FAQSearchEmail

humanlevelartificialintelligence.com   

  
 chapter7

Home | Videos | Contact Us   

 
Home
HLAI
UAI
Videos
Books
Patents
Notes
Donation

     
 

                      

                         << content                                      Chapter 7

4.  Reading science books

In this chapter, I will increase the complexity by discussing how the robot learns knowledge from science books.  In order to learn science knowledge, all three previous learning methods have to be used.  In science books, there are facts/knowledge about objects; and there are diagrams that explain how objects work.  In the previous learning methods, we discuss how knowledge is learned and stored in memory.  These knowledge are based on sentences and paragraphs.  What makes science books particularly difficult to learn is that it contains diagrams and flow charts to explain a process.

In addition to reading sentences and paragraphs, the robot has to also learn to understand diagrams and how they work.  The sentences in science books explain how diagrams work.  They give facts about various areas of a diagram and also to give the reader a fabricated animation of the sequential steps of a process in the diagram.

The pathways that the robot stores in memory contain both static data and linear data.  Diagrams are considered static data because they are still pictures in the science book.  Some diagrams are flow charts, so they are actually an animation that represents a process.          

 

A neural network example 

If you read an AI book about neural networks, they will give you a diagram similar to FIG. 34.  The author will explain using sentences how the neural network works.  First, the book will explain the parts of a neural network.  There are three layers in a neural network:  the input layer, the hidden layer, and the output layer.  The explanation of the parts related to a neural network gives facts about what each part does.  By understanding the individual parts, the author can then explain the linear functions of the neural network.  The steps to a neural network includes:  inserting data into the input layer, then the output layer will delineate the desired output, and finally the neural network modifies all of its connection weights in all three layers.  The learning of the neural network happens during the modifications of the connection weights.

FIG. 34

The above information only outlines a general idea of the neural network.  The science book will explain further the details of how the neural network works.  In order for the robot to organize data, it has to use one type of diagram for one type of explanation.  For example, the diagram in FIG. 34 is used for a general explanation.  In FIG. 35, the diagram is used to explain how the connection weights are modified.  The robot will see the steps that the neural network will go through to change the weights.  Equations are given in the science book and the robot has to use these equations and apply them to the connection weights of the neural network. 

FIG. 35

In some cases, the teacher will ask the robot to write down the flow chart of all the connection weights and how they are modified.  The book will tell you how to apply the equations to the connection weights.  The robot has to follow the steps and write down the numbers on each connection weight step-by-step.  If the robot did the task correctly, the connection weights will have the correct numbers. 

For every detail diagram of the neural network, the information will be stored in their respective area.  The connection weight diagram in FIG. 35 will be stored in the general neural network diagram in FIG. 34, in the connection weight area.  These diagrams will also self-organize and their optimal area of storage will be outlined.   

If there was a diagram to explain each layer of the neural network, then each diagram will be stored in the general diagram in their respective places.  The detailed diagram for the input layer will be stored in the input layer, the detailed diagram for the hidden layer will be stored in the hidden layer, and the detailed diagram for the output layer will be stored in the output layer. 

Another type of diagram is used to show the different variations of neural networks.  Instead of 3 layers, some neural networks have 4 or 5 layers.  The author can explain what kind of variables can be used in each layer.  The author can also explain what kind of techniques to use to determine the variables in the desired output.

 

Sentences store different timelines of events

In history books, the authors give information about certain events in history.  The author will tell you, through the books, when these events occurred.  The constitution convention occurred in 1787.  The author will then give you information about what happened during that time period.  The fabricated movie created by the robotís conscious, while reading the section on the constitution convention, should record the event occurred in 1787.   

History books arenít a very good example to use.  Instead, I will use comic books.  Referring to FIG. 36, the fabricated movie can actually be other timelines besides what is currently happening.  It could be a scene that happened in the past or the future.  It can also be a scene that is currently happening now, but isnít located in the current pathway.  The current scene (the fabricated movie of a current non-viewable scene) can be a scene that is happening 1 mile or 1,000 miles from the current pathway.      

FIG. 36

The primary story is based on the current pathway (the sequential panel recognitions of the comic book).  The fabricated movies of other events happening at different times are secondary events.  The characters in the comic book are explaining past events or future events.  Maybe Charlie brown is planning to buy a lawnmover tomorrow or Charlie brown bought a new bike at the department store yesterday.  These types of information doesnít exist now, but it either happened in the past or will happen in the future.  Sometimes, Charlie brown will describe a 2 minute event that happened yesterday.  The reader has to visualize what that 2 minute event is like (a fabricated movie).

