<< content
Chapter 8
Decision Making and Planning Tasks
There are many different levels on decision making and each level influences
the way the robot makes decisions. Below is an outline of the level of
factors that will influence the robot (the AI program) in terms of decision
making.
Levels of Decision Making:
1. innate reflexes based on pain--when a person is in
great pain reflexes are most likely to trigger. These reflexes are wired
into pain so that when the pain reaches a certain point it triggers the
reflex. No conscious decision was needed to trigger this action. Some of
these innate reflexes are: when a person is in great pain he/she yells out
loud, when the knee cap is hit with a hammer the leg moves automatically,
etc.
2. Learned decisions based on past knowledge--the
conscious guides the robot to make decisions. These decisions are either
based on future predictions or logical decision making.
3. Pain and pleasure built into the
robot--attractiveness or ugliness, physical pain (degree of pain) and
physical pleasure (degree of pleasure) are factors that the robot uses to
make decisions. Is the robot going to eat lobster for dinner (the robot
loves lobsters) or is the robot going to eat rice? (the robot eats rice only
if he has to). These pain/pleasure factors that are built into the robot
will make decisions.
4. Daily routine--Learned things that the robot was
thought everyday by teachers are factors in decision making. Some of these
daily routines are so natural that the robot doesn't need to make a decision
to do them. Daily routines such as: waking up in the morning, brushing your
teeth, eating 3 meals a day based on the time, going to sleep at night,
using the bathroom when you need to go, and going to work or school on
weekdays.
These are the levels of decision making. The higher levels overshadow the
lower levels in terms of decision making. For example, innate pain
overshadows learned decisions because innate pain is a reflex and is
triggered by pain while learned decisions uses conscious thoughts to make
decision. In other words, innate pain is triggered without conscious thought
and overshadows learned decisions.
Another example is learned decisions can overshadow pain and pleasure. This
form of pain and pleasure doesn't trigger reflexes, it's just a lower degree
of pain/pleasure from innate reflexes--a degree where the robot can manage
the pain. If a person has an itch on his butt and this person is walking on
the street, the person can make a decision to scratch his butt or not. He
can wait until he gets to a private area before scratching his butt. This is
one demonstration of learned decision having higher priority than
pain/pleasure. Even something like using the bathroom require learned
decisions. If you have to go the pain is unbearable. However, you can't take
a dump on the street or in a classroom. You have to make a decision to go to
the bathroom and take a dump. Even though the pain is so great the learned
decisions guided the robot to take the appropriate actions.
Pain and pleasure is another factor that is used for hidden objects. The AI
finds these patterns and wire pathways with pain and pleasure. The strongest
pathways have their powerpts strengthened because it's wired to pleasure and
the weak pathways have their powerpts lowered because it's wired to pain.
The learned decision encapsulates pain/pleasure to plan out tasks and make
decisions for the robot. The main function of decision making is always to
pursue pathways that lead to pleasure and stay away from pathways that lead
to pain.
Daily routines such as brushing your teeth, sleeping at 9 pm, and waking up
at 7 am are just things that we learn everyday and this type of learning is
so normal that we do them without thinking. Learned decision can overshadow
these things because we can control when we sleep by conscious thought.
Instead of 9 pm we can sleep at 2 am. Instead of eating cereal for breakfast
we can eat a hamburger. This daily routine is also another factor that can
be encapsulated into learned decisions to plan out tasks and make decisions
for the robot. The AI program finds patterns concerning daily routines and
use this pattern in a hidden object. This hidden object will then be
assigned to words/sentences as meaning of words/sentences.
In this chapter we will concentrate on learned
decisions. Learned decisions are lessons that are taught to the robot by
teachers. The teachers teach the robot how to make decisions. Many
decision making lessons will be taught to the robot. When the robot is in
the supermarket to buy food he has to decide what foods to buy. When the
robot has free time, he has to decide what he wants to do. When the robot
has extra money in the bank, he has to decide what to buy and for whom.
Thus, for each situation the decision making process is
different. In order to create a universal pathway to make a decision for
the current environment the robot must be taught decision making under many
situations.
Computer programs inside the conscious
The robot will sense data from the environment through
its 5 senses: sight, sound, taste, touch and smell. Target objects are
objects the robot recognizes from the environment. The brain of the robot
will extract element objects that have strong association with their
respective target objects. All element objects from all target objects will
compete with one another to be activated in the mind. These activated
thoughts are known as the conscious of the robot.
