Table of Contents

ATAI-20 Main

ATAI-20 Reykjavik University



Final Project


Description


For the final project you are asked to implement your own original task into the TestChamber (see Introduction to the Final Project), implemented by you and your collborator.

Your task must have:

  1. A causal chain with a length of at least 3 (E.g. Get the Key, open the door, activate the switch, bake some pizza, eat the pizza) - If you already have done more than 3 or feel like you have enough time to do more you can do so, everything over 3 will give you bonus points.
  2. Multiple solutions independent of each other:
    1. at least two of them must be mutually exclusive
    2. at least two of them must include different objects that need to be handled by NARS
    3. at least two of them must have a causal chain of at least 5
  3. At least one own object that is included in the task (e.g. WireCutter to stop active wires):
    1. For this you might have to include a new Action Operator for NARS
    2. You might have to change parts of the existing code to include the new actions and objects
  4. An own set of predefined knowledge which can be loaded in the GUI to realise NARS solving the task. (Add a new “knoMenu” to the EditorPanel.java file in the grid2d folder of the opennars-lab source code which includes knowledge of the new object and/ or additional information which can be useful for NARS) You can use this set of predefined knowledge to teach NARS about the different causalities instead of forcing actions repeatedly
  5. A “training scenario” must be provided in which NARS learns about the single causal connection (e.g. Key→Door and switch→oven and oven→pizza) independently, such that it can learn before trying to complete the full task by trying things - or being shown - in the training room.

Write a report on your findings (max. 5 pages + appendix). For this you can choose to focus on one or both of the following:

  1. Analyze the abduction of NARS to abduct causal chains from single causal connections. Some questions that you may want to answer with respect to that might include:
    1. What role, if any, does abduction play in NARS’ behaviour?
    2. What are the abduction mechanisms that NARS is deploying?
    3. Is there a limit to the abduction abilities of NARS (e.g. length of causal chain)?
  2. Analyze the induction of NARS to generate new rules which result in the possibility to solve a novel task. Some questions that you may want to answer with respect to that might include:
    1. What role, if any, does induction play in NARS’ behavior?
    2. What might be the limits of NARS’ induction in this world? Is it limited in some way (by the reasoning methods of NARS or the task-environment)?