Fabricated movies should generate a 5 sense data representation of any grammar sentence.  A sentence might include an object belonging to Charlie brown or Charlie brown taking an action.  Thus, the fabricated movie uses 5 sense data to generate a meaningful representation of the sentences read.  These fabricated movies should represent all objects, events, and actions in the current pathway and establish relational links between them in terms of physical properties, hidden thoughts, object interactions, time, and so forth. 

The fabricated movie can also reference events that happened in the current pathway in the past or future.  Reference pointers will be established between events in the fabricated movie and past/future events of the current pathway. 

 

5.  learning science lessons from teachers

A more powerful way to learn information is to have a teacher demonstrate examples and give facts about different subject matters.  Itís one thing to read a science book and learn the information, but to have a teacher teach the lesson is a totally different experience.  We learned how to study when we were in grade school.  We learned the procedural steps that are needed to pick up a science book and to learn the information.  Reading a comic book is totally different from reading a science book.  Reading a Chinese book is different from reading an English book.  There are rules that have to be followed based on a given book.  These rules decide how to read a book, how to look at diagrams, and how are the chapters in the book arranged.  (I will be discussing how to form intelligent pathways in memory to read a given book in the next chapter).  

The reason why a teacher is needed to teach the robot knowledge is because he/she serves as a guide that helps the robot focuses on certain objects and illustrate, using a chalkboard (or monitor), how things work in a sequential manner.  If the teacher is trying to teach the robot an addition problem, the teacher has to write things on the board, erase things on the board, focus the robotís eyes on certain things on the board and so forth.  While the teacher is illustrating a lesson, he can also give facts and information about that lesson.

The robotís brain has to store all these lessons in memory in an organized way.  If you combine all 5 learning methods:

1.  reading novels.  2.  reading comic books.  3.  reading history books.  4.  reading science books.  5.  learning science lessons from teachers, 

the robot should be able to organize the knowledge it experiences from the environment in an intelligent way.  Not all information is stored from the environment.  The robot uses intelligent pathways to focus on limited data from the environment and it stores these data in memory in an organized manner.  The teacher that is teaching the lesson is the guide that will help the robot to focus on what matters from the environment and what information should be stored in memory.  For example, the robot looks at the teacher and sees the way he/she dresses and he sees the furniture in the classroom, including the chalkboard.  However, despite all this information, the lesson on the chalkboard is the primary data that should be stored in memory.  The robot will give top priority to the lesson on the chalkboard and low priority to anything else. 

This type of teaching is much easier to learn compared to reading a book because books has fixed diagrams and text.  If the robot doesnít understand something he can always ask the teacher.  If you read a science book and you got questions, the book canít give you an answer, you have to search for the answer yourself. 

Learning information from teachers should be applied to mathematics.  If you try to learn math from a book, it would be very difficult.  People learn information mainly through visual images.  Diagrams in math books are very limited because they take up too much space.  For example, flow diagrams take up a lot of space in books.  Usually, the author will provide a detailed diagram and expects the reader to use common sense to fill in any missing data.  This confuses the reader and will make it more difficult to understand the lesson. 

Teachers, on the other hand, understand the lesson and he/she will teach the class so that the students can understand the lesson. 

Believe it or not the conscious mind of the robot is formed by lessons from teachers.  Even when a person is reading a science book, the conscious is like a teacher that tells you things about the content.  In other words, you are able to read and understand the science book because of lessons from teachers about guidance. 

Teachers have a finger and can point to certain areas on a chalkboard.  The pointer is a guide to focus the robot to a certain area.  The teacher can also write things down, erase things and draw action lines on the chalkboard.  Most of the lessons on the chalkboard can be categorized as static data or linear data.  A math equation demonstrated by a teacher will show existence of numbers (static data) and linear steps (linear data) in order to solve.  Through self-organization of many similar examples, the robot will know which are static data and which are linear data. 

Teachers can also use finger pointing and diagram drawing to show relationships between objects.  These relationship patterns can also help in configuring data for storage purposes.  For example, the teacher can draw 5 diagrams on the chalk board and he/she will point to one diagram and say that object1 is a parent to objects2-5.  This lesson will create a diagram in the robotís mind and store object1 as a parent node and objects2-5 as child nodes.  The teacher might draw a diagram of a line and he puts object2 on the left side and object5 on the right side and say that there is a range between object2-object5.  Object2 is low and object5 is high.      

