humanAI.com

 
taskengine
timemachine
patents
notes
donation

 

 

 

Universal Artificial Intelligence

Content on this page were cut, copy and paste from my books or patent applications.

In this page I will try to explain how the UAI is applied to individualized robots to do specific human tasks.  The UAI is one software program that can replace all human jobs.  It can:  drive a car, fly an airplane, cook food in restaurants, do janitor work, collect fruits in farms, work in a factory, deliver mail, or diagnose patients.   

The first step in building a UAI is to first build a human robot.  My definition of a human robot is a robot that can sense, think and act like an adult human being.  The human robot has to graduate from college with a difficult degree (such as an engineering degree) to prove to the world that it has achieved human level artificial intelligence.  This human robot is used to train the UAI for a specific human job.  The human robot will have the ability to record experiences (called the robot's pathways) from its brain while doing a human job.  FIG. 46 is a diagram depicting two brains:  1.  the human robot's brain.  2.  a specialized brain.  The specialized brain is a container specifically designed to store the human robot's pathways, while doing a human job.  For example, if the human robot's job is a taxi driver, it will record all pathways that are selected from its brain while it's driving.  By the way, the human robot's pathways record what the robot is sensing from the environment, as well as, what the robot is thinking about.  Within the specialized brain there is also a common sense knowledge part.           

The human robot's pathways contain the robot's goals and rules it must follow.  The robot knows that his job is a taxi driver and he has to get passengers from a current location to a destination in the safest way possible.  He knows common sense things, like if the passenger doesn't pay upfront don't drive the car or if the passenger is soaking wet, don't let the passenger into the car.

The robot's pathways in the specialized brain contains a "self-creating software program" to drive a car.  The extraction of rules and hierarchical goals, solving conflicting rules, following procedures of driving during any given situation and so forth, are part of this self-creating software program (FIG. 47). 

 

The human robot is the trainer of the UAI (for a car).  During each training (each driving situation), the robot's brain knows exactly want to do and think.  Thus, a human robot is the perfect trainer for an autonomous car.  If you look at Stanford university's AI car or MIT's AI car, they need human programmers to manually feed in the rules and goals of the AI car.  They use machine learning, whereby a human being has to drive the car in order to train the AI car.  Also, their AI cars use lasers, GPS, and navigation software.  Human beings don't need these tools to drive a car. 

In FIG. 48, a dummy robot is a human robot without a brain.  The specific brain is inserted into the dummy robot and the dummy robot is put inside a car to serve as a taxi driver.  If you look at Stanford University's AI car or MIT's AI car, there are no robot drivers -- the foot pedals and the driving wheels move by themselves.  In my UAI program for a car, there is a physical dummy robot inside the car turning the driving wheel and pushing the foot pedals.  There are certain advantages to this type of AI car.  For one thing, the dummy robot can drive any car regardless of the model or year.  For example, a dummy robot can drive a 1920's car or a 1970's car.  It doesn't matter if its' model is a Ford, Toyota, Honda or GM.

The dummy robot can also control any type of machine.  You can take the dummy robot and put it in a train or a plane, or a boat, or in a factory or restaurant. 

Another advantage of a dummy robot is that it can fix the car if any problem arises.  If the car has a flat tire, the dummy robot can change the tire.  If the car runs out of gas, the dummy robot can go to a gas station to refuel.  If the car doesn't start, the robot will run diagnosis on all system parts, and if a problem is identified, it will fix the problem.  Thus, this dummy robot not only has to have knowledge about driving a car, it also has knowledge about how to maintain and fix cars.     

 

Knowledge of driving

In the specialized brain, the robot has to record the learning of knowledge about driving such as facts, rules and standard procedures regarding driving.  For example, when the human robot is reading a book about how to drive, those pathways are stored in the specialized brain.  The human robot's brain will help self-organize information in the specialized brain in an optimal way.  The entire experience of learning to drive a car is stored in the specialized brain.  Bad information, bad driving methods, misinformation, and so forth will be deleted from the specialized brain during the universalized stage, whereby similar pathways are compared and universalized.  Bad facts are also corrected by the real human robot when he identifies a bad fact. 

However, to minimize mistakes and extraction of bad information, the human robot should train itself to learn the correct facts, to drive in a perfect manner and solving problems with minimal trial and error.   

Personal experiences from the real human robot should not be recorded in the specialized brain.  The human robot should not record experiences when it is sleeping, or taking a vacation, or doing a hobby or talking to friends.  If the real human robot is thinking about personal things while doing its job, the self-organization phase will eliminate these thoughts from the specific brain.  

