Overview Index

Experiment

Experiment is the core part of the learning process (another part is knowledge download).

Experiment checks and updates representation of the world in the main memory.

There are two types of experiment:

1. Active experiment.
2. Passive experiment.

Active experiment

Active experiment includes these steps:

1. Accomplish an action.
2. Get the response(s) from the world (collecting events).
3. Analyze correlation between action and response (events).
4. Update relations between involved concepts.
5. Analyze how suitable were concepts involved into the experiment.
6. Updating desirability of involved concepts.

Example:

1. Strong AI accomplishes action: “Send message over ICQ ‘Hi, dude!’”.
2. Strong AI receives an event: “’Hello!’ response over ICQ”.
3. Event correlation analyzer analyzes correlation between action “Send message over ICQ ‘Hi, dude!’” and event “’Hello!’ response over ICQ”.
4. Coherence attribute in cause-effect relation table is updated between concept “Hi, dude” and concept “Hello”
5. Reward generated by super goals is calculated.
6. The reward is distributed among all cause-concepts responsible for the event.

Passive experiment

Passive experiment consists of these steps:

1. Getting an initial event from the world.
2. Collecting subsequent events from the world.
3. Analyzing correlation between initial event and subsequent events.
4. Update relations between involved concepts.

Example:

1. “Lightening has flashed” event happened.
2. “It is thundering” event happened.
3. Event correlation analyzer analyzes correlation between “Lightening is flashed” event and “It is thundering” event.
4. Coherence attribute in cause-effect relation table is updated between “Lightening is flashed” concept and “It is thundering” concept.

Experiment in general

Active experiment and passive experiment are very similar with each other.

Experiment in general consists of 3 major parts:

1. Action.
2. Receiving event.
3. Analysis and memory update.

Take into account that an action can be considered as “initial event”.

“Analysis of ‘Action->Event’ correlation” is made by Event correlation analyzer. This analysis is pretty simple. Results of the analysis are saved into cause-effect relation table and into concept table (desirability attribute).

Experiment as the key part of the learning process

Experiment is the core process of learning.

Knowledge download (reading books and listening to others’ thoughts) may be very efficient. But “knowledge download” works pure without follow up experiment. That’s why I consider experiment as the key part of the learning process.

Strong AI constantly learns through experiment:

Strong AI considers every action and every event as parts of experiment. That means that every action and every received event are used for the learning process.