Hi all. I chose to create a new thread for posting a summary of ideas about the system that we will be developing. I will not get into descriptions of algorithms here, I believe that we need to focus on the architecture of the system, and then start analysing the details of each system. Nevertheless, my opinion about algorithms is that we should implement many different ones and embed them in our system, so that we will be immitating the multitude of neural pathways in the human brain.
I chose to put this proposal in a separate thread because, although it is a summary of my ideas, it is still a little lengthy. I hope it will not be difficult or tiresome to read.
I will put 5 postings: description, general information, archetypes, model, and the system.
The description a short well of information on the type of system we are trying to develop as I have seen it.
General information will give a background in the complex meaning of intelligence-knowledge.
Archetypes will provide a short discussion of my understanding on how archetypes should look like.
Model discusses the architecture that I am in favour of.
The system discusses some inherent limitations of our system.
Thank you,
Alexios Tsiaparas
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
We want to develop a system that will show elements of intelligence. The underlying theory of conditional AI - stating that intelligence is a large (possibly infinite) number if if-then statements - will be used as our baseline.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
We need to understand the basic functionality of the human brain before we attempt to make something that tries to immitate our intelligence.
One point that we all agree on so far is the internal storage, i.e. archetypes. Forms, initially blank, that will be filled with information. The question is what kind of information. This answer is provided the anatomy of the brain.
The brain has specific areas that handle different types of archetypes. Speech, reading or smelling trigger different parts of our brain.
The mind stores archetypes that can be received by our sensory organs. The brain thus links archetypes with the different medias of input. In terms of programming, this means different objects for speech, text, etc.
For example, our brain does not store information on ultra-violet light. Not because of any electro-chemical incapabilities, but simply because we do not see ultra-violet light. The brain does not contain any centers for handling such information.
That said, this is the same limitation of our system. Its intelligence is bounded by the different kinds of input and output it can handle.
For example, everyone knows about the word 'AND' and also how to use it. But if one asks 'what does AND mean?', noone can give an answer on the meaning of this word. Any attempts fall into an explanation through use, not the actual meaning. So, we use it because we have just seen that one can merge two things by putting this word in between. Moreso, what is love, we all know the word, but unless you possess the capabilities of filling the thrill of being with someone you love, then you cannot explain the word sufficiently.
These examples show that our intelligence is directly linked with our input-output capabilities.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Let us suppose that we do not know anything about cars. Not ever even heard the word. If a car passed next to us, we would have stored the image of the car, but we cannot say 'this is a car'. If we hear the word 'car' in a discussion, we also store this information, but we cannot say 'this is a car' and point to one. But if someone shows us a car and tells us that it is one, then we link the two archetypes together.
I tend to think of archetypes as graphs that link one archetype to others. The edges are the meaning, the 'actions' that connect the two archetypes. In the previous example, we have two archetypes 'WORD513' and 'IMAGE1566'. The edge between them is an 'IS' relation, or 'action', where 'WORD513' is car and 'IMAGE1566' is the image of a car.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I still favour the model :
INPUT->PRE-PROCESSING->BRAIN->POST-PROCESSING->OUTPUT,
or the more refined
INPUT->SYNTACTIC ANALYSIS[in]->SEMANTIC ANALYSIS[in]->BRAIN->SEMANTIC ANALYSIS[out]->SYNTACTIC ANALYSIS[out]->OUTPUT
because if we allow our primary model -the generic one- to include elements such as speech, then we would confine our analysis capabilities.
Speech, vision, or hearing are different components of the syntactic, semantic systems. What exactly is syntact and semantic analysis and brain?
Syntactic analysis is the system that translates between the signals received or the signals to be transmitted and their internal capabilities. For example, when we move, the syntactic system knows what signals to send to our muscles in order to lift our feet. Our phonetic chords can output a wide range of voices and speech-symbols. Nevertheless, when we learn a new language, we cannot instantly pronounce correctly the new symbols. Our syntactic system has to learn how to signal the chords in order to produce the desired speech-symbol. When we succeed in this, then our syntactic system starts learning how to effectively pronounce whole words, instead of slowly reading each letter. And the task continues to perfection.
Sematic analysis is the place where we direct, select actions to perform. This is the place that we take archetypes and we apply the actions that we can do with them. This is the place where we discover new actions and new archetypes. For example, I have the capability of writting 'You are very clever' (As a proof, I have just done that :) ). But normally, I would not have written something like that, unless my syntactic system was instructed to do so from my semantic system, which would know that you are very clever.
The brain is the system that learns new knowledge and selects knowledge. If I tell you to show me a car, it would be you brain that would have linked the word 'car' with the image 'car' and the sound 'car', providing your semantic system with all these archetypes so it could select the image 'car' and forward it to me (assuming of course that we had some way of transmitting this image, this would have been an output media that currently humans do not normally possess).
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
As I have stated in the introduction, I will not analyse any particular algorithms here, because according to me, we need many different algorithms so that our system becomes more intelligent.
I think that we need to discuss first on the decomposition of the system at its entirety, and then we can move to defining simple and complex algorithms in order to address each step in our decomposition.
I hope that I was not too long and tiresome. Thanks to everyone who reads these posts. As far as I have seen the concepts of RCL and CC algorithms, they look very promising candidates for the semantic analysis phase in the model proposed. We can certainly employ the RCL in order to identify patterns in the data that we are supplied and failing that, CC can take over constructing the actions from the ground up.