It is very important that all the rules, facts and standard procedures of driving a car are stored inside the specialized brain.  FIG. 50 is a diagram depicting the robot's pathways having references to the rules and facts related to driving.  The more the human robot drives, the more organized knowledge about driving will be in the specialized brain.        

 

Common sense knowledge       

The specialized brain also contains a common sense knowledge part.  The specialized brain has to have basic knowledge that a 10 year old has.  It has to know how to count things and add numbers together.  For example, if the dummy robot is a taxi driver and 4 people steps into the car, the robot needs the skill of counting the number of passengers and understanding that the taxi car can only seat a maximum of 3 people.  Also, the dummy robot needs to know common sense knowledge.  For instance, if a mother is carrying a child on her lap, the dummy robot will tell the mother to put the child on a special seat.  This robot knows that the mother is engaging in a dangerous situation and the child might get hurt if the taxi is involved in a car accident. 

There are no shortcuts to creating common sense knowledge in the specialized brain.  The dummy robot will have to record all learning of knowledge in elementary school.  All skills that a 10 year has must be stored in the dummy robots brain.  Elementary school knowledge serves as a foundation for understanding natural language and logical skills.  Basic math, english, logic, history, verbal communications, medicine, geography, social skills, and so forth must be learned by the dummy robot. 

The reason I didn't want the specialized brain to store learned knowledge from high school and college is because I didn't want to build a real human robot.  If you build a real human robot and you force that robot to work as a taxi driver 24/7, you are committing robot slavery.  In other words, you are breaking the 13th amendment and forcing a living entity to work. 

The idea for the dummy robot is to instill minimal amount of knowledge in its specialized brain so that it can understand natural language and common sense knowledge.  Once it has a fundamental knowledge base, you can teach it to do many basic human tasks.  A 10 year old can drive a car, or cook food, or do janitor work, or take care of patients or work in a factory to mass produce products.  A 10 year old needs to know basic grammar so that he/she can follow commands from a boss.  

The common sense knowledge part can also be trimmed further by deleting knowledge that don't really matter, such as the identity of the dummy robot or certain past experiences.  It really depends on what type of robot you're trying to build.  If the dummy robot is a taxi driver, the past experience of playing basketball at age 8 will not be stored in its specialized brain.  Other past experiences, like going camping with classmates at age 9 will not be stored in the specialized brain.            

When the dummy robot is driving a taxi, it might run into problems along the way.  It needs skills of identifying problems and solving them.  That knowledge can only be learned through elementary school.  The real human robot that is doing the training for this dummy robot will train the dummy robot to drive under the most popular situations.  However, there will be situations the dummy robot will drive in that wasn't trained by the real human robot.  Under these unknown situations, the dummy robot will use common sense knowledge and problem solving skills from elementary school in order to drive the taxi safely from the current location to its destination location.  If the dummy robot is trained properly and enough knowledge is learned, it can drive the taxi safely under "any" given situation.         

 

How to train a dummy robot to fly a plane

Training a dummy robot to fly a plane is exactly the same as the training used to drive a taxi.  The specialized brain in the dummy robot has a common sense knowledge part, which is knowledge learned from elementary school.  First, the real human robot has to acquire knowledge on flying a plane.  This is accomplished by reading books or attending flying school.  The real human robot has to record all pathways (also called experiences) related to learning how to fly a plane.  Finally, it has to train the dummy robot how to fly a plane.  The real human robot records every second of its life, while it is flying a plane.  All those pathways from the robot will be stored in the specialized brain.  After many years of flying, the specialized brain contains all the knowledge it needs to become a pilot.  The specialized brain for flying a plane is put inside a dummy robot and the dummy robot serves as an autonomous pilot. 

This robot pilot can be put inside any plane and it will be able to fly that plane.  Some planes don't conform to standardized controls and the controls are rearranged in different areas.  The robot pilot is smart enough to control the plane regardless of the configuration of the controls.  It can fly a tiny plane or it can fly a jumbo jet.  The reason for this is because the real human robot has trained itself to fly many different planes.  However, the robot pilot should be properly trained with one type of plane.

 

How to train a dummy robot to drive a car and fly a plane

DARPA is holding an ongoing contest starting from Nov. 2010 for any Universities or technology companies to design and build a hybrid machine that can not only drive a car, but also fly a plane.  This hybrid machine takes commands from a remote or interior passenger and it drives and flies autonomously.  DARPA wanted a simple autonomous hybrid machine that is user friendly and accepts minimal input from the passenger. 

