A general intelligence is learning and problem solving abilities, that may be applied in any situation.
A separate machine for a word processor or calculator is not required. A universal turing machine (or it descendent the PC) may be used with an appropriately chosen program.
General intelligence is the equivalent idea applied to the learning problem solving domain.
Intelligence probability. edit
Intelligence can be measured as the probability of the intelligence solving any given problem weighted by the probability of that problem.
A problem is a time limited interaction with an environment, with a goal that is either achieved or not achieved.
- The intelligence probability is the probability of an intelligence solving any problem given to it.
For an intelligence there is a probability of it being posed any problem. The problems may form a continuum so the probability must be considered as a probability measure.
So,
- The measurement of intelligence for each problem is either,
- 1 - The intelligence will achieved the goal.
- 0 - The intelligence will not achieve that goal.
- The intelligence probability is,
- The weighted sum over the probability measure of problems of the measure of intelligence for each problem.
What is an intelligence edit
An intelligence is a machine or algorithm that may be applied to any situation where there is input sequence, an output sequence, and a goal (or reward sequence). The machine must choose outputs that achieve the goal, or maximize the sum of the values in the reward sequence.
The output is chosen by the intelligence. The intelligence has no prior knowledge of how the input and the reward are related to the sequence of actions.
Probability edit
Probability relates to information. Given a coin toss in advance I have no information about whether it is more likely to be heads or tails. I say that there is a 50% chance of it being heads or tails.
I do not need a series of trials to arrive at a probability. A probability reflects my knowledge (or lack of knowledge) about the coin. The coins behavior does not need to be inherently random for there to be a probability. Lack of knowledge about the result of the coin toss makes the result unpredictable for the intelligence.
Probability is a value that belongs to the intelligence. A different intelligence, which has different information, or different initial prior probabilities may give a different probability, for the same event.
Algorithmic probability edit
Probability is about finding patterns in the data. Ray Solomonoff interprets a pattern as a program that generates data. A program has inductive power if the program is shorter than the data it represents. Algorithmic probability is the theory of probability defined using the length of the shortest program that represents the input from the world.
Decision theory edit
Decision theory determines the best choice of actions that maximize the expected reward.
The theory considers the sum of the rewards after each action. Each action invokes a response from the world, that alters the probabilities. This implicitly takes into account the information gained by taking each step
The reward function need not be similar to human rewards. The regard may be the accomplish of a goal. The purpose of the reward as a real value function allows values to be placed on each outcome.