There are two types of conscious thoughts: open
activation and hidden activation. The open activations are element objects
that are presented to the conscious and the robot is aware of the data. On
the other hand, hidden activations are element objects that are not
presented to the conscious and the robot is unaware of the data. Extremely
complex tasks in memory, such as problem solving, will require both open and
hidden activation.
Think of the conscious as a “memory bubble” that takes in and modify
data; and the modified data instruct the robot to take action (FIG. 42A -
42B). Inside the conscious is a computer program (or a series of
interconnected computer programs) that manipulate data in the conscious.
The intelligent pathways extracted from memory and used by the robot
generate this computer program.
FIG. 42A
The intelligent pathways control the data in the
conscious by adding, deleting and manipulating data. They also control the
searching and extracting of data from memory. The instructions inside the
intelligent pathways will have search functions that will extract relevant
data, based on a situation, and put them into the conscious. Some data will
be extracted from memory, while other data will be extracted from long-term
memory.
There are also other things the intelligent pathways in
memory can do. They can control the comparing of data; or analyze two or
more data from the conscious. These intelligent pathways can even predict
the future and generate a computer program to output future events of what
might happen in the future based on the current environment.
The conscious actually does many other things for the
robot. Some of these things include: giving knowledge about an object,
providing meaning to language, solving problems, answering questions,
following orders, planning tasks, solving interruptions of tasks, managing
multiple tasks at that same time, following the rules of a game, coming up
with imagination and so forth. The intelligent pathways in memory allow the
conscious of the robot to do all these things (and more).
Important data in the conscious
The conscious does many different things for the
robot. Four of the most important things the conscious does are: 1.
manage tasks. 2. establish rules to follow, based on the tasks. 3.
solving problems and planning steps to achieve tasks. 4. know identity.
Referring to FIG. 42B, there are four containers the computer program in the
conscious generated as a result of intelligent pathways: the task
container, the rules container, the planning container and the identity
container. All data from all four containers influence each other one way
or another. For example, the rules will influence what tasks to
follow/abort and the planning information will influence what rules to
follow or what tasks to do.
These containers are just temporary caches inside the
conscious that was generated by intelligent pathways in memory. Based on
the current environment, the robot selects an optimal pathway from memory
and that optimal pathway has instructions to create containers so that
groups of data could be manipulated and logical thoughts and actions can be
had by the robot. The intelligent pathways create any type of computer
program or discrete mathematical functions to manipulate data in the
conscious -- a database system, an operating system to manage multiple
threads, a word processor, an image processor, a search engine, a state
machine or any software program.
Managing multiple tasks
The robot’s brain has a task container that manages
tasks. When he wakes up in the morning, knowledge will pour into the
robot’s conscious in terms of 5 sense data. Most likely sentences will pour
into the task container. Natural tasks that the robot does everyday will be
the first things to populate the container. Natural tasks will include:
brush your teeth in the morning, eat breakfast, eat 3 meals a day, take a
shower at night, go to work at 8:30 am, leave work at 5:00 pm and so forth.
Previous tasks that the robot decided to do will also
be included in the task container such as decisions made yesterday or last
week or last year. For example, if the robot decided 2 weeks ago that he
was going on vacation next year December, that task might be included in the
task container. If the robot decided yesterday that he was going to eat
hamburger for breakfast the very next day (which is today), then the task of
eating hamburger for breakfast will be in the task container.
Tasks in the task container can be ongoing as well, if
the robot was a student in school, then his primary task in the task
container will be to go to school at 7:30 am every morning. If the robot
was a business man, then his primary task in the task container will be to
go to work at 8:30 am every morning. The tasks in the task container are
not fixed and are changeable. For example, if the robot graduates from
school and found a job at a local business, then every day that he wakes up
the task of going to work at 8:30 am will be there.
Most of the task in the task container is “at the
moment” and the robot does a task, then move on to other tasks and so forth.
Future tasks are remembered in memory and it’s the job of the robot’s
conscious to remind the robot what tasks to do at certain times. This robot
forgets things so sometimes, he forget to do tasks. Sentence commands can
also define a task to do at certain times or certain places. For example,
if a friend says, “call me when you get home from work”, the robot will
remind himself when he gets home to activate the thought: “call the boss”.
