Taiiwo/AI.md

## AI.md

      
    Raw
  

              AI.md
            
          
    The Basis of an Artificial Intelligence

So the general concept behind an artificial intelligence is quite simple. It's
basically just a replication of what we see in other sentient beings. It has been
abstracted to mean things like speech, the ability to play a game well, to guess
a person's motives and intentions, and to use that against them in order to win.
The reason this seems like a form of intelligence to us, a being with sentience,
is because as humans, we see anything with an understanding of anything to be sentient.
Almost, anyway. An understanding is a very important part, but where it really hits
home, is when the being shows a capacity to use it's understanding to achieve a
specific goal. Sometimes goals we share. This definitely makes it easier to relate
to it. Nobody ever made a conversational AI that is antisocial, simply because we
want it to talk back. These AIs have tendencies to try to make us talk back. It asks us
questions, tries to put us on the spot, tries to take control of the situation in
order to reach its main goal: to pass the turing test. The designers of these systems
add quirks for this, things like how the AI handles things it doesn't know. If it
threw an error, it would break the immersion, so it doesn't. It realises that it
has a flaw, and tries to avoid it, usually by asking a question or changing the subject.
What I want to talk about here is essentially my understanding of what an AI is.
I want to create an AI, and I'm writing this document to try to get things straight
with myself. What are these things I'm trying to replicate? How do I define them?
They are all very simple. Almost every human on earth must have an understanding
of what they are. All I need to do is break them down into a synthesizable abstraction.
So, here we go. I'm going to try starting from the ground up, and explain piece by
piece, what I think a consciousness is.
Information

A lot of the words I'm about to use are actually synonymous, so it's understandable
if you'd not know what I mean when I say them, because coming from anyone else, they
could mean something totally different. Data, as I was reluctantly taught in A Level
ICT, a course I failed on a technicality, is a raw fact or figure without meaning.
19 is a number. An integer. It is data. 19 is my age. It is now information. The
purpose of this section is to create an abstraction of this process, and to explain
to you, just to share an understanding that I totally made up, and how it's done.
I believe, for the purposes of this demonstration, that information is made up of
3 things:

Elements
Attributes
Values

Elements

So, the element is the subject. The body. The object, the parent. There are already
countless similar abstractions to this one. The wording to this one specifically,
I took from HTML, simply because the concepts are comparable. A piece of information
must have an element. It is the thing the information is about. It is an important
part in the identification process. Without an element, we just have a jigsaw piece
that doesn't fit anywhere.
Attribute

This is the thing we know, in the information. The name of the type of information
this is. We already know we have an element, but what about the element do we know?
We know the Attribute.
Value

What is the attribute, if we know it? We know we know the attribute of the element,
but what is it? The value is the value that is assigned to the attribute that is
assigned to the element.
What Happens if we Fuck with Information?

So, if you're of any level of respectable intelligence, and this isn't me judging
you, but me simply defending myself from feeling arrogant, and that my simple theories
aren't actually as complicated as they are in my head; you'll have gotten to this
point thinking that I've just displayed a very partial element to an understanding
of what a basic fact is. Like I've said, this abstraction has been used many times
before me; but this is where it gets cool.
These three things aren't just a list of things that I've identified as important,
there are no hidden other ones. There is nothing I haven't thought of, but not because
my understanding is perfect, hell, I wouldn't need to write all of this down if it
was. The reason I know there's nothing more, is because if you manipulate these three
things, you start to see deeper into the process of learning.
Take a sentence made up of all three of these things. If something has all three
of these things, we can call it a fact. There are issues in this theory that reside
within the english language, and I'm yet to come across a language that doesn't have
these quirks. Let me just give you an example of a fact:
I am 19 years of age.

Pretty boring, right? Not something you'd find one the back of a cereal box, but
it does the job, and it serves its purpose as something you probably didn't know
before. In this example, I am the element. The attribute is my age, and the value
is 19 years. We talked about data above, and all of these things individually,
without context, are examples of that. Just as a side note, years could be a forth
thing, I'd probably call it the unit, but it fucks with everything and makes it more
complicated for no extra value other than the possibility of changing the unit later,
but let's say for now, that all of the values are actually stored as their base unit,
seconds in the case of time, and then converted as required.
So we get to inspect some merits of the english language when we start moving these
things around. You'll notice that I keep calling them things, because I've already
taken up all the good names, and I don't want to confuse things by throwing another
label around. The fact above is in the order [element][value][attribute]. You
could shift those around, and get 6 different sentences, all meaning the same thing.
Not very useful, but it displays the kind of concept that I'm getting to.
The more interesting stuff happens when you start removing them. One of the annoying
quirks of the english language is that removing the attribute just makes us guess it.
I am 19.. But if we remove one of the other two, the only logical sentences that can
be made are questions. Less value: How many years old am I?, Less element: Who is 19 years of age?. As a side note, these questions also have one other possible formation.
If we had a piece of software, and I'm working on one, that could turn these 3 things,
in any given order, into a sentence, what we would have is a way of automatically
creating sentences out of a databank of information. We need only
supply our questions in the form of two, or even one of the three things (where
the attribute is assumed), and we would be given a sentence, assuming the answer is
in the databank.
Add, if you will, another piece of software that reverses the process (another work in
progress). That takes a sentence, and deconstructs it into it's base components. It would
give us two things: A way to ask the system questions in plain english, and more interestingly,
a way for the machine to read content, and learn information from it. It could be the most
intelligent program of all time, if taught how to read something like the web,
or a set of up-to-date encyclopedias.
Not Good Enough.