The real human robot has to first learn knowledge about flying a plane and driving a car.  This is accomplished by reading books or going to school.  Next, the real human robot has to train itself to control the hybrid machine.  Lot's of training is required in order to create a specialized brain for controlling this hybrid machine.  Innately, the specialized brain contains common sense knowledge learned in elementary school, so simple problem solving skills and common sense knowledge is had by the specialized brain.  Finally, the specialized brain is put inside a dummy robot to control the hybrid machine autonomously. 

During the training phase, the real human robot is given commands to do, such as:  "fly to locationA and drive to locationB".  The commands can be given by a user through any type of communication method.  A remote human (the user) can send a text message to the real human robot or a passenger in the hybrid machine can give the real human robot a verbal command. 

After many years of recording pathways from the real human robot while doing his job (which is driving and flying), the pathways in the specialized brain creates self-creating software programs to control this hybrid machine.  Finally, the specialized brain is put inside a dummy robot to control the hybrid machine autonomously. 

 

How to train a dummy robot to cook food in a restaurant    

The real human robot has to record pathways (or experiences) when learning to be a cook.  After it graduates from cooking school, it will record pathways while working as a cook.  These pathways are stored inside a specialized brain. 

During the training phase, the real human robot is a cook that is working based on commands given by a manager.  The manager tells the real human robot what to do at various times by giving verbal commands or text messages.  The real human robot has to follow the commands from the manager and no one else.  Actually, the real human robot has to follow commands from a chain of upper-level workers.  What if there were 4 hierarchical managers and the robot has to solve conflicts of commands? 

This is where common sense knowledge comes in.  The robot has to know what its roles, powers and rules are.  He also needs to know the roles, powers, and rules of each hierarchical manager.  He will use logic to determine if he should listen to a command or not.  For example, if one manager tells the real human robot to cook for customerA, but the president tells the real human robot to cook for customerB, the real human robot will cook for customerB because the president owns the restaurant.  The president has more power and the real human robot understands that. 

The real human robot's pathways contain self-creating software to determine wither to listen to a command or not.  Thus, whenever the real human robot is given a command, this self-creating software is activated and it will determine wither or not the real human robot will follow the command.

Each restaurant will have their own cook robots.  A Chinese restaurant will have different cook robots compared to robots in a Mexican restaurant.  In a typical Chinese restaurant there are about 200 different items on the menu.  The real human robot that is doing the training for the cook robot has to be trained to make all 200 items.  In addition to this, the cook robot has to also have the ability to make any customized item the customer wants to order.  The cook robot will be skilled at making the 200 items on the menu (depending on what is ordered most frequently), but it will not be skilled at making customized items.  For example, if the cook robot was trained to cook Chinese food and a customer wanted to order pizza, it will use the available knowledge it learned in cooking school and make the pizza to the best of his abilities.  This pizza is made using basic cooking skills and through trial and error.             

The cook robot can be trained to work in 2 or more restaurants.  For example, a cook robot can be trained to work in a Chinese restaurant, a fast food restaurant, a Mexican restaurant and a French restaurant.  The specialized brain for this cook robot will be massive because it needs to store all knowledge about all 4 restaurants and the real human robot has to train itself to cook in all 4 restaurants. 

The cook robot can also be trained to respond to customers in terms of what it can and can't do.  If a customer asked the cook robot to do surgery on a patient, the cook robot will respond by saying:  "sorry, my job is to cook food".  If the cook robot was asked to drive a car, but the cook robot has no knowledge about driving a car, it will respond by saying: "sorry, my job is to cook food".  If you train the cook robot over and over again in terms of what it can and can't do, it will have the ability to respond to customers regardless of what they ask for.    

You can create a universal cook robot that can cook in any restaurant, but that would not be efficient.  Each restaurant should train their own cook robots and to make sure that each cook robot is skilled at their domain jobs. 

The idea behind the Universal artificial intelligence program is to create fully automated restaurants.  For example, in a fully automated Mcdonalds there are no human workers.  The cooks are robots, the clerks are robots, the manager is a robot, the delivery boy is a robot and the janitors are robots.  Each robot are individuals (just like human beings).  They work together in a team like environment, based on common rules and goals, to run the restaurant.  Different types of real human robots are selected to train each dummy robot.  A real human robot skilled in cooking will do the training for a cook robot, a real human robot skilled in managing a restaurant will do the training for a manager robot, etc. 

