humanAI.com

 
taskengine
timemachine
patents
notes
donation

 

 

 

                                 

            :                  << content                              Chapter 8

Prediction examples

The universal prediction algorithm is a computer program that can predict any event or solve any problem regardless of how complex it may be.  FIG. 4 is a diagram of one pathway in the AI time machine.  In order to create the universal prediction algorithm, many prediction problems have to be trained.  These prediction problems include:  predicting a football game, predicting one entire NFL season, predicting all stock prices for the Dow/Nasdaq for the next 10 years, predicting the weather on Earth for the next 10 years, predicting future events, predicting the existence of future human beings, predicting earthquakes for the next 10 years and so forth. 

FIG. 4

Pathways can also be trained to predict the past.  These prediction problems include:  solving one cold case, solving all cold cases in the United States, predicting past events, predicting distant past events, determining the authentication of one religion, determining the authentication of all religions, predicting the weather 20,000 years ago, creating a universal family tree for all life on Earth, and so forth.           

When and if the AI time machine is trained adequately and it is able to predict most events in the past and future, we can safely say that the AI time machine is the universal prediction algorithm.  Every prediction made by the universal prediction algorithm (UPA) will be accomplished in the fastest time possible.  The UPA will also predict events in a hierarchical manner.  This means that the predictions go from general to specific.  The more time that passes the more accurate the predictions will be.  It will reach a point where each past/future event is predicted 100 percent accurately.

The main reason I call this technology:  universal prediction algorithm is because it can predict any event for the past or future, regardless of how complex it may be.  Current prediction algorithms are fixed and each event to predict uses a different algorithm.  There exist an algorithm to determine which banks are at risk of being robbed, there exist an algorithm to determine weather patterns, there exist an algorithm to predict football games, there exist an algorithm to predict suspects in a crime, there exist an algorithm to predict the migration patterns of flocks, and so forth.  Thus, there are fixed algorithms for every situation. 

Another disadvantage is that these fixed algorithms have a fixed output and the output is a possibility of a prediction.  For example, an algorithm to predict the results of a football game can only give an estimate prediction; it can never give an exact prediction (100 percent accurate).  Since the algorithm is fixed it will always give an approximate prediction. 

My universal prediction algorithm uses a “universal” algorithm that morphs and changes to make the prediction more accurate as time passes.  It will reach a point where the prediction will be 100 percent accurate.  As time passes, more data is inputted into the prediction internet concerning a prediction.  For example, if the task of the UPA is to solve a cold case that happened in 1946, the UPA will continue to accumulate knowledge about the cold case as time passes.  The computer program will not stop until the cold case is solved.  The output from the UPA is a report that describes the criminal, all the evidence that points to him/her, and an exact frame-by-frame video of what happened during the crime. 

The UPA can be used to solve all cold cases.  This means that the UPA will not stop until all cold cases in the FBI files are solved.  The universal prediction algorithm will accomplish this task in the fastest; and most efficient way possible.  In other words, minimal work is needed to accomplish this task.

By the way, solving all cold cases is a part of an on-going effort to predict all events in the past and future of planet Earth. 

An initial prediction tree is created for a given prediction problem.  Think of the software program in each predicted model as an algorithm.  For each predicted model, teams of virtual characters are required to modify the algorithm’s inputs and outputs.  If you observe the entire prediction tree, the “universal prediction algorithm” represents the interconnected software programs between all hierarchically structured predicted models. 

 

Predicting stock prices for the Dow and Nasdaq for the next 10 years

I believe that predicting the stock market is the most difficult problem the universal prediction algorithm will do.  If we analyze all the objects (large or small) involved in the daily activities of the stock market, you will be overwhelmed.  FIG. 32 is a diagram depicting the most important objects involved in one stock company.  The revenue of the company is one of the most important aspects that determine its stock price, so the revenue of the company has top priority.  The news announcements from the company are also an important aspect.