These types of lessons coming from teachers help to organize the material being learned.  In the future the robot can use these demonstration lessons and apply them to understanding materials in science books.  For example, if you read the last paragraph again you will create the diagrams in your mind.  I did not physically write on the chalk board the two diagrams.  I simply written a paragraph describing a teacher drawing the diagram.  Because you are intelligent at a human level you are able to fabricate a movie of a teacher drawing the two diagrams.  The first diagram is a hierarchical tree of a parent node represented by object1 and 4 child node represented by objects2-5.  The second diagram is a line with object2 on the left and object5 on the right.  The range from left to right is low to high.  

 

Self-organizing knowledge from multiple books in the brain

Books are read in the classroom or at home or at a park.  Because robotís store data in a 3-d grid, the location of where the robot read the book is where the data will be stored. 

The next factor is that the visual book also helps organize data.  The book is a visual media and the robot is storing information from the book in a linear manner (page by page) of what it had read.  Facts/knowledge, diagrams, pictures, flow charts are all arranged in a 3-d manner of where it was read in the book.  All data will also have relational links to each other. 

If the robot read a math book in school, at home and at a park at different times, then the information is stored in their respective locations (FIG. 37).  During the self-organization phase, the data in the math book read at school and at home will self-organize (they combine the data and share them).  The data from the park is too far away and isnít subject to self-organization.  Self-organization prevents repeated data from being stored in memory.  After self-organization, there are two areas where the data from the math book are stored:  the location between school/home and the park.  The masternode will represent the copy that has the strongest powerpoints, which in this case is 23.      

FIG. 37 

Letís say that in the future, the robot reads the same math book (the same chapter) at a different location.  The robot is in a library and is reading the book and the library is located very far from the school or park.  The robot will initially store the math problem in the library location, however, after reading the data for a short while it will start to store the data in the school area (the masternode).  The reason for this is because the data in memory regarding the book is so strong in the school area that when the robot reads the book and is unaware of its environment, the robotís brain found the school area to be an optimal area to store the data from the math book.  This method demonstrates that storage of data doesnít totally depend on the current visual environment. 

Referring to FIG. 38, as stated before, the contents learned in books are stored as linear data, organized page by page.  Information are stored in an estimation.  In real world examples, the brain forgets information and it confuses sequential pages.  The diagram shows facts/knowledge being stored, along with diagrams and pictures.  Most of the time, the knowledge is referring to the diagram and reference pointers are included.  

FIG. 38

 

The color of the book, the title of the book, the author, the subject matter and so forth are information used to organize the content read in the book.  Despite all the content inside a book, the robot will self-organize many books in his lifetime.  The self-organization will depend on the subject matter.  Physics books will be closer to chemistry books, love poetry will be closer to romantic books, math books will be closer to computer science books and so forth.  These books will self-organize all their content.  As the robot learns more information, they will be structured in a hierarchical manner (FIG. 39).

 

 

 

 

Organizing data through the robotís conscious thoughts

The above methods show that the robotís brain can store books in the place that they were read, and also, by the visual book itself.  The content of the book will be stored in an organized way through the robotís conscious thoughts.  Referring to FIG. 40, when the robot recognizes something the conscious activate sentences or words to id objects.  The robot will id the topics contained in paragraphs or id important facts from several pages.  The robotís conscious might also summarize several pages of reading.  All this information that comes from the robotís conscious is used to organize data. 

All facts from topic1 will be stored in topic1, all facts from topic2 will be stored in topic2 and all facts from topic3 will be stored in topic3.  Hierarchical data storage is also possible.  Topic1, topic2, topic3 and summarizing facts are all stored in the 2 hour reading.

This hierarchical data storage is an estimation and data in memory forget information.  The beginning and the ending of topic2 is just an approximate estimation.  Pathways are broken up into a plurality of fragmented pathways when they are forgotten.  Some data might even migrate to distant areas.   

 FIG. 40

 

Intelligent lessons learned from teachers

The lessons learned in school form intelligent pathways in the robotís memory.  The robot can do math, science, draw, speak and so forth.  Some intelligent pathways require the robot to write down things such as an addition problem.  Other times, the robot can do simple tasks in its mind, like planning routes to drive a car from the current location to a destination location.  When planning routes, the robot will id its current location, activate a fabricated map in its mind, draw lines to connect where it is and where it wants to go, and remember the linear streets it has to travel. 