Thank you,
Alexios Tsiaparas
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi all. I chose to create a new thread for posting a summary of ideas about the system that we will be developing. I will not get into descriptions of algorithms here, I believe that we need to focus on the architecture of the system, and then start analysing the details of each system. Nevertheless, my opinion about algorithms is that we should implement many different ones and embed them in our system, so that we will be immitating the multitude of neural pathways in the human brain.
I chose to put this proposal in a separate thread because, although it is a summary of my ideas, it is still a little lengthy. I hope it will not be difficult or tiresome to read.
I will put 5 postings: description, general information, archetypes, model, and the system.
The description a short well of information on the type of system we are trying to develop as I have seen it.
General information will give a background in the complex meaning of intelligence-knowledge.
Archetypes will provide a short discussion of my understanding on how archetypes should look like.
Model discusses the architecture that I am in favour of.
The system discusses some inherent limitations of our system.
Thank you,
Alexios Tsiaparas
We want to develop a system that will show elements of intelligence. The underlying theory of conditional AI - stating that intelligence is a large (possibly infinite) number if if-then statements - will be used as our baseline.
We need to understand the basic functionality of the human brain before we attempt to make something that tries to immitate our intelligence.
One point that we all agree on so far is the internal storage, i.e. archetypes. Forms, initially blank, that will be filled with information. The question is what kind of information. This answer is provided the anatomy of the brain.
The brain has specific areas that handle different types of archetypes. Speech, reading or smelling trigger different parts of our brain.
The mind stores archetypes that can be received by our sensory organs. The brain thus links archetypes with the different medias of input. In terms of programming, this means different objects for speech, text, etc.
For example, our brain does not store information on ultra-violet light. Not because of any electro-chemical incapabilities, but simply because we do not see ultra-violet light. The brain does not contain any centers for handling such information.
That said, this is the same limitation of our system. Its intelligence is bounded by the different kinds of input and output it can handle.
For example, everyone knows about the word 'AND' and also how to use it. But if one asks 'what does AND mean?', noone can give an answer on the meaning of this word. Any attempts fall into an explanation through use, not the actual meaning. So, we use it because we have just seen that one can merge two things by putting this word in between. Moreso, what is love, we all know the word, but unless you possess the capabilities of filling the thrill of being with someone you love, then you cannot explain the word sufficiently.
These examples show that our intelligence is directly linked with our input-output capabilities.
Let us suppose that we do not know anything about cars. Not ever even heard the word. If a car passed next to us, we would have stored the image of the car, but we cannot say 'this is a car'. If we hear the word 'car' in a discussion, we also store this information, but we cannot say 'this is a car' and point to one. But if someone shows us a car and tells us that it is one, then we link the two archetypes together.
I tend to think of archetypes as graphs that link one archetype to others. The edges are the meaning, the 'actions' that connect the two archetypes. In the previous example, we have two archetypes 'WORD513' and 'IMAGE1566'. The edge between them is an 'IS' relation, or 'action', where 'WORD513' is car and 'IMAGE1566' is the image of a car.
I still favour the model :
INPUT->PRE-PROCESSING->BRAIN->POST-PROCESSING->OUTPUT,
or the more refined
INPUT->SYNTACTIC ANALYSIS[in]->SEMANTIC ANALYSIS[in]->BRAIN->SEMANTIC ANALYSIS[out]->SYNTACTIC ANALYSIS[out]->OUTPUT
because if we allow our primary model -the generic one- to include elements such as speech, then we would confine our analysis capabilities.
Speech, vision, or hearing are different components of the syntactic, semantic systems. What exactly is syntact and semantic analysis and brain?
Syntactic analysis is the system that translates between the signals received or the signals to be transmitted and their internal capabilities. For example, when we move, the syntactic system knows what signals to send to our muscles in order to lift our feet. Our phonetic chords can output a wide range of voices and speech-symbols. Nevertheless, when we learn a new language, we cannot instantly pronounce correctly the new symbols. Our syntactic system has to learn how to signal the chords in order to produce the desired speech-symbol. When we succeed in this, then our syntactic system starts learning how to effectively pronounce whole words, instead of slowly reading each letter. And the task continues to perfection.
Sematic analysis is the place where we direct, select actions to perform. This is the place that we take archetypes and we apply the actions that we can do with them. This is the place where we discover new actions and new archetypes. For example, I have the capability of writting 'You are very clever' (As a proof, I have just done that :) ). But normally, I would not have written something like that, unless my syntactic system was instructed to do so from my semantic system, which would know that you are very clever.
The brain is the system that learns new knowledge and selects knowledge. If I tell you to show me a car, it would be you brain that would have linked the word 'car' with the image 'car' and the sound 'car', providing your semantic system with all these archetypes so it could select the image 'car' and forward it to me (assuming of course that we had some way of transmitting this image, this would have been an output media that currently humans do not normally possess).
As I have stated in the introduction, I will not analyse any particular algorithms here, because according to me, we need many different algorithms so that our system becomes more intelligent.
I think that we need to discuss first on the decomposition of the system at its entirety, and then we can move to defining simple and complex algorithms in order to address each step in our decomposition.
I hope that I was not too long and tiresome. Thanks to everyone who reads these posts. As far as I have seen the concepts of RCL and CC algorithms, they look very promising candidates for the semantic analysis phase in the model proposed. We can certainly employ the RCL in order to identify patterns in the data that we are supplied and failing that, CC can take over constructing the actions from the ground up.
Thank you,
Alexios Tsiaparas