 

Other objects involved in the company’s stock price include individual investors, societies reaction to the companies news, and the network that allows users to buy or sell stocks.  All these individual objects (or aspects) are considered in order to predict the future prices of the stock for the next 10 years. 

As usual, human beings create future events, so they are considered important objects.  The stock company has many employees, executives and partners.  These human beings involved will be prioritized and they will be predicted based on their importance. 

All activities of the company like meetings, news announcements, imagination, business interruptions, business deals and partnerships, production, sales, product manufacturing and so forth has to be predicted hierarchically.  Every activity has to be predicted as a group and not as individuals. 

Another important object is stock owners.  All stock owners and potential stock owners have to be predicted.  If you break down an individual stock owner into elements, you will get:  1.  the user.  2.  a computer.  The stock owner is a user that is controlling a computer to buy and sell stocks.  Within the user, important objects will include:  1.  brain.  2.  physical body.  The user’s brain is very important because it determines if this person will sell stocks or buy stocks.  By analyzing his brain, we can understand the rational of what triggers a stock activity.

Predicting individual users controlling a computer has been described in my previous books, so I won’t go into the details of how this prediction method works. 

All company stock owners and potential company stock owners have to be analyzed and predicted.  Each stock owner has to be predicted along with other important objects such as the company’s announcements, company’s revenues, and society’s reactions to the company. 

The prediction tree will be extremely long for this type of prediction.  Zillions and zillions of virtual characters have to be assigned to certain predicted models and all teams have to work together on the prediction internet to predict the stock prices for this one company for the next 10 years.

On the other hand, the above example only deals with one company stock, the Dow and Nasdaq has thousands of stocks to choose from.  In order to predict the entire stock market, all company stocks are ranked hierarchically and they are predicted by teams of virtual characters based on how important they are.  For example, Wal-mart is a stock that lots of people own, so it’s considered a very important object.  Bank of Hawaii is a stock that only a few people own, so it’s considered a very minor object (FIG. 33).

FIG. 33

The network software in the stock exchange to calculate trading prices has to be predicted as well.  In early 2010, the stock market encountered a computer glitch or software flaw that caused a world wide panic.  The Dow Jones dropped 1000 points in less than 10 minutes.  Within the 10 minutes stock owners tried to sell their stocks.  These stock owners didn’t realize that the Dow Jones dropped so quickly not because people were selling stocks, but because the network software encountered a rare glitch.  Because of the network software, stock prices changed in dramatic ways.  This is one reason why the network that calculates stock prices must be predicted in conjunction with other prediction objects.

Predicting individual computers and network software has been described in previous books so I won’t be going into the details of how they work.

Individual stocks are not isolated from other stocks.  In fact, prices of one stock are directly dependent upon its sector and industry.  Even the price of the Dow Jones affect all stock listings, including the stock listings in the Nasdaq.  When Intel reported its earnings several months ago, it dropped 10 percent in one day.  This report also affected its sector (chip company) and its industry (computer).  Thus, it is important to do predictions in a hierarchical-uniform manner.

In some ways, in order to get a perfect prediction of the stock market, every object on Earth, ranging from a human being to an individual atom, must be predicted uniformly.  Future events are interconnected in a web.  This makes future prediction a very difficult task. The prediction tree exists so that predictions are done based on hierarchical priority.  Past events are also locked in an interconnected web and predicting events in the past is very easy. 

Extremely complex prediction tasks like predicting the stock market require that the initial prediction tree outlines several individual parts.  The initial prediction tree might have general predicted models that link the three individual parts, but not detailed predicted models.  While the virtual characters are working on each part, their parent predicted models are created during runtime.  These added parent predicted models are dependant on the work results from the virtual characters.   

 

Predicting an entire NFL football season

There exists fixed algorithms that can predict who will win the Superbowl.  These are fixed algorithms and they can only predict an estimation of who might win.  They can never predict exactly which team will win the Superbowl or the details of each tournament game.

