giving agents artificial intuition - trinity universityyzhang/reu/2008/program/reports/final... ·...

16
Giving Agents Artificial Intuition Jason Leezer Computer Science REU

Upload: others

Post on 13-Jul-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

Giving Agents Artificial Intuition

Jason LeezerComputer Science REU

Page 2: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

What is an Agent?- Perceives it's environment- Reasons- Acts upon it's environment

What is Deliberation?- Slow- Effortful- Deliberately Controlled

What is Intuition?- Fast- Associative- Automatic

Page 3: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

A bat and a ball cost $1.10 in total. The bat costs $1 more than the ball.How much does the ball cost?

Shane Frederick (personal communication, April 2003)

Page 4: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

Ball Costs $0.10Bat Costs $1.00

Right?

A+

Page 5: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

Wrong!

(Ball) (Bat) (Total) $0.05 + $0.05 + $1.00 = $1.10

Page 6: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

Motivation

Reproducing Human Behaviour is Difficult- Can't use utility theory- Humans don't always act rational

Human Simulations Could Aid Social Sciences- Allow situations/scenarios they would be

unable to create in the natural world- Allow experiments to be performed on a

massive scale in a much shorter time frame

Page 7: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

How to Reproduce IntuitionIntuition is based strongly on accessibility – Khaneman

Accessibility - “the ease with which particular mental contents come to

mind”

Our Method:If an agent's current state is close to a state in

memory, then the best action can be inferred from the past experience

Page 8: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

Intuitive Decisions:- Restrict amount and type of information used- Information accessible from current state

Deliberative Decisions:- No restriction- All information is accessible

Q – Learning:- Assigns a “Q-Value” to every state action pair- In each state the action that corresponds to the

highest value is chosen.- Q-Value's are updated after each state transition- α is the learning rate, r is the reward, γ is the

discount factor

Page 9: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

Intuition Deliberation

Yes No

Am I Familiar With This?

Satisfying Choice

The Environment

Actions

Perception

Page 10: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

I'll take x, you get 10 - x

Accept:Person A get x,

Person B gets 10 - xReject:

Both get $0

Person A

Person B

The Ultimatum Game

Page 11: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

How people play the game:

Human Players

Rational Players

Page 12: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

Rational Players

0 1 2 3 4 5 6 7 8 9 100

20

40

60

80

100

120

140

BothAll IntuitionAll Deliberation

Split Value

Num

ber t

ime

s ac

cept

ed

ResultsIntuition Deliberation Agents

Human Players

Page 13: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

Rational Players

Human Players

ResultsIntuition Deliberation Agents

Page 14: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

Rational Players

Human Players

ResultsIntuition Deliberation Agents

Page 15: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

Conclusion & Future Work- While difficult, human behavior can be reproduced-Both Intuition and Deliberation decision making needed-- The learning algorithm plays a big role

What's Next?

-Investigate the effectiveness of other learning algorithms

-Restrict other types of information

-Reproduce human social behavior

Page 16: Giving Agents Artificial Intuition - Trinity Universityyzhang/reu/2008/Program/reports/Final... · 2008-12-13 · Motivation Reproducing Human Behaviour is Difficult - Can't use utility

Thanks to

Dr. Zhang

Dr. Lewis Dr. Drennon