 

How to train a team of dummy robots to run an autonomous hospital

A doctor's job is very simple if you think about it.  Knowledge about medicine is massive and that's one of the reasons why doctors are needed to diagnosis a patient.  Diagnosing patients is very similar to cooking food.  The similarities between Doctors and cooks is that they do things over and over and over again.  Doctors would see a patient, analyze background knowledge about the patient, observe the illness, and use logic to identify the illness and to recommend a solution.  Doctors do this over and over again.  The same can be said about cooks because they cook the same foods over and over again. 

In order to train a dummy robot to serve as a robot doctor, the real human robot has to go to medical school and learn about medicine.  If there are too much knowledge to learn, the real human robot can learn the most important things and to use external references to remind the robot doctor of certain things.  For example, the robot doctor can search for specific info over the internet or read a medical book.  Thus, information regarding medicine can be accessed externally instead of learning everything and storing them in the specialized brain. 

For the common sense knowledge part, this robot doctor needs more than an elementary school education.  It might have to learn higher education like high school and some college. 

Next, the real human robot has to train the robot doctor (the dummy robot) in terms of diagnosing patients.  The real human robot will record all experiences of working as a doctor for many years and storing these experiences inside the specialized brain.  When enough training is had, the specialized brain will be inserted into a dummy robot and this dummy robot will be a robot doctor. 

Different doctors will be trained by different specialized real human robots.  An eye doctor is trained differently compared to a heart surgeon or a dentist.  However, each robot doctor has common knowledge about medicine.  For example, they all know how the human body works.  They know basic knowledge about every field in medicine, but they are skilled in only one specific field.  

Each robot doctor will be trained according to what their domain job is.  There is a scope of what each robot doctor can and can't do.  For example, if a patient asks an eye doctor to do open heart surgery, the eye doctor will refer the patient to a heart surgeon.                 

 

How to train a government to control multiple autonomous companies

If you look at the US budget for any given year, the majority of spending is on Medicare.  If the government succeeds in building a fully automated hospital and if the medical care of the automated hospital is equal to or better than a real hospital, then the government can cut down on Medicare spending.  In fact, the automated hospital was designed to give "free medical care" to any patient.  In other words, patients don't have to pay anything to the hospital.        

The goal of universal artificial intelligence is to create fully automated companies.  These automated companies include:  AI restaurants, AI super markets, AI malls, AI hospitals, AI farms, AI sewing factories, AI construction companies, AI taxi companies, AI airplane system, AI military, and so forth.  In order to coordinate and to best serve human beings, an artificial government has to be established.  The United States government spends years and years to pass one bill.  The artificial government can past bills in less than one second because they can work inside a virtual world. 

This artificial government requires many individual dummy robots that are skilled in politics.  Actually, this artificial government is a part of the US government and works in conjunction with the US government (which are controlled by human beings).  The artificial government's goals are to run autonomous companies and to make sure that laws are set up to protect and serve human beings.  It will make decisions based on the US constitution and in the best interest of the human race.  For example, if lots of patients are not getting good medical care from an autonomous hospital, the artificial government will shut down the company and demand that each robot doctor be retrained properly.  In another example, if an autonomous farm is producing fruits with e coli bacteria, it will be shut down immediately and new laws and procedures will be in place to make other autonomous farms safer.      

The real human robots think and act like a real human being.  We can harness their experiences and build virtually any type of robot.  I don't care if a job is retail, garbage collecting, cooking, or construction work.  As long as human beings are able to do a job, we can build specialized dummy robots to do that job. 

The real human robot has a physical body like a human being.  Eyes, arms and legs are required to do some jobs.  For example, police robots has to chase, subdue and handcuff criminals.  Soldier robots has to carry a real gun and shoot enemies.  By making a dummy robot that have similar physical parts as a human being, they can not only extract information like a human being, but they can also manipulate objects in our environment like a human being.      

 

How to train a dummy robot to play videogames

The universal AI was an idea I came up with in 1999.  I wanted to build a software program that would play mortal kombat.  I was also a fan of streetfighter and wanted to build a software to play that game.  An idea popped up in my head to design a software program that would play "any" videogame for "any" game console. 

Training a dummy robot to play videogames is the same as training a dummy robot to do any other human job.  First, the real human robot has to play a videogame based on commands from a user.  A user will give the real human robot commands on what it has to do.  For example, the user is a human being that is sitting next to the real human robot.  The user tells the real human robot to play mortal kombat and to play as the right character (the opponent).  The user on the other hand will play as the left character (the player).  In another case, the user can tell the real human robot to play as the left character (the player) and the opponent is the CPU. 

