Georgeon, Olivier L. :: IDEAL MOOC

1. Teaser
2. Lecture 1: Embodied Paradigm
Backlinks
- perception being solution
  - (Systems with knowledge base)
- developmental skill aquisition with general purpose actuator
  - (Footnotes)

1. Teaser

https://projet.liris.cnrs.fr/ideal/mooc/

2. Lecture 1: Embodied Paradigm

2.1. main idea

agent.input != agent.perception(environment)

agent is not ~~passive~~ observer of reality
agent construct perception of reality through active interaction
agent be part of reality for the active interation to happen specifically, agent’s action is taken into account in the perception process. they are aware of the current being the consequence of their actions

my interpretation of embodiment is making a principle more vivid through trials and examples.

2.2. experiment-centered design

2.2.1. traditional view - input = environment

sensor data is a representation of the environment.

2.2.2. embodied view - (experiment, result) = environment

the representation/perception of the environment is a cause-effect relation(function). The experiment is initiated by the agent input(sensor data) is result of the experiment(action) and the state(true, comprehensive state of the environment, not accessible)

2.2.3. implementation - function

or precisely, mapping.

in traditional view, input to the agent is implemented with o = f(s), where s is the true state of the environment of infinite length and not accessible. function f map the comprehensive state of environment to a few values that is outputed by the sensors.

in embodied view, such input(sensor data) is regarded as a similar mapping, only it is affected by the comprehensive state and the experiment action, hense r = f(e,s)

2.2.4. rudimentary

2.3. stateless environment

The perception of the environment could be: action a1 would always yield result r2, rather than a set of states of sensory readings.

Note that this comply with the pattern of discrete functionalities of consciousness not necessarily implemented by highly complex biochemical process involving lots of values and particles.

2.4. The agent is the environement

in the demo on the website, the distinction of agent code and environment code is not obivous. Every bit of code is in fact, the agent, including the perception of the environment: the agent is not interacting with a simulated environemnt that is represented as a state of values in the computer – that representation is also part of the agent.

2.5. stream of intelligence

intelligence is best represented as a stream of thoughts(literally). a stream of intelligence could be awareness of the experiment and the results.

2.6. how would the agent deal with multiple interactions

how exactly? say it encounters e1,r1 and e1,r2. What would it anticipate on e1 then? would it take some kind of previous result into consideration?

2.7. literature

2.8. working log

0:00:00: starting to read syllabus. the first lesson is the embodied paradigm. so let’s do it!
0:07:25: wasting time on getting literatures. Which is not important because I have readings at hand that are not read.
0:21:01: texted with PP for 4 minutes.
0:23:12: gothic kid talk.
1:19:30: I checked all the code in projects/IDEAL-mooc/lesson1/src. I make some notes about compiling and runing java programs.
1:20:22: I’m satisfied and tired.

2.8.1. conclusion of this session

I diverted 3 times. The log make me feel good and it did record things I have now forgotten. It reminded my that I’m just supposed to dive into this paradigm, not looking other stuff, making comments everywhere and post tensions (to impress). and I do make my own organisations and interpretations where I see fit, which I’m happy about.

The workload of this mooc series is light. it is about 1 hour each session, for 7 sessions.

Backlinks

perception being solution

(Systems with knowledge base)

Those systems typically stores their perceptions, their experiences in a knowledge base, so as to use it in later reasoning.

In Georgeon, Olivier L. :: IDEAL MOOC, an approach of recognizing regularities of interactions in past experience and use that regularity to reason of a best action to perform is discussed.

In NARS, observations are stored in database and queried with its Non-Axiomatic Logic when reasoning for an action to be done to achieve a goal.

developmental skill aquisition with general purpose actuator

(Footnotes)

3 key figures are from French: Pierre-Yves Oudeyer, Jean-Christophe Baillie, Georgeon, Olivier. scattering around the world; usually coorporate with psychology department on the contrary of increase of efficiency; optimization in a very narrow problem such as “this image is an apple” Implementation of DEvelopmentalAI Learning