The reason that the robot is able to plan routes in its mind is because of doing the same task in the real world.  In school, teachers give the robot worksheets to do.  A simple worksheet is a mouse in a maze and the robot has to draw lines so that it can get the mouse from the start location to the end location.  The robot has to analyze all the routes and plan the routes by drawing lines.  If the robot is stuck with a path, it can erase the line and take other paths. 

By doing worksheets of plotting routes, the robot is able to use this lesson on any type of similar problem.  It can take a toy car and push it around to a destination location or take a pencil and draw routes to get a car to a destination location. 

In the case of planning routes in the robotís mind, the robot has to activate a map of the city.  Next, it will use the lesson from worksheets related to planning routes, to determine what streets to drive to.  Finally, the robot remembers the linear streets to go to and apply them to the real world.  Planning routes on the road is a simple example, and the robot can do this all in its head.  More complex route planning will require the robot to write down on paper the routes needed to get to a destination location.  Usually, human beings plan at the moment.  If the robot had to drive to a far distance area, he has to plan 3-4 times to get there.    

I canít emphasize how important it is for the robot to have the ability to do all worksheets.  The worksheets establish vital intelligent pathways in memory to do things.  The robot needs to learn how to cut and paste images together in order to fabricate make believe movies in its head.  When reading a book, a fabricated movie is created.  The fabricated movie is based on intelligent pathways that take lessons from cutting and pasting images in the real world.  If you have a worksheet that wants you to cut objects out and paste them together in a meaningful way, then the robot has to observe the objects and see which objects belong to each other.  A dress should be pasted onto a girl and high heals should be pasted onto a women.  A tuxedo should be pasted onto a man and a toy should be pasted onto a baby.  The robot also has to know where they should paste these objects.  The head of a girl should be pasted onto the top-middle of a dress.   

Lessons about coloring books and cutting out boundaries of an object can be used for the robotís internal image processor.  This image processor can delineate an image from a picture or movie sequence without any programmed computer codes.  The robot learned how to cut out images from lessons in real life. 

Looking at forms in the real world will generate template forms to store background knowledge on objects (like human beings).  The observation of the form and all of its slots teach the robot how background data on a person should be stored and structured. 

The search functions in the pathways of the robot are done by patterns found between similar examples.  No programmers are needed to write the computer codes to the search functions.  The robot does all the hard work by comparing and searching for these patterns between similar pathway examples.     

 

More on reading science books

The robot learns a science book very similar how it learns a history book.  There are slight differences.  For one thing, in the history books, the reader is more interesting in creating a fabricated movie of scenes from the text.  Itís almost like watching a movie of the history book.  On the other hand, the science book requires looking at diagrams and understanding flow diagrams. 

The understanding of a neural network require, first, understanding the element parts of the neural network:  input layer, hidden layer and output layer.  Next, the overall idea of the neural network is learned.  The programmer has to define the output layer based on trained input data.  Next, the programmer has to train the neural network with many training examples.  During the training phase, the connection weights adjust between the input layer and output layer so that it can ďlearnĒ.  When the neural network is trained properly, it can be used by people. 

In FIG. 41, the diagram of the neural network is the bulk of the information from the science book.  All the text are explaining how the diagram works.  The various diagrams from the book will organize itself into hierarchical levels.  For example, an overall diagram of the neural network will be at the top and a diagram of the input layer will be located in the input layer area of the overall diagram. 

FIG. 41 

        

In science books there are diagrams as well as flow diagrams.  The flow diagrams show you step by step how the neural network works.  The robotís brain can use previously taught lessons from teachers (an intelligent pathway) to convert the flow diagram into an animated flow diagram.  The intelligent pathways can fabricate a movie of what the flow diagram should look like.  For example, a fabricated movie of how a neural network work can be a diagram of a neural network, a user inputs data into the input layer and the connection weights change, and finally, the output layer outputs something. 

The fabricated movie will require the flow diagram and the sentences that tell the reader how the steps work linearly.  The diagrams along with the sentences will generate a more detailed animation of the functions of a neural network. 

These fabricated movies are also put into the overall diagram according to its associations.  If the fabricated movie creates steps to how the input layer works, the fabricated movie will be stored right next to the input layer in the overall diagram. 

Even if the book doesnít have a flow diagram, instead, have a still diagram, the robot can still generate a fabricated movie based on the sentences in the book.  The still diagram might be the overall diagram of the neural network.  The sentences around the diagram will explain the linear steps to how the neural network operates.  This fabricated movie of the function of the neural network will be stored in the same area where the overall diagram is stored in memory.        