The sentence has a pattern, which is: when you recognize your home,
activate this thought: “call the boss”.
The robot can be fickle minded as well. The robot
might decide to go to Panda express for lunch, but when he gets there he
decides to go to Burger King, next door instead.
Deciding on tasks, like I said before, is based on at
the moment. If the robot is a student in a college studying computer
science, and the professor gave him a software assignment, then the robot
will plan when he will write the software program. The professor might give
the students 2 weeks to finish the software program. The robot knows that
he has lots of time so he decides to write the software program tomorrow.
If the robot is writing the software program and he has to sleep, he will
plan the task of continuing to write the software program tomorrow in his
free-time. If the robot is writing the software program for 5 straight
hours and he is tired, he will decide to take a 1-hour break, before
continuing his work. Within the 1-hour break, the robot will use logic to
find something to do to relax, such as deciding on watching a movie or
playing videogames.
In other cases, the robot might be in trouble and he
has until tomorrow to finish the software program. He will decide to stay
up all night to finish the software program. This task will defeat any
pre-existing task in the task container, such as sleeping at 9 pm or eating
a sandwich before sleeping.
Rules in the rules container
When the robot gets up each day, knowledge in terms of
rules will pour into the rules container. These rules dictate the
boundaries of certain behavior. In the identity container, the robot has
many rules that it follows. These rules include: likes, dislikes, goals,
limits, how to do things, what rules that must be followed while doing tasks
and so forth.
Since there are soo many rules in life, the robot has
to follow rules at the moment. That’s where the intelligent pathways are
needed. When the robot predicts the future, there are actually “event
pools” stored at certain times when events happen. In the event pool are
rules ranked in order that activates in the robot’s conscious. The computer
program in the robot’s brain will manage these rules and determine what
rules to activate at certain times.
If the robot was playing baseball, and he is batting,
the rules of conduct (regardless of how complex they may be) will be
populating the rules container. The computer program will manage these
rules and activate the most relevant for that given situation. Most of
these rules are just sentences that have discrete math properties such as
if-then statement, for-loops, while-loops, functions, encapsulated rules and
so forth. The most famous is the if-then statements. If the robot
recognizes this situation then activate these set of rules.
Finally, the task container and the rules container go
hand in hand. The robot accomplishes tasks based on the rules. These are
the rules that you must follow in order to accomplish this task. These set
of rules are needed in order to solve a math equation. These set of rules
are needed in order to play chess. The rules are boundaries set that the
robot must follow in order to accomplish his objectives.
Deciding to do a task
The decision maker is ultimately the robot and he
controls what decisions to make. The intelligent pathways in memory make
decisions and extract relevant data from memory to make decisions. Because
of this, the robot must have information about its identity (FIG. 43). In
the identity container, there exists information that is important to the
robot. Things like dislikes and likes are located there.
Depending on the situation, the data in the identity
container varies. If the robot is shopping for groceries at the local
supermarket, then likes and dislikes of food will be stored there. If the
robot was in a clothing department, then likes and dislikes of clothes will
be there. When the robot consciously decides to buy groceries from the
supermarket, information about food and possible purchases pour into the
robot’s conscious like a facet valve. The more he thinks about buying
groceries the more information is extracted from memory.
FIG. 43
The robot is alive because of all the containers in the
conscious. The robot can control what tasks and how to carry out each
task. The robot controls the rules that it must follow in order to fulfill
the tasks. The identity container contains the likes and dislikes of the
robot and what its values are when making decisions. The robot is
self-aware because of the control of itself. External objects like people
and the environment influences the robot to make decisions, but it’s the
robot that ultimately makes the decisions.
The identity container also establishes relational
links with other currently encountered objects. Family members are
established and friends/foes are established. The things that the robot
possesses are known in the identity container. Things like the robot has 2
arms, 2 legs, a brain, two eyes, 1 mouth, 2 ears and so forth. He knows
that the visual images coming into the eye belongs to him and the things
that he hears from the environment come from his ears. The robot knows that
he can control most of the physical aspects of its body, such as focusing
the eyes on desired areas, focusing the ears to hear certain sound,
thinking, moving its body and so forth.