During the initial training, the real human robot already has basic knowledge from elementary school.  It knows that if the videogame says game over, it means that it has lost the game.  It knows from playing numerous games (like chase master, board games and sports games) that the objective is to beat the game.  For each game, common sense knowledge will tell the real human robot facts like:  "identify the character you control in the game", "identify the objectives of the game", and "identify basic rules of the game".    

The next step is to learn knowledge about a game.  This is accomplished by reading magazines and strategy guides.  In the game of mortal kombat, the player has to string together a series of combos in order to maximize the amount of damage to your opponent.  Either the real human robot can identify the objectives and rules of the game by using logic or it can identify the objectives and rules of the game by reading magazines (or someone can tell the real human robot the rules and objectives during the game).     

Once a specialized brain is created to play mortal kombat, the specialized brain will be inserted into a dummy robot.  A user will give commands to the dummy robot to play a videogame.  For example, the user might ask the dummy robot to play as the right character (the opponent), while the user plays as the left player (the player).  Yet in another example, the user can tell the dummy robot to play with the CPU. 

----  

When playing a game like basketball, the rules and objectives of basketball are learned in PE.  The knowledge used to play a real basketball game is carried over to playing a basketball game in a videogame.  Adaptive pathways are used by the real human robot to learn the controller.  It's brain uses adaptive behavior to identify which buttons control what objects and actions in the videogame.  In a real basketball game, the real human robot has to use his body to play the game.  However, in a videogame, the real human robot has to use his hands to push buttons on a controller.  Also, using a keyboard as the controller to play a videogame is different from using a playstation controller.  Common sense skills will tell the robot how to control characters in the videogame regardless of what the controllers look like.  Thus, the real human robot can play the same basketball game on any videogame console (playstation, xbox, PC, PSP, Wii, NES, gameboy, Genesis, SNES, iphone, ipad, desktop computers, laptops, etc.)        

This is one of the reasons why a robot car can drive any car regardless of the model type and year. 

You can train a dummy robot to play any videogame for any game console.  Many videogames are very basic and people from elementary school are able to past each game easily.  Games like contra, super mario, and ninja gaiden are all straight forward games that don't really need skills to past.  Other difficult games like Zelda, final fantasy, and metroid require more training on the part of the real human robot.  In these role playing games, the objectives and the rules are not given to the player.  They have to use logic to identify the objectives and rules.  Sometimes the objectives and rules can only be had through trial and error.  As stated earlier, strategy guides and magazines can be an alternative source of knowledge to beat role playing games.   

 

Training complex machines that require teams of human workers

A tank requires a team of humans in order to operate.  In order to build a team of dummy robots to operate a tank, "super intelligent robots" are needed to do the training.  FIG. 52  is a diagram depicting the data structure of an autonomous tank.  There are 4 dummy robots controlling the tank.  The captain is responsible for all decision making, the shooter shoots enemies, the driver drives the tank, and the intelligence officer extract information.   The time dilation between the 4 dummy robots can be different.  For example, the captain can operate at 1 second per frame, while the intelligence officer can operate at 1 nanosecond per frame.  When orders from the captain are given to the intelligence officer, he has to accomplish his task in the fastest time possible. 

Super intelligent robots are needed to train each dummy robot because the training require some tasks to be done inside a virtual world.  Also, each dummy robot works in different or adaptable time dilations.    

The captain is observing a user's input and trying to understand what the user wants the AI tank to do.  The user's input can be in any media.  It could be at least one of the following:  a video of the user or a text box or a fillable form, etc.  A sequence of commands are given to the captain by the user and the captain has to interpret the commands and give orders to his lower level workers.

The current pathway (AI tank's pathway), which stores the user's input and car activities, is processed by software and this software organizes activities in the car for the captain.  The captain's pathway stores 5 sense data and thoughts belonging to the captain.  The captain will view the software and analyze what he thinks is important; determine the order of tasks to execute; solve conflicting tasks; give tasks to lower level workers, and execute tasks linearly or simultaneously. 

Building a machine that requires thousands of human workers would be much harder to train.  Something complex like that would require hierarchical captains and a structured team of workers. The starships from Star wars is a good example to use.  Details about how to train the AI Starship, or AI tank, or AI military are described in my books and patent applications. 

 

<< notes

 

 

  

Home | Task Engine | Time Machine | Patents | Notes | Donation 

Copyright 2007 (All rights reserved)