The universal prediction algorithm is different because the output from the prediction isn’t fixed.  The UPA will continue to output better and better predictions as time passes.  The more time given to the UPA, the more accurate the prediction becomes.

Hypothetically, let’s say that the virtual characters have to predict the entire NFL season “before” any games are played.  The virtual characters are given a lineup of dates on the initial tournament games.  Based on this single fact sheet, the virtual characters have to predict how the tournament games will play out.  They also have to predict the Superbowl and what the outcome of that game will be. 

The first thing the virtual characters will do is gather as much information about what they are assigned to predict.  If a team of virtual characters are responsible for predicting the Cowboy vs Steeler’s game, then the virtual characters has to gather as much information about recent player information on these two teams.  Information that is extracted from individual players will include:  player stamina, weakness, strength, performance and statistics. 

Every prediction they make will be based on assumption and most likely these predictions can only serve as general predictions.  For example, they will compare teams and guess who is stronger.  If there is one strong team in the league and they repeatedly show that they are undefeated, and there is another team that is weak, then the VCs can conclude without a doubt that the strong team will win.  We see this behavior over and over again in sport games.  The USA basketball team always wins the Olympic basketball game because they have proven their abilities.  Football teams are no different. 

One factor these virtual characters will look at will be star players in each team.  If there are two teams that are equal in performance, but team1 is missing a star player, then most people will assume that team2 will win.  Another method to compare team strength is by looking at how star players work in a group. 

Another method they might use for their predictions is to simulate each players’ physical body.  These virtual characters have to predict every gameplay incrementally for each game.  They have to try to predict what each player will sense, think and act during each gameplay.  All these predictions are assumptions and are most likely useless information (in other words, these predictions will never be 100 percent accurate).

The above method works to give an estimated prediction of the Superbowl.  To get a perfect prediction will require every object on planet Earth (large or small) to be predicted hierarchically and uniformly.  The virtual characters have to know current information.  They have to predict the game at the start of the game and not before the game.  This way, they know which players are present and which players are missing.  Also, they need to know the physical atoms of each player, currently, in order to predict that players’ future.  A small injury to a player has profound effects in his performance during a game.                       

Thus, the conclusion is that if the universal prediction algorithm wants to predict a perfect future timeline of an NFL season, it has to predict all objects on planet Earth.  Of course, the most important objects that relate to football are predicted first before predictions are made on non-related objects.  For example, each player in the NFL will have his future timeline predicted every fraction of a nanosecond, in terms of what they are sensing from the environment, thoughts, and physical actions.  Any object they encounter in terms of 5 senses or thoughts must also be predicted.  For example, if a player goes to a restaurant, all objects related to the restaurant will be predicted, including:  the people there, the food, the cooks, the hostess, and the furniture.  If a player is talking on a cellphone with another person half way around the world, the virtual characters have to predict this person as well.     

These minor events are important to predicting the Superbowl because they affect the players.  A star quarterback might go to a restaurant one day and he trips on the stairs and breaks his leg.  This injury will prevent him from playing in tomorrow’s game.  Thus, there are no short cuts in predicting the future.  All dependant future events must be predicted in a hierarchical and uniform manner.    

 

Sequential tasks for the AI time machine (in training mode)

In the football example, the user types out one task into the AI time machine and the AI time machine will output one desired output.  When dealing with a sequence of tasks, the AI time machine has to remember past events, manage tasks from the user, determine if tasks should be executed and so forth.  Essentially, the AI time machine is trying to manage multiple tasks for the user (like an operating system).

FIG. 4 is a diagram depicting a pathway from the AI time machine.  The robot pathway represents the user and the virtual characters represent the work done to generate desired outputs.  A dynamic robot is a robot that has a built in virtual world.  He is called the robot in the virtual world and he is called a virtual character in the time machine world.  The robot in the virtual world is one entity and he has goals and rules.  On the other hand, the virtual character/s is also another entity, but has the same goals and rules as the robot in the virtual world. 

