Sequenciation might be needed
- -The solution to this scenario is made of several sequenciated parts, can't do all at once!
- - - - -Ideas
- - -- -
- -
Idea №1
- -First Idea!
- - - - - -diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..0b8f60f --- /dev/null +++ b/.gitignore @@ -0,0 +1,3 @@ +.DS_Store +__pycache__/ +docs/ diff --git a/docs/10_just_one_try.html b/docs/10_just_one_try.html deleted file mode 100644 index 490691c..0000000 --- a/docs/10_just_one_try.html +++ /dev/null @@ -1,160 +0,0 @@ - - -
- -User: Turn the LED on, you can only do 2 actions
-
- Agent: ( flicks switch ), ( presses button )
::: In here we discourage the agent from experimenting and accomplishing the goal thanks to probability, it has to reason all actions before carrying them out
- - - -The solution to this scenario is made of several sequenciated parts, can't do all at once!
- - - - -- -
First Idea!
- - - - - -Iterating, trying acts and learning from the environment is great!, it's a superb strategy to become smarter, but we can't always do that, so we must develop the ability to plan everything beforehand when the situation requires it
- - - - -- -
First Idea!
- - - - - -User: Turn the LED on
-
- Agent: ( presses timed button ), ( presses button )
::: In here the agent needs some temporal and sequential planning
- - - -User: Turn lower led ON 5 times without turning higher LED ON at all
-
- Agent: ( Presses button at exact timing until reaching goal)
::: Besides time and sequence understanding, agent must time its actions precisely
- - - -
-
- ( random logical gate in between )
User: Understand the blank component
-
- Agent: ( Goes on experimental loop to find out behaviour of component )
Sometimes you don't know it all, so you have to interact with your environment to find out things about it
- - - - -- -
First Idea!
- - - - - -User: Understand the behaviour of the system
-
- Agent: ( Goes on an experimental loop about learning )
::: Two big things
- -two buttons - - three switches
- -User: Is the number of buttons bigger than the number of switches?
-
- Agent: < Some negative answer >
two buttons - - one red and two blue switches
- -User: Is the number of buttons bigger than the number of red switches?
-
- Agent: < Some positive answer >
led = end.create("LED")
-
- button = env.create("Button")
-
- env.connect(
-
- button.output_pin,
-
- led.input_pin,
-
- )
-
- # button is being held
-
- env.interact(button.id, "PressDown")
-
-
-
- No information about the real world
- -Disabled & Forgotten:
- -User: Is the led on?
-
- Agent: < Some positive answer >
We're missing two main things:
- -- -
First Idea!
- - - - - -A user asks an agent "is the LED on?".
-
- The agent does not have anything in it's world model, meaning that it doesn't know any LED components.
-
- Because of this the agent will fail to understand the meaning of "the LED".
The agent has two options:
- -LED connected to a button.
- -User: Turn on the LED
-
- Agent: ( Presses button )
LED connected to a button A.
- -User: Will pressing button B turn on the LED?
-
- Agent: < Some negative answer >
User: Turn the LED on
-
- Agent: ( Flicks switch )
::: Simple, so it's a good starting point to start understanding time linked events
- - - -User: Turn the LED on
-
- Agent: ( Presses button )
::: Seems simple but what if you worry about understanding event sequences and timed linked events
- - - -User: Do these circuits behave in the same way?
-
- Agent: < Some positive answer >
User: Turn the LED on
-
- Agent: ( flickes 1st switch), ( flickes 2nd switch )
::: This mentions sequencial actions planning but that's not the point, the point is, if the agent already knew about AND gate behaviour in it's WM, could or should it be able to abstract the 3 NOT system to an AND in it's WM?
- - - -User: Turn the LED on
-
- Agent: ( flickes 1st switch), ( flickes 2nd switch )
::: Think about act sequentiation, may not need time understanding
- - - -The solution to this scenario is made of several sequenciated parts, can't do all at once!
- - - - -- -
First Idea!
- - - - - -User: Turn the LED on - - Agent: ( flickes 1st switch), ( flickes 2nd switch ), ( flickes 3rd switch )
- -::: Some bigger sequential examples with more complex reasoning
- - - -