The important thing to remember is that the information stored in memory about the neural network are based on 5 sense data, mainly diagrams and fabricated movies.  The extraction of knowledge is based on the intelligent pathways in memory and not the data stored in memory.  For example, letís say that a teacher asked the robot this question:  ďwhat color is a human heartĒ, the robot will only extract a picture of a human heart from memory.  An intelligent pathway is needed to analyze the human heart and determine that the color is red.  Itís the intelligent pathways that really do the hard work and not the data stored in memory. 

How exactly does this analyzing intelligent pathway form in memory?  Teachers will teach the robot in real life using examples.  The teacher will take out a picture of a heart and ask the robot:  ďwhat color is the heartĒ.  The robot will focus on the color and say:  ďthe color of the heart is redĒ.  In order to understand how to answer this question the teacher uses supervised-learning, whereby he/she has to ask the question and give the answer.  The robotís responsibility is to find the search pattern between the question and the answer. 

This intelligent pathway was trained to analyze a picture that is currently focused on by the robot.  It can also be used to analyze a picture in the robotís mind.  The teacher can trick the robot and ask this question:  ďI want you to remember the heart picture, what was the color of the heartĒ.  The robot will activate the heart picture presented earlier and it will use the intelligent pathway of analyzing the picture to output:  ďthe color of the heart is redĒ.    

 

Intelligent lessons learned from teachers

The lessons learned in school form intelligent pathways in the robotís memory.  The robot can do math, science, draw, speak and so forth.  Some intelligent pathways require the robot to write down things such as an addition problem.  Other times, the robot can do simple tasks in its mind, like planning routes to drive a car from the current location to a destination location.  When planning routes, the robot will id its current location, activate a fabricated map in its mind, draw lines to connect where it is and where it wants to go, and remember the linear streets it has to travel. 

The reason that the robot is able to plan routes in its mind is because of doing the same task in the real world.  In school, teachers give the robot worksheets to do.  A simple worksheet is a mouse in a maze and the robot has to draw lines so that it can get the mouse from the start location to the end location.  The robot has to analyze all the routes and plan the routes by drawing lines.  If the robot is stuck with a path, it can erase the line and take other paths. 

By doing worksheets of plotting routes, the robot is able to use this lesson on any type of similar problem.  It can take a toy car and push it around to a destination location or take a pencil and draw routes to get a car to a destination location. 

In the case of planning routes in the robotís mind, the robot has to activate a map of the city.  Next, it will use the lesson from worksheets related to planning routes, to determine what streets to drive to.  Finally, the robot remembers the linear streets to go to and apply them to the real world.  Planning routes on the road is a simple example, and the robot can do this all in its head.  More complex route planning will require the robot to write down on paper the routes needed to get to a destination location.  Usually, human beings plan at the moment.  If the robot had to drive to a far distance area, he has to plan 3-4 times to get there.    

I canít emphasize how important it is for the robot to have the ability to do all worksheets.  The worksheets establish vital intelligent pathways in memory to do things.  The robot needs to learn how to cut and paste images together in order to fabricate make believe movies in its head.  When reading a book, a fabricated movie is created.  The fabricated movie is based on intelligent pathways that take lessons from cutting and pasting images in the real world.  If you have a worksheet that wants you to cut objects out and paste them together in a meaningful way, then the robot has to observe the objects and see which objects belong to each other.  A dress should be pasted onto a girl and high heals should be pasted onto a women.  A tuxedo should be pasted onto a man and a toy should be pasted onto a baby.  The robot also has to know where they should paste these objects.  The head of a girl should be pasted onto the top-middle of a dress.   

Lessons about coloring books and cutting out boundaries of an object can be used for the robotís internal image processor.  This image processor can delineate an image from a picture or movie sequence without any programmed computer codes.  The robot learned how to cut out images from lessons in real life. 

Looking at forms in the real world will generate template forms to store background knowledge on objects (like human beings).  The observation of the form and all of its slots teach the robot how background data on a person should be stored and structured. 

The search functions in the pathways of the robot are done by patterns found between similar examples.  No programmers are needed to write the computer codes to the search functions.  The robot does all the hard work by comparing and searching for these patterns between similar pathway examples.    

 

<< content               next chapter >>

 

 

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

Copyright 2006 (All rights reserved)