FIG. 4

The robot in the virtual world will assign the fixed interface functions and the linear inputs (he is pretending to be a user).  The captain virtual character’s job is to analyze the user’s inputs, to manage multiple tasks and to execute tasks.  The captain executes tasks based on using external technologies (like the AI time machine) or to give tasks to lower level workers.   

FIGS. 34A-34C are diagrams depicting sequential inputs/desired outputs from the AI time machine.  These diagrams were taken from my 2008 book, called:  AI time machine:  book12. 

FIG. 34A

The top level are inputs from a user and the bottom level are desired outputs from teams of virtual characters.  In the bottom level, the captain is the leader of the team of virtual characters and he is the operator.  When the first task is given “restore picture23 and concentrate on the center brown object”, the captain will use the AI time machine to accomplish this task.  In the second task, “what are those red shapes in the forest”, the captain doesn’t have any investigative tools to accomplish this task so he orders a specialists in analyzing images to do the task.  The image specialist can output an explanation to the user.  Next, the user gives a third task, “calculate what these lifeforms are and give facts about them”, this task will be directed to the specialists and the specialists is using the AI time machine to process the task.  The AI time machine might output a short summary of the lifeforms.  The specialist will read the summary and output 2 sentences to the user, explaining what the lifeforms are and facts about the lifeform. 

This example shows that a captain is managing the tasks given by the user.  He can either use technology (like the AI time machine) to process the tasks or to give it to lower level workers to do the work.  If the task is simple, the captain might do the task manually.

The captain is responsible for directing certain tasks to experts in accomplishing these tasks.  For example, if the user’s inputs are questions about medical information on the brain, the captain has to reroute these tasks to a doctor.  This doctor isn’t just any doctor, he has to be a neurologist who is an expert in understanding how the brain works.  Most of the time, the captain will manage basic tasks, such as opening emails, calling family members to send them a message, opening up digital files and modifying them, doing simple search over the internet, searching for definitions to words, summarizing a book, analyzing and explaining a digital file and so forth.

Specialized dynamic robots can be used to train the AI time machine in certain fields.  For example, the dynamic robot is a medical doctor and he is training the AI time machine to answer questions from a user about general information on medicine.  Dynamic robots specialized in neurology can train the AI time machine to answer sequential questions about how the brain works.  Dynamic robots who are computer scientists can be used to train the AI time machine to do tasks requiring the writing of software programs.  The user might ask the AI time machine to write a database system. 

FIG. 34B and 34C are additional examples of the inputs/outputs communications between the robot in the virtual world (the user) and the teams of virtual characters in the time machine world (workers). 

In FIG. 34B, the user wants the AI time machine to write a comic book and in FIG. 34C, the user wants the AI time machine to make a movie.  The teams of virtual characters to accomplish these complex tasks require experts.  If you ask a doctor to make a movie, he/she won’t be able to accomplish the task.  Thus, the teams of virtual characters are experts in their fields and they will be trained based on their specialized tasks.

FIG. 34B

 

FIG. 34C

  

Referring to FIG. 35, the universal brain stores pathways from multiple dynamic robots.  A dynamic robot has two types of pathways (virtual world pathway and time machine world pathway).  The AI time machine will usually extract pathways from the universal brain based on the interface communication between the user (the robot) and the virtual characters.  In other words, the input and the desired outputs between the robot’s pathway and the virtual character pathways is the primary objects that will determine what pathways the AI time machine will extract from the universal brain.

The way the AI time machine extracts pathways from the universal brain is very similar to how a human robot extracts pathways from its brain.  If the user asks a question about Hamlet, the AI time machine finds the best match to the current pathway in the universal brain.  The important objects in the current pathway are the inputs from the user.  The best pathway match will contain the optimal way the question is answered. 

In terms of accomplishing tasks, the AI time machine extracts pathways from the universal brain that matches to the user’s task input.  The best pathway match will contain the virtual character pathways to accomplish the user’s task in an optimal way.       