If the robot has money in his pocket, he knows that the
money belongs to him. Possession is a learned thing by teachers. Teachers
will use sentences to associate how one object can possess another object.
If the robot pays money for a domain name (which is a non-physical object),
the robot knows that he possesses the domain name. The sentence best
describe possession: “if the robot pays for the domain name, then the
domain name belongs to the robot”. The word belong is a word that has
hidden values that mean possession.
The robot is taught at a very young age common sense
knowledge and what decisions benefit itself. Simple decision making are
taught to the robot at its early stages. A lesson like selecting between
two items based on price is taught. The teacher gives the robot two
identical ice cream and asks the robot to select one. One cost 2 dollars
and the other cost 8 dollars. Using common sense, the robot will determine
that both ice creams are identical, but one is more expensive than the
other. The teacher forces the robot to select the cheaper ice cream.
The teachers are there to supervise the learning by
guiding the robot with sentences. These sentences tell the robot what is
right and what is wrong. If the robot selects the 8 dollar ice cream, the
teacher will ask the robot why. At that point, he/she will say: “that is
wrong, you have to select the cheaper ice cream”. The teacher will also
explain why. The teacher will repeat the lesson until the robot selects the
correct ice cream and makes the right decision.
Learning will be gradual, whereby the teacher will
teach more complex decision making examples. For example, the teacher might
display 5 different chilli can and the prices for each can are different.
The robot has to select one of the 5 chilli cans. The prices have to be
compared, the presentation of the chilli cans have to be compared, the
amount of money the robot currently has, and the brands of the chilli have
to be compared.
In the identity container contains a very important
fact: “the robot is allergic to onions”. If 1 of the chilli cans has
onions in it, the robot should rule that one out. The robot doesn’t care if
the chilli can is cheap or the name-brand is well known. The fact that he
can’t eat the chilli tells the robot that he won’t be buying that chilli
can. Thus, the teacher will teach the robot to rule out possible chilli
cans. Next, the teacher will teach the robot to compare and narrow down the
choices to 2 or 3. In order to do this, a sophisticated form of logic is
required to compare prices and name-brands. Facts about the 4 remaining
chilli cans are listed below.
1. $2.50 2. $10.95 3.
$3.95 4. $1.99
well-known best brand
well-known unknown
The robot will know that he currently has 5 dollars in
his pocket. Chilli can 2 will be ruled out because he can’t afford it. The
next step is to compare the other 3 chilli cans in terms of price and
name-brand. The robot likes the $3.95 choice, but only slight above the
other two choices. The robot likes the $1.99 choice because its cheap, but
dislikes the taste of the chilli. After an elaborate debate, the robot
decided to buy the $2.50 because it is considered not too cheap and not too
expensive. He also likes the taste of the chilli.
Decision making is a personal thing, but the teacher is
there to help the robot make the best decision and to guide him in the
decision making process. If I were to select from the 4 choices above, I
would probable pick the $1.99 choice because I like to buy cheap things.
Buying cheap is one of the rules that I follow when I buy anything.
In another example, the teacher can put 10 fruits on a
desk and ask the robot to select one fruit. In the robot’s memory all 10
fruits are stored there and each fruit has powerpoints to determine its
desirability. The decision at this point is to search for each fruit in
memory and compare their powerpoints. The robot might be taught to select 3
fruit out of the 10 quickly and then compare the 3 fruit in terms of
different aspects to ultimately decide on 1 fruit. When the robot does
select 3 fruits, he will have to compare aspects of the fruits such as size
and taste. Let’s say the robot selects a banana, an apple and an orange.
The banana is quick to eat and taste good. The apple last longer, but it
doesn’t taste good. The orange last long and it taste good. Based on these
three logical thoughts, the robot will determine which of the 3 will benefit
itself. It’s obvious from the 3 logical thoughts that the orange is a good
choice because it taste good and it last long.
Thus, decision making is simply generating facts about
each object and selecting among these facts. Comparing facts can be used as
well, such as: “this chilli can is larger in size than that chilli can”,
“this chilli can’s brand-name is popular and the price is cheap”, “this
chilli can is expensive, but the chilli is very good”. Based on these
logical facts generated by the items for selection, the robot has to
determine which item is the best. In some cases, a decision is a debate in
the robot’s brain.
