• Erik Luo

An idea for realistic human AI


An idea for realistic human AI:

A human has:

The concept of self (All human experiences)

The instinct to avoid pain and pursue pleasure

A degree of happiness

A degree of suffering

Fear

Love

Death

Life

The purpose of self is to survive

And to experience

The self has the ability to learn, because the cause of all suffering is fear, and fear is caused but ignorance.

By learning, we overcome fear.

The root of happiness is love.

Love is the act of selflessness.

It is the will to help others overcome their fear.

A realistic human AI must have these things, but how?

What if we teach the code to code itself

To modify itself

A program that we can give a task and let it make it happen

Let's say we want it to learn math

What is the first thing we need?

A self.

What is a self?

How do we make programs which is made out of 0s and 1s understand that they exist?

Well, similar to computer programs, human take input and have an output.

Input such as the 5 senses.

We learn from experiences, if we lose all our senses, we will not have a self.

Knowledge is built from experiences.

So we first need an input.

But how do we do that if the program doesn't have any senses?

What if we let them have senses but in a virtual reality?

If we can simulate photo-realistic visual graphics, we can also simulate other senses.

So the AI can have experiential input, not by the player, but by the reality itself.

For example, this human AI will have a part in its program that takes visual input.

Visual information: Color, light, value, perspective, the ability to see in 3d.

Then the program will take that information and give an output, such as a thought.

But to have thoughts, one must first have a language.

With the program, we should be able to make it so that we can teach it languages.

One of the first thing that is needed in order for this program to work is the ability to process information, which is the ability to give meaning to things.

It will get information from the virtual reality, then assign meaning to them through experience.

For example, same as a human, they will have to learn that fire is hot, by touching it with their hand.

There will be a program for all senses input.

Visual: Color, light, value, perspective, the ability to see in 3d.

Smell: All smells. Our smells are very complex. It doesn't have a degree of description. But we can create one. Good smell and bad smell.

Touch: Hardness, temperature, dryness, smoothness.

Sound: High and low, frequency, Music, the brain likes order. How the frequency was arranged will determent whether or not they sound good.

Taste: Good and bad. Things tasted the way they are for a reason. For survival. An apple is sweet because it has things that our body needs. Other things that don't taste good means that they are not what our body want. However, there will be a system to keep in check of the amount of "Good things" the AI eat. There will be a causation to every action in this world so there will be a cause for every intake of things.

After making the input program, the next step is to make the processing system.

A system that takes information and process it bases on the level of survival that information provided.

It will have a system to value the information and order them into the most important to least. The degree of importance depends on the condition and the need of the AI at that moment. It will follow the pyramid of human desire. Go from survival to self-actualization.

There will be another system call the system of causation. Which will give causation to information and things.

For example, the fire has information stored in it for all five senses, and when the Ai touch the fire, the processing program will take the information and output the likely cause.

There will also need a data management system.

Which take the information and store it.

In time, the AI can refer back to it at its own need of survival.

If an information has not been referred back in a certain period of time, it will delete itself.

So the AI can forget things and make room to remember more.

Next is the movement system, this will be a system where the Ai can control all aspect of its 3d "Body" in this space, obeying the laws of physics.

Then we need a system that allows the AI to make changes to and interact with things in the reality.

Interact system.

For example: If you put water into fire, the information in the fire will change, such as the temperature, the color, the smell, and that will have a causation.

What is causation?

Information that influences other information.

Now, Let's imagine I already made this.

A human looking 3d model who is in a blank world with a campfire.

What will the AI do?

It will take all the information and make a decision based on the information(thought) output through the processing system.

This will have great research and experimental value.

Imagine making a game out of this AI.

The AI will have the ability to generate stories by interacting with other Ais or the player.

It will be different every time.

A multiplayer without other players.

A world that changes and evolves itself.

An infinite possibility for experiences.

You can create realities at your own will.

We will be able to run experiences on these realities and help humanity for the better.

(This is just a basic idea)

How do I make this?

Let’s say I am making this.

I will have Player object.

For the visual program, we can have a range object. It will be in front of the AI object.

It will collide with things and determent if they can be seen or not.

If they can be seen, then the range object will collect all the visual information about those objects.

Information such as color, light, value, distance, and position.

Then the information will be taken by the player object, and then it will process it.

The processing system will take that information and compare it to the survival range and the happiness range.

If the object has information that is helpful for the survival, the AI will seek action.

If the object has information that is not helpful for the survival, the AI will avoid it.

The causation system/Interact system.

If the player object collide with other objects, the interact system will take the information of the object and the AI, then find the logical outcome.

Then it will output it.

The movement system:

We can have two types of movement system.

  1. Self-learning system

  2. Pre-determent animation assets

For the Self-learning system, the Ai will have the ability to play around with the causation system.

For example, we have a AI, and an apple, but the two objects have a far distance from each other.

In order to get the apple, the AI must know how to walk.

We will make the hunger level extremely low, so that the AI will move like it’s life depends on it.

This will require an animation system that is not controlled or pre-determent.

Everything must be running on the causation system and obey the laws of nature.

There will be need for an complex physics engine.

For the pre-determent animation system, we will have a wide range of set animation for all possible situation.

The use of animation will be a part/result of the output system.

Yes, to make it 100% like a human would be tough, but I think we can use this concept to make our current AI better.

Let's take the input system as an example.

We don't need to make it simulate 100% of how human senses work, that will be way too hard for our current technology to do.

We only need to make it have some kind of input from its world.

Information doesn't have to be super complex, it just needs to give the AI an idea of the things it's perceiving.

The visual system as an example, we don't need it to know exactly how red something is like a human would, we just need it to know enough so that it acts like a human.

#Insights #GameDev

Erik Luo

Game Designer

VR/MR Developer

Contact

Follow

©2017 BY ERIK LUO.