Transcript
Page 1: Incorporating Tutorial Strategies Into an Intelligent Assistant

Incorporating Tutorial Strategies Into an Intelligent Assistant

Jim R. Davies, Neal Lesh, Charles Rich, Candace L. Sidner, Abigail S. Gertner,

Jeff Rickel

Page 2: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

2

Organizations Involved

• College of Computing, Georgia Institute of Technology (Davies)

• Mitsubishi Electric Research Labs (Lesh, Rich, Sidner)

• The MITRE Corporation (Gertner)• USC Information Sciences Institute

(Rickel)

Page 3: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

3

Motivating Example

• Long camping trip

• Someone tutors you on how to set up a tent

• As time passes, that tutor becomes an assistant

Page 4: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

4

Research Goal

• To show that assisting and tutoring are two points on the same spectrum by building an agent that can transition between both behaviors.

Page 5: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

5

Intellectual History

• Collaborative Assisting Agent (COLLAGEN)– assists with software applications

• COLLAGEN generated interest in learning to use applications.

• We are extending COLLAGEN so it can tutor and built an agent called Triton.

Page 6: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

6

COLLAGEN (COLLaborative Agent)

• Middleware

• Discourse theory of collaboration

• Shared plan theory• Rich, C. and Sidner, C. L. (1998). COLLAGEN: A

collaboration manager for software interface agents. User Modeling and User-Adapted Interaction, Vol. 8, No. 3/4, 315—350.

Page 7: Incorporating Tutorial Strategies Into an Intelligent Assistant
Page 8: Incorporating Tutorial Strategies Into an Intelligent Assistant
Page 9: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

9

Assistant Scenario

• User: Let’s plan the second leg of the trip.

• Assistant: Okay.

• User: [selects Washington] What next?

• Assistant: [points to San Francisco] I will select San Francisco.

• User: Okay.

• Assistant: [selects San Francisco]

• User: What next?

• Assistant: Let’s remove an airline constraint.

Page 10: Incorporating Tutorial Strategies Into an Intelligent Assistant
Page 11: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

11

ITS Background

• Domains: physics and algebra

• Domain Knowledge

• Student Model

• Pedagogical knowledge• Wenger, E. (1987). Artificial Intelligence and

Tutoring Systems. Morgan Kaufmann Publishers, Inc. Los Altos, CA.

Page 12: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

12

Tutoring Scenario (part 1)

• User: What next?

• Triton: Please specify the route of the trip.

• User: What next?

• Triton: Select Atlanta and then select Boston.

• User: What next?

• Triton: Please select Atlanta.

• User: What next?

• Triton: [points to Atlanta with cursor] Please click on the dot near Atlanta on the map.

Page 13: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

13

Tutoring Scenario (cont.)

• User: [clicks on Atlanta] What next?

• Triton: Please select Boston.

• User: [clicks on Dallas]

• Triton: Oops! What you did was select Dallas; what you should have done was select Boston. I will bring you back to the previous state by clicking undo. [undo] Please select Boston.

• User: [clicks on Boston]

• Triton: Great job!

• User: What next?

• Triton: Select the dates and times of travel.

Page 14: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

14

Goal of this research

• Add to COLLAGEN to give it tutoring behaviors

• Adjustable parameters

• Triton

Page 15: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

15

Worked Examples

• Worked examples are kinds of recipes

Page 16: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

16

The User is Not Always Right

• Determining when a task is completed

• Responding to Errors

Page 17: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

17

Responding to Errors

• Intervene after n unrecognizable actions

• What the intervention looks like:– Say what the student did– Say what the student should have done– Undo to get to previous state

Page 18: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

18

Tutors are not Maximally Helpful

• Because of learning goals

• Waiting for Student Initiative

• Suggesting actions without doing them

• Explaining

• Demonstrating

• Pointing

Page 19: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

19

Learning Goals

• Usually task goals are in service of learning goals, but not always

Page 20: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

20

Waiting For Student Initiative

• In assisting, always try to help

• In tutoring, get student to try herself

Page 21: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

21

Suggesting Actions Without Doing Them

• Should you force the user to do all actions?

• Agent suggests doing, but doesn’t do.

Page 22: Incorporating Tutorial Strategies Into an Intelligent Assistant

Explaining

Page 23: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

23

Explaining (cont.)

• Composite Actions– list of task descriptions

• Primitive Actions– application-level description of what to do

on screen

• Stored as explanation recipes

Page 24: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

24

Demonstrating

• Behavior:– Do a sequence of actions– Undo them

• Stored as explanation recipes

Page 25: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

25

Pointing

• In assisting, point when proposing

• In tutoring, point when explaining a primitive

Page 26: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

26

Summary of Parameters

• When to intervene after error detection

• Who defaults to do actions

• When to point

Page 27: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

27

Contributions

• Middleware

• Use of recipes as a single representational structure for:– abstract actions– utterances– explanations– demonstrations

Page 28: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

28

Conclusions

• This work bridges the gap between tutoring and assisting

• Smoothly transitions between them

• Based on collaborative discourse theory

Page 29: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

29

Future Work

• Student Model

• Automatic Shifting between assisting and tutoring

Page 30: Incorporating Tutorial Strategies Into an Intelligent Assistant

http://www.cc.gatech.edu/~jimmyd/research/triton/

30

URLs

• http://www.cc.gatech.edu/~jimmyd/research/triton/• http://www.merl.com/


Recommended