Another way that decision making is learned is by doing
grade school worksheets. Decision making worksheets are given to the robot
and he has to accomplish the assignment. The robot will be graded on how
well it makes decisions. If the robot has a good grade, that means he made
the right choices. On the other token, if the robot has a bad grade, that
means he made terrible choices. The teacher will correct the errors in
front of the robot so that he can see what mistakes he made.
Yet another way of learning decision making is by
watching TV shows like the price is right. People are put on the stand to
make decisions. Some decisions are based on logic while others are based on
random selection. If the robot watches 10 scenes and in each scene the task
was to select from 3 random doors, the robot will copy the average behavior
of the contestants. For example, 7 out of 10 times the contestants selected
the 2nd door, then the next time the robot has to select a random
door, he will select the 2nd door.
FIG. 44 is a diagram showing input/output again. The
robot has to find out the pattern between the input and the output. In this
case, the input is the question and the output is the selection. When the
robot watches people decide on things, he tends to mimic their behavior.
FIG. 44
Decision making can also be corrected by pain and
pleasure. What if the robot buys a chilli can from the supermarket and
found out that the chilli had tiny onions in it? The robot will remember
the brand and never buy from this brand again. The pain of finding out the
chilli he bought had onions in it caused the robot to put negative points on
this type of chilli.
Tasks to do and rules to follow
When the robot makes a decision, that task will be
inserted into the task container (FIG. 43). If the robot decides to eat
lunch at Subway, it will insert instructions, via sentences, into the task
container. The computer program inside the conscious keeps track of the
tasks and remind the robot when to do these tasks. The computer program
generates an operating system to manage tasks. Things like listing linear
tasks, solving interruption of tasks, rearranging tasks, executing multiple
simultaneous tasks, reminding when to do tasks, erasing tasks, modifying
tasks, managing hierarchical tasks are all done by the computer program
inside the robot’s conscious.
Distant tasks can also be inserted into the task
container. If the robot decides one day that he will take a vacation next
week Monday, then he will store that task in the task container. Tasks in
the task container are not fixed and can be changeable. For example, if
Monday comes and the robot is too busy, he can procrastinate and say that he
will take the vacation next month.
Tasks can come from both external commands or internal
commands. A teacher telling the robot to do a task is considered external
command and logical thoughts to do tasks is considered internal commands.
Playing chess example
Traditional chess AI programs rely on generating next
step moves and ultimately selecting one move to play the game. A human
being doesn’t play chess like an AI chess program. In fact, the reason the
best chess player in the world lost to a computer is because the AI chess
program was cheating. The human being can only see 2-3 steps into the
future. The AI chess program can see 300 steps into the future for all
pieces.
I will explain how the robot plays chess using human
level artificial intelligence. The reader can compare my methods with the
traditional AI chess programs and note the dissimilarities.
FIG. 45 is a diagram depicting the intelligent pathway
that the robot will follow to play the game of chess. The first step is to
consciously decide on playing chess. The robot determines that the opponent
is an expert. This prompts the robot to play the game with great
attention. Not only will the task of playing chess be in the task
container, but the other task is to play the game with great attention.
The next thing that happens is that knowledge will pour
in from memory into the conscious and populate the containers. The rules of
the game are poured in, the objectives of the game are poured in and the
strategies to use are poured in. The fact that a data is a task and another
data is a task groups them together and stored in the task container. The
fact that a data is a rule and another data is a rule groups them together
and stored in the rules container. Thus, data coming into the conscious is
organized based on association. The data should be structured in a
hierarchical manner so that only important data enters the conscious.
As the robot play chess during runtime, the intelligent
pathways selected will define the computer program in the conscious to play
the game. Every action is based on at the moment. The only exceptions are
the rules of chess. The rules container contains rules that might or might
not happen. These rules are extracted because the robot’s brain thinks that
the data might be needed in a given situation. One of the rules to follow
is that the player (the robot) and the opponent have to take turns when
playing the game. Each can only make one move per turn. This rule is
followed by the robot throughout the game.
The computer program will do things for each given
moment in time. When the opponent moves, the robot will plan strategies and
guess the opponents next move. When it’s the player’s turn, he has to plan
several future steps and select only one to act. The player knows that one
of the rules of playing chess is to make your move quickly. Thus, the robot
has a time limit to make his move.
