Definition of Universal AGI

Definition of Universal Artificial General Intelligence

Artificial intelligence that can solve all problems as much as possible.
(If we define "solve as much as possible" and "artificial intelligence", we get the following.)
For all problems, a constant part of the program(Code, Constants) where the number of verified(Correct/Incorrect) answer choices increases over time without misinterpreting the problem.

■Definition of Artificial Intelligence

A program for solving a problem is divided into variable parts (variables, input information, knowledge) and constant parts (code, constants).
constant parts is defined as "artificial intelligence".
In order to have the same problem-solving ability as humans, we need the same knowledge as humans, but the superiority of knowledge is considered separately from the superiority of intelligence.

■Definition of problem

・Tell by questioner to respondent.
-・Range of possible answer choices (e.g. Yes/No).
-・Criteria for judging correct/incorrect for each answer choices.
-・Information required to verify each answer choices.
-・Constraints on the verification method (e.g. time limit).
Anything that can be described above is defined as a "problem".
・Interpret the question content.
・Solve the problem using the below.
-・Available computer.
-・Information given by the questioner.
-・Information already have.
・Tell by respondent to questioner.

■Definition of "solve problems"

・Example: Choosing a answer choice at random, even if the answer is the correct by chance, the problem is not solved.
・Example: If you choose a answer choice at random, verify that it is correct, and answer it, the problem is solved.
・Example: In the problem of predicting the highest probability, if the calculation is correct, the problem is solved even if the prediction is wrong.
・Definition of "solve problems":The respondent correctly interprets the correct/incorrect judgment criteria and verifies that a given choice is correct.
・It does not matter how the correct answer choice is specified (can be random).

■Categorization of "why the problem cannot be solved"

・Misinterpretation of the range of possible answer choices.
・Misinterpretation of criteria for judging correct/incorrect for each ansewer options.
・Misinterpretation of verification method constraints.
・Conflicting information between questioner and respondent.
・Missing infromation from questioner to respondent.
【Insufficient ability】
・Insufficient computing power (e.g. computing time, memory).
・Insufficient information (not only information from the questioner, but also information that already have).

■Definition of "solve as much as possible"

Misinterpretations can be rectified at the programming stage.
Absence of misinterpretation is a requirement of "solve as much as possible".
Given computing power and information can be used efficiently, but cannot be increased by program.
The most efficient use of a given computing power and information is not a requirement of "solve as much as possible".
There is no need for maximum efficiency, but progress towards optimal solutions must be made.
The number of verified(Correct/Incorrect) answer choices increases over time is a requirement of "solve as much as possible".
Definition of "solve as much as possible": The number of verified(Correct/Incorrect) answer choices increases over time without misinterpreting the problem.

■Defining Levels of Universal ArtificialGeneral Intelligence

If the computational power is infinite, the optimal solution can be obtained, but the efficiency with limited power is expressed as a level.
【Level1. Optimal accuracy】
Only choice the correct choice, but does not take into account the computational power required.
Best within infinite computing power and limited information.
【Level2. Optimal calculation order】
Efficiently narrow down choices within limited computing power
【Level3. Optimal redundancy】
Avoid unnecessary calculations
Best within limited computing power.

■Classification of Elements of Universal AGI

Categorize the program elements so that when layer n and below are completed, level n is completed.
【Layer1. Accuracy】
determine the conditions for consistent interpretation.
All possible combinations may be explored in a brute force manner.
【Layer2. Calculation order】
Decide on an efficient search method (breadth-first, depth-first, etc.).
Decide the conditions for terminating the search and the processing to be done when it is terminated.
【Layer3. Redundancy】
Minimize the number of program comparisons and branches required.