the current state to look for information.  This question only relates to what the robot did today, possible a few hours ago.  The search function will not search for data that happened yesterday or 1 month ago.  Also, the search function is searching for prominent events that got the robot’s attention.  Simple events that the robot didn’t pay attention to will be searched last (or ignored).

When the robot is studying and is preparing for a test, he will use conscious thoughts to remember important data.  Intelligent pathways will identify important information in books.  In fact, all data that the robot reads in will be stored in memory in an organized manner (via intelligent pathways).  The way that the robot remembers important information is to control its focus.  The robot might read in a fact and he will activate the thought:  “remember this fact”, the fact will be stored as strong data, both in the long-term memory and the brain of the robot.  When the robot takes the test the very next day and questions are asked “related” to a fact, the fact will activate because the robot remembers that information.

The conscious has computer programs that organize data in the long-term memory.  The data in long-term memory has reference pointers to data in memory (the robot’s brain).  Because data in long-term memory are new, those data are stronger than data in memory and can be remembered easily.

FIG. 17 is a diagram illustrating how data in long-term memory are organized through computer programs in the conscious.  The robot has played the legend of Zelda for 3 hours; and during the game, the robot has talked to a dozen characters in Zelda land.  All quotes from characters are important because they are the primary source of instructions to follow in order to beat the game.  In the diagram, all quotes are organized by where they were encountered in a map (provided by the game) and visual sequences of encounters are stored in the map in terms of when they were encountered.     

These types of quotes are summarized information that relate to a quote from a character.  For example, a 2 minute speech by one character can be condensed to 1 sentence or 2 pictures.  The robot will extract the information from this map and use intelligent pathways to analyze the data further.  By having the conscious organize data in the long-term memory, the search function can find data very quickly.  If the robot is stuck in the game and doesn’t know what to do, it will first search important data in memory.  The quotes from characters in the game will be searched first.  For example, the robot might remember that one of the characters told him to search for a hidden cave in lake hylide.  This memory will instruct the robot to go to lake hylide and search for a hidden cave.  This hidden cave might have a character that will tell the robot his next mission.    

FIG. 17


Controlling the variables the robot has to search in memory

When the robot is asked a question, an intelligent pathway can be used to identify important search variables.  These search variables will then be used to search for data in memory.  If the robot was asked a question such as:  “what are the disadvantages of playing sports?”, the search variables might be:  bad, sports, output things.

FIG. 18 is a diagram of what the search variables should look like.  The search variables are actually a fabricated diagram or movie sequence, whereby the variables are identified and the relationships of the variables to actions are established.  Even the outputs of what to search for are outlined.  In this case, the search function needs to output the “things” that are bad in terms of sports.  

FIG. 18

Actually, the computer programs inside the conscious will do things in linear order.  It will first search for the variables:  bad, sports; and extract information about these two variables.  A valve of information will flood the conscious and another intelligent pathway will be used to interpret the data and to extract specific information.  The task of finding out “things” that are bad in sports require another intelligent pathway to analyze and extract information from the data in the conscious. 

On second thought, questions and answers are learned in school and each type of question has a unique search function.  It’s up to the robot to learn how to answer “all” questions in school.  The searching and processing instructions to answer a question is based on finding patterns between very similar examples.  For example, the question:  “what are the disadvantages of playing sports?” require many examples stored in memory.  The teachers have to guide the robot to say the right answers for that question.  The robot’s brain will do all the hard work of comparing similar examples and to come up with an optimal way to search for the right answer for that question.  FIG. 19 shows that when the search pattern is found, it will be stored right after the question.  The next time the robot is asked a question, the unique search pattern will be used to output the correct answer. 

FIG. 19

All questions and answers are learned through fuzzy logic.  This means the robot can answer similar questions from the ones learned in memory.       


Controlling activation of facts about target objects

In conventional grammar software, words are parsed using a language parser.  Each word is assigned to their fixed definitions.  For example, the text “bat” is a noun or the text “run” is a verb.  Any intelligent program that uses grammar rules and language parsers are not self-learning.  They require computer scientists to input information. 

In my human robot (an AI program) there are no language parsers or grammar rules.  Words that are recognized by the robot do not go through a language parser. 

When the robot recognizes a text such as “bat” the robot will activate the meaning of the word.  A picture of a bat might appear or the sound of the text might activate.  All element objects (meaning) for the target object (text of bat) will compete to activate.  The text word of “bat” has other meanings as well.  The text word “bat” is also a noun.  The question is how does the robot activate the correct facts related to a target object?

The conscious is the answer.  The conscious controls the activation of element objects related to a target object.  If the robot was doing a grammar worksheet and the instructions are to identify nouns and verbs, when the robot reads in the text “bat”, the conscious will extract the fact “noun” from memory and assign that noun to the text “bat”.     

On the other hand, if the robot was doing an identification worksheet and the instructions are to assign visual images to texts, when the robot reads in the text “bat”, the image of a bat will activate.  The robot will compare the image activated to the image on the worksheet and determine that they are similar. 

Thus, based on a problem, the conscious has computer programs set up to extract specific element objects from recognized target objects.  This information will be used to solve a given problem.  The computer programs in the robot’s conscious are actually searching for the target objects in memory and extracting element objects in a heuristic manner.