Ask it How old is Barack Obama's cat?, and it might know the answer. Ask it for
a description of the quadratic formula. Ask it for it's applications. It's creator.
When was it created? Where was it created? Why? You might get answers to all of these
questions, depending on the information being available, and whether or not the
learning algorithm was able to parse it correctly, and we must also consider things like
the possibility of the AI being tricked. Misleading, incorrect information. It doesn't
know the answer. It only know what someone thought the answer was. So what's next?
If people were just these things that stored these databanks of, well, data,
then would we be sentient? Or just walking databanks? We spoke at the start about
sentience not being about information, but about understandings, and above that,
the applications of those understandings.
An understanding is a collection of pieces of information that support concepts.
You could ask the AI we outlined above What is the quadratic formula?, and it would
try to find an appropriate answer for you. Ask it Teach me the quadratic formula.,
and what would it do? Well, realistically, you've only given it the element, so
naturally, the algorithm sees it as a request. You didn't specify an attribute, so
it's going to have to guess that too. It would probably end up giving you a textbook
description of the algorithm, but if that was all we needed to be taught, then why
would we have both teachers and textbooks? This AI would never ask you a question.
And why not? I believe that the answer comes from the systems restriction to information.
Each fact is in the database by itself. Waiting for the specific question that
flags it up as an answer. It knows nothing of it's neighbours. They're unrelated to him.
The records in the database are probably ordered by time, meaning the description
of the quadratic formula might be right next to the age of Obama's cat. This is
a description of a problem, that also leaves us with a room for improvement.
Understanding of Information

This is pretty much where I'm at. I've figured out up to the information level,
and I know the text step, but if I'm to be honest with you, I'm writing this document
in the hope that the next part will come pouring out of my fingers, so let's go.
I believe the order of things to be something like:
Universe¬
    Realities¬
        Sentiences¬
            Understandings¬
                Information¬
                    Data

Looking at that is quite daunting, actually. It seems I have a long way to go. I
define the singularity as the complete artificial abstraction of these things. Or
any kind of full, all encompassing understanding of these things. I once had a theory
that the universe was a multi dimensional fractal, and that if we found the formula,
we could theoretically know anything. Limited only by the time it takes us to learn
and comprehend. I imagined a formula using the variables x, y, z, time and d. Imagine
the universe has a center. All of these variables, if we are to use integers to represent
them, are the smallest possible units that the universe is capable of. Imagine a ring
around the center of the universe, with a near-infinite number of points. The radius
of this ring is the same unit of measurement as the variables. The smallest possible.
Now imagine a ring that surrounds that one, with the same number of points, this
number is constant. The distance between this ring and the next is again, the smallest
distance possible. This goes on until the edge of the universe, and as it goes on,
the universe is greater still. The points on the circles are selected with x. The
circles themselves are selected with y. This pattern is rotated by z, creating a
series of spheres. By selecting any combination of these variables, we are given a
fixed point in space. Time is our distance, in units of the smallest possible unit
of time, from the beginning of the universe, and finally, the throw a curveball,
if we want to support the multiverse theory, d gives us the number between zero and
infinity that represents the dimension of the universe that we're talking about.
Given these values, the formula would return the value, probably a one or an zero,
of the space at that given point at that given time in that given dimension. Using
this, we could construct an accurate simulation of anywhere, and any time in the universe,
past, future, present, or other-worldly.
So. Let's get started. We need to understand the next step. The understanding step.
If an understanding is a representation of a set of information, then how much
information do we need to construct an artificial understanding. I guess the lowest
level of this is knowing that it exists. That it is a thing in the world. If it can be
seen, what does it look like? If it can be heard, what does it sound like? Etc.
The next most important thing is probably a practical knowledge of the thing. What
is it used for? How do we use it, and how does it work? After that, I think we need
to know how it compares to other things. If there are a line of identical objects,
but one is green, and the rest are blue, then one is a green object, and the rest
are object. We need to know what sets it aside, what makes it different to the rest.
What we've outlined here are the basic three components that are required to accurately
convey an understanding of what something is. To teach, in other words.

