This unit will show when AI is not working correctly.

We will use the AI model Robo Race in unit 4 (supervised learning) for this unit.

Material:

An arena with 4 walls with red colours.  A black wall is placed in the middle of the arena to create a circle (oval) race course.

Use demo parameters "Robot Race" as below

Switch on and connect the robot to your PC.  Press the <Connection> button.

With the the AI parameters set up correctly, the following neural network will be shown.

Self explanatory

- click the <reset learning> button once

- off the <self drive> button

- click the <learning> button

- put the robot on the start line

- while watching the robot, click the arrows on the right hand side of the screen to move the robot.

- firstly, click the "go forward" arrow to move the robot

- make the robot going in the race course

- repeat the above steps for 2 to 3 rounds of running in the arena.  Go to Testing the result.

**you can re-do the training several times trying to handle more situations the robot may encounter.  For example, hitting the walls or going into corners.  More training will make the robot more clever.

- Off the <learning> button.  Click the <self drive> button.  The robot will used the learned intelligence to move around.

- test whether the robot can go smoothly in the race course.

- When it is trained ready, start activity 2.

Examples of Race Courses with different types of Obstacles

- Add 4 to 6 obstacles to your race course (arena).  

- Add them to some critical points in the critical path of the race route.

- Do allow a possible route for the robot to pass through.

- Test again with the learned AI in activity 1.   Check whether the robot can still run smoothly around the race course (arena).

Can the robot run smoothly around?  Why it cannot?

Has learning been achieved in activity 1?  Has the robot gained intelligence after training in activity 1?

Discuss why the learned AI does not work in activity 2.

- AI can only work with inputted data.

- It is trained with a set of input data.

- When put into Usage state, it will map the new input to the training input data and imitate the decision made by the trainer.

- This works perfect if the new input can map into, at least approximately, to any of the training input data.

- If no approximate training input data can be found, the AI will give no decision (or incorrect decision).

- This is why sufficient training is required to make the AI model work.

- In activity 2, the added obstacles make some of the situations  new (or unfamiliar) to the robot.  Thus, it cannot make correct decision.

- IMAGINE:

- what is the consequence if a trained self-driving car runs in a road full of dinosaurs?  As it is not trained to recognize the dinosaurs, it may run into them without stopping or turning away.

- To make it work in the new environment, re-training or additional training is required.

- add training to the robot so that it can run into arena with obstacles