It really depends on what problem the robot wants to solve.  If the robot was doing a grammar worksheet, then it will activate grammar rules related to texts it recognizes.  The robot’s mind will simply activate grammar facts about the text it reads.  If the robot was a teacher and is correcting student essays, then a computer program in the robot’s conscious will specifically cater to identifying texts in terms of grammar, identifying sentence/paragraph errors, identifying misspelling or wrong words to use in sentences and so forth.  The computer program will know that its goals are to grade essays and its job is to find errors in essays.  Lessons learned in grammar class will activate and managed by the robot’s conscious.  

Another example would be reading Shakespeare books like Hamlet.  A more analytical pathway must be used to read text from Shakespeare.  A more logical form of intelligence is needed to understand what sentences mean and how multiple scenes relate to each other.  The meaning to sentences and paragraphs by Shakespeare go beyond what is presented in the text.  He uses artistic expressions that make his work a masterpiece. 

If the robot was reading text from a novel, he doesn’t have to worry about grammar rules or deep understanding of the text.  All he needs to do is read the book and create a fabricated movie that explains, in a visual way, what the book is about.     

Teachers teach the robot what to do for a specific task.  If the robot’s task is to grade student papers, the teacher will teach the robot to identify grammar errors, misspellings, content of paper and so forth.  Thus, the next time the robot has the task of grading papers, he will use grammar rules, check grammar errors, and everything English teachers taught the robot about grading essay papers.  On the other hand, if the robot was reading a novel, the teachers would teach him to visualize the texts or teach him nothing at all.  The teacher didn’t ask the robot to pay attention to grammar structure or identify sentence types.  It’s natural at this point for the robot to fabricate a movie, while reading a novel.  However, hierarchical activation or random activation might tell the robot that this word is a noun and this word is a verb, etc.  For the most part, reading any material with text, will activate English lessons because of association.  For specific tasks, specific English lessons will activate. 

More advance novels like Hamlet or the scarlet letter will activate lessons in advance English classes, whereby the robot has to analyze and logically come up with facts that are not described in the text.   


The conscious can generate a map of the environment

When the robot plays a game like the legend of Zelda, it is very important that the robot have a map of the environment.  In the game, sometimes maps are given to the player and other times maps are absent.  In the event maps are absent, the player has to use temporary memory to remember where he has gone in the past and what the environment probable looks like. 

Some dungeons in the game resemble complex mazes.  Sometimes, the player has to enter rooms and remember what these rooms look like.  For example, the player can have the option of entering 3 rooms (FIG. 20).  The player enters room1 and found out it leads to a dead end.  The player goes back to the main room and enters room2.  Room2 is another dead end, so the player goes back to the main room and enters room3 where he found the level boss. 

FIG. 20

The robot’s brain has to remember what is in each room and to remember where it has gone in the past.  The computer program in the conscious will generate a temporary 3-d map of the environment, based on a short past, of where it has gone and what information was stored in each room.  This 3-d map will be used for logical purposes to make future decisions. 

This 3-d map is very basic and the robot can only remember a short area of the environment.  Referring to FIG. 20, only several rooms and their locations to each other are presented in the 3-d map.  If the environment is too complex the robot will be confused in terms of what the environment looks like.  This is one of the reasons why a human being will get lost if they are put into a complex maze. 

FIG. 21


As the robot encounters more rooms the old map is forgotten, and a new up-dated map will be generated.  By the way, the map in FIG. 21 is a map that was learned through example.  In a past gameplay of Zelda, the robot goes through several rooms (called movie sequence1) and then presses the start button, which shows a map (called map1).  If the robot learns many various examples of movie sequence1 and map1, then a pattern begins to emerge. 

In the future, when the robot encounters a situation similar to movie sequence1, then a fabricated map1 will activate in the conscious.  The fabricated 3-d map will look similar to map1. 

In addition to the maps that are generated by the robot’s conscious, fabricated maps can be created from data in memory.  A fabricated movie of lines drawn on a surface will show the robot where it is and where did it go in the last several minutes (or hours). 

Also, maps that are seen in the past and remembered can also activate to help the robot navigate its surroundings.  For example, if the robot learned all the streets and cities of Hawaii, then it can activate a map of Hawaii; and come up with logical outputs.  Taxi drivers have a very good map of the environment because it’s part of their job – to drive people around the city.  When the robot is driving a car and needs to go somewhere, a previously learned map will activate which will tell the robot where it’s currently at and where its destination location should be.  The robot will use intelligent pathways to draw lines on this remembered map to plot a route to its destination location.        

Fabricated maps can be general (comprising lines and circles) or detailed (comprising remembered city maps).  A general map of a country can be created or a detailed map of a small area in the country can be created.  The intelligent pathways generate these fabricated maps.  If the map is big like the United States, the mind can travel from one state to the next using animation zooming (zoom in or out).  Maps can be traveled using car speed, human walking speed or plane speed.  The robot’s conscious can even draw lines and diagrams on the fabricate map to indicate a function.  Traveling on states can be represented by animated or static arrows.  The fabricated maps can also convert from general to detail at any given moment. 

These types of intelligent pathway to manipulate and fabricate maps are based on the robot visually experiencing these behaviors in real life.  For example, your local news station shows the United States map and the camera zooms in to Nebraska.  This animation is learned by the robot and intelligent pathways mimic its behaviors.   


<< content               next chapter >>



Home | HLAI | UAI | Books | Patents | Notes | Donation

Copyright 2006 (All rights reserved)