Objective
Practical
Comparative

Let's say that objective is a list of value-based properties: colour, size, weight,
etc. Practical is a set of instructions for each of it's uses, as well as the limitations
and restrictions of using each. Finally, the comparative requires the first two
sets of information, and compares them to the first to two sets of everything else.
It needs to find things close enough to be considered similar, but different enough
to not be considered the same thing. So we need a list of those things, along with its
difference from each.
An explanation of something can be given, therefore, in 3 paragraphs. A list of
sentences that give the trivial properties of the thing, a list of its practical
applications, and their requirements, limitations, and restrictions; and then
finally, a list of its alternatives, and it's differentiations from them.
A Cake

A cake's shape is commonly round, but also can commonly be square or rectangular in
shape. A cake is commonly sweet, and created as a dessert or for a celebration. There
are many types of cake, ranging in flavour, colour, size, and shape.
One can make a cake, as long as one has the necessary ingredients and the utilities
required. The amount of cake one can make is restricted by the amount of ingredients
one has, and the limitations of the utilities used. One can eat a cake. One generally
requires eating utensils, but this is optional. One can only eat so much cake, based on
one's appetite, and willingness to eat the cake.
Cake is one of a range of desserts including pies, tarts, icecreams, and confectionery.
cake stands out from these as it has a sponge-like texture, although cake can be used
as an ingredient within other desserts. Cake is one of many celebratory traditions, amongst
parties, champagne and other alcoholic beverages, cigarettes and cigars, and gifts.
Cake stands out from these as it can be eaten, and is often done so alongside enjoying
the other celebrations. Cake is a type of food, and has a nutritional value. Compared to
other foods, cake is generally high in sugar, and therefore considered unhealthy.
Why did I just read that?

That was a 3-part description of my understanding of cake. There were a lot of things
omitted, and algorithms that utilize this set of theories will have to index these
by importance, and create sentences until each paragraph reaches a desired length.
Okay. We're Sentient. Now What?

Theoretically, with enough information in the right format, our AI now knows everything
there is to know about everything. Not just what things are, but how to use them,
and what using them does. How to pick the best thing for the job. This is where the
reality layer kicks in. The reality layer is not like the other layers we've attempted
to understand so far. It's more of a way to expand what we have. Because once
we have sentience, we, by default, have a concept of reality, but our ability to
change that reality lies within the abilities of the entity we embody. For example,
place this AI inside a program that types, and it will be able to follow the instructions
for writing, say, and essay. As long as it has the specified requirements, it can
do anything, as long as it knows how.
AI, write me an essay on volcanoes, please.. Okay, sure. How do we write an essay?
Here's some sets of instructions. Let's start with the first one, and see if I have
the requirements. It says I need a connection to the web. Check. Now I need to the
ability to type on a keyboard. Well, how do I do that? Here's some sets of instructions.
Looks like I need hands, and a keyboard. I have hands already. I'd need to go find a keyboard,
but that's possible. I know where one of those is. Looks like I have all of the
requirements for this option, and I've checked all of the other options, and this one
is the fastest, so let's go. I'll go to the keyboard. The first instruction is to
write an introductions. I need to write a sentence about what I'm talking about.
I know what a volcano is, so I'll write a description of that. Now, how do I type?
Apparently, I take my fingers, and push them down of the keys corresponding with
the letter I'd like to input. I can do that. I'll just go through each letter of
the sentence I want to write, and type that letter.
I'll spare you the other steps, but that's how it does things. It just keeps looking
up how to do the individual tasks, and then puts the mall together to complete the
main task. This works recursively, and can accomplish any task that the AI has
sufficient capabilities for, and and appropriate selection of instructions for.
Universe = Singularity

Well, There's only one step left. This step is sort of hard, and we can no longer
embody a single being. We must have full control over everything in the universe.
Then we need an objective. The meaning of life. We explore the options, and choose
the fastest one, or the best one over a given time frame. Now, here's what sets
aside the universe-AI from the sentient-AI: the ability to create our own instructions.
The perfect AI, has a list of every interaction it can have within the universe,
and its requirements and effects. All it needs to do, is check every possible sequence
of actions, and score it's viability, in terms of success probability, speed, quality,
etc. of the outcome, and pick the one with the highest score.
You're probably thinking, that's not the singularity. I can do that, to some extent,
I just don't have control over everything in the universe. True, humans are capable
of universe-level thinking, but once it's done, we simply add that to the list of
things we know how to do, and move on. A true universe-level AI, does not do this.
Yes, it adds it to it's list of instructions, but each time it is required to complete
a task, regardless of whether it has been asked to do this before, it will always
re-calculate its options.
A human implementation of this, at this level, is dangerous. A universe-level AI
can only be as perfect as the person who creates it. While the concept of the AI
will be striving to achieve perfection, any mistake in its valuations and concepts
of morality could be catastrophic, and bring about the types of things we see in just
about every AI movie ever.