The intelligent pathways will generate an optimal
computer program to manage tasks for the game of chess. The game will
continue and the player and opponent will keep taking turns until one wins.
One of the strategies used to play chess is to predict
the future by imagining a fabricated movie of 3 future chess moves (one
future move per chess piece). The robot has to fabricate 3 future chess
piece and decide which of the 3 moves to select. The method to do this is
to imagine the pieces moving in this space or that space. While the
imagining takes place the robot is also imagining possible opponent moves.
The brain manages these fabricated future movies and selection of each
future move is done by comparing with other fabricated future moves.
My guess is that the brain fabricates one future move
and it fabricates another future move, compare the two fabricated moves to
determine which is better, then selecting one of the two future moves.
Next, the brain will fabricate another future move compare it with the
previous selected future move and determine which is better. Previous
unselected future moves can be reconsidered too. The robot’s brain will do
this until he has to act.
Each human player uses their own personal methods, but
most chess players use the method described above. Some expert players can
actually fabricate 3 step moves for each chess pieces and determine which is
better.
While the robot’s brain is managing the task of
fabricating future moves, it will also follow the rules of chess. If the
robot is trying to fabricate a future move, he has to first identify the
chess piece and determine how it can and can’t move. The horse piece can
only move in an L movement. This rule will activate at the moment the robot
wants to fabricate a future move using the horse piece. If the chess piece
is the pawn, he can only move one step forward. While making decisions on
which fabricated future move is better, the robot has to consider the worth
of the chess piece. The king is a very important piece because it can do
all the moves that any of the other pieces can do. The pawn is the weakest
because it can only move forward. During the decision making process, the
robot has to weigh all the fabricated future moves and only decide on one
action. The computer program inside the conscious will also remind the
robot he has a limited time period to decide.
Playing chess is a lot like shopping for groceries.
There is an internal debate that determines which logical thought to
choose. Lessons from teachers guide you to decide on things.
In terms of learning how to play chess, a teacher or a
friend will teach you how to play the game. He/she might tell you this:
“imagine several future moves and select only one”. However, the learning
of chess comes from a universal pathway that can play any game. The
teachers have taught you to teach yourself how to play an unknown game.
When you don’t know how to play a game you have to first learn the rules and
objectives. This can be done by reading instruction manuals or having a
teacher teach you.
Learning the objectives and rules of a game is one
thing, but going through trial and error to determine what strategies are
good and bad is another. Trial and error makes the robot more skilled at
the game. Trial and error will keep the good strategies and delete the bad
strategies. The robot can also be taught to generate strategies and try
them to see if they are good or bad strategies.
Also, teachers can teach the robot to identify bad
things or good things in the game. Having lots of chess pieces on the board
for the player is a good thing. Having less chess pieces on the board is a
bad thing. Having strong chess pieces on the board is a good thing and
having weak chess pieces on the board is a bad thing. By understanding the
bads and goods of the game, the robot can learn to be better in the future.
Playing chess isn’t a very good example to use in terms
of identifying bad and good things in a game. If you play videogames, it is
easy to identify the bads and the goods. When a player dies, that is
considered bad. That bad move will give the player pain, which is necessary
to improve in the future. Bad things in a videogame will include: dying,
losing life energy, missing a powerup, getting hit by an enemy, jumping into
a lake, losing the game, not having high scores and so forth. The good
things in a videogame will include: having high energy level, gaining
powerups, avoiding enemies, defeating enemies, having high scores, winning a
level, beating the boss, hurting a boss and so forth.
If you play videogames, the identifying of good and bad
things is a vital part of the game. If the player does good things, it will
steer the game in the right direction. If you do bad things the game will
steer the game into the wrong direction. The robot should also have the
knowledge to identify the objectives of the game. In some unknown games, I
have to play many times before I logically identify the objectives. For
example, in donkey kong, I didn’t know what the objective of the game was at
first. I tried something and died, then I tried something else and I died
again. However, by trying something and identifying what is good and bad in
the game, I was able to use logic to identify the objectives.
The point I’m trying to make is that when the robot has
to play an unknown videogame, the identifying of good and bad things in the
game will help it generate better strategies to play the game. If the robot
has good strategies to play the game, then he will eventually succeed in
passing all the levels (refer to my previous books and patent applications
regarding how the robot plays videogames such as contra and the legend of
zelda).
<< content
next chapter >>