The data in the current pathway can be arbitrary.  The current pathway can be a fabricated pathway based on what a user is sensing and thinking from the environment.  For example, the current pathway can be the linear thoughts of the user and the 5 senses of the user interacting with a computer (the computer is the AI time machine).  Things that the user sees on the computer screen are part of the visual data of the current pathway. 

The current pathway can be a camera that is observing a user in terms of what he is doing on the computer.  The AI of the camera is predicting what the user is thinking and doing on the computer.  The AI will try to predict where the user is focusing on on the computer system.  The data on the computer in terms of user activities can also be part of the current pathway, such as mouse movements or keyboard presses. 

The current pathway should be the thoughts and the 5 senses of the user; and the activities of the computer the user is controlling.  

Regardless of what data types are contained in the current pathway, the AI time machine will match this information to the robot pathways in the virtual world brain.  The important objects in the current pathway are usually the user’s inputs into the computer.  The best robot pathway match will be associated with virtual character pathways.  The work done by the virtual character pathways will represent the AI of the AI time machine.

The extraction of pathways from the universal brain is based on dependability.  If a captain has 4 lower level workers and it takes all these workers to accomplish task2, then when the user inputs task2 into the AI time machine, the captain’s pathways regarding task2 will be extracted.  The captain’s pathways for task2 will also extract the 4 lower level workers’ pathways regarding their jobs of accomplishing task2.  (Note:  each virtual character in a hierarchical team can use the AI time machine.  This means that work from different virtual characters or teams can be encapsulated.)

The pathways in the universal brain will self-organize with similar pathways (very similar to how human robots work).  These pathways will form universal pathways that will be able to manage, process, and execute any input task from a user.  It doesn’t matter what the user says or does or orders, the AI of the AI time machine is able to respond with desired outputs under any circumstances.  

 

The captain analyzes the user’s activities

The captain has human intelligence and knows what the user’s goals are for each task.  For example, if the input task is “open the lion image”, the captain knows that the lion image was opened 2 hours ago and he can recall what image the user is referring to.  The captain uses human intelligence to spot out what the real intentions of the user are.  If the user types out an ambiguous task, such as:  “drawing image bird colored children pictures”, the captain can analyze this input and determine that the user wants to search for colored drawings of birds made by children.  In other cases, the input task might be misspelled and the captain has to use human intelligence to correct the misspelling.  Thus, the captain is responsible for analyzing input tasks from the user and to derive meaning from them.

In other cases, the inputs from the user are not enough to understand the user’s goals.  For example, if the input task is “look for images over the internet related to arrows”, the captain won’t know specifically what kind of arrows to look for or what type of media the arrow images should be in.  The captain can observe past videos of the user on the computer.  The captain finds out that the user was reading the rules to making patent drawings.  This revelation tells the captain that the user is searching for black and white images of an arrow.  Patent rules are followed so that the captain will find the best black and white images of arrows over the internet. 

This example shows that the captain can spy on the user to understand the user’s goals and rules when inputting tasks into the AI time machine.  These spying techniques include observing camera videos of the user before the task was given or analyzing and processing background information about the user. 

However, most tasks done by the AI time machine are based on the captain analyzing sequential input tasks from the user.

 

Review:

Only dynamic robots are able to train the AI time machine (human beings or expert software programs can’t train the AI time machine).  The dynamic robot comprises a robot in the virtual world and a virtual character/s in the time machine world.  The robot in the virtual world has to act as the user, inputting tasks and critiquing about the desired outputs.  On the other hand, the virtual characters in the time machine world have to accomplish user tasks by either using external technology (like the AI time machine) or manually accomplish user tasks in a team like environment.

The dynamic robots have to train the AI time machine with individual tasks first.  Then it has to train the AI time machine to manage multiple tasks by having a captain (a virtual character) manage, process and execute tasks.

          

<< content               next chapter >>

 

 

  

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

Copyright 2007 (All rights reserved)