Dynamics of Rule Revision and Strategy Revision in Legislative Games - PowerPoint PPT Presentation

1 / 22
About This Presentation
Title:

Dynamics of Rule Revision and Strategy Revision in Legislative Games

Description:

Buying books for the purpose of donating them. Donating cars that do not run ... Allowing deduction for only car's value realized on sale ... – PowerPoint PPT presentation

Number of Views:79
Avg rating:3.0/5.0
Slides: 23
Provided by: RPL6
Category:

less

Transcript and Presenter's Notes

Title: Dynamics of Rule Revision and Strategy Revision in Legislative Games


1
Dynamics of Rule Revision and Strategy Revision
in Legislative Games
  • Moshe Looks
  • Ronald P. Loui
  • Barry Cynamon
  • Washington University in St. Louis, USA

2
Basic Idea
  • Legislators don't always
  • Say what they mean nor
  • Mean what they say
  • Hart there is a limit to the use of language
  • Sometimes the rule-makers don't even agree
  • Be deliberately vague
  • Toss the issue to the courts

3
Legislation is Worth Studying
  • Rules change
  • Often they change in response to
  • Agents behaving badly
  • Agents discovering unintended strategies
  • North there is institutional learning
  • Legislate to perform strategy extinction

4
How to Study Legislative Institutional Dynamics?
  • Multi-agent systems simulation
  • Can we build a model that exhibits the
    interesting phenomena?
  • Agent modeling
  • Institutional modeling
  • Plausible dynamical modeling
  • Would anyone (outside AI) be able to work with
    rules
  • As text?
  • As logic fragments?
  • As procedures?

5
Idea!
  • All edict takes the form of
  • an objective function on k variables
  • to be maximized
  • Legislative revision change of function
  • Legislative abridgement projection
  • Onto subspace
  • I.E., use only a subset of the variables

6
Target Phenomenon I
  • Tenure-granting colleges often publish rules
  • Will count journal publications
  • Will count student evaluations of teaching
  • Will count amounts of external research funding
  • Legitimately interested in
  • Productivity
  • Intellectual impact
  • Teaching ability
  • Published criteria
  • Observable
  • Apparently precise

7
Target Phenomenon I
  • Those seeking tenure subvert the spirit of the
    rules by
  • Joining long co-author lists
  • Reporting research in minimal-publishable-units
  • Avoiding teaching difficult courses
  • Giving inflated grades
  • Doing research for the sake of funding
  • Adding their names as Co-PI to big projects
  • In the worst case, there is misdirection
  • Papers written for the resume, not for the
    scholarship
  • Teaching aimed at good feedback, not long-term
    student growth
  • Research aimed at getting funding, not
    intellectual impact

8
Target Phenomenon I
  • Tenure Committees Respond by
  • Normalizing papers by author count
  • Evaluating the five best publications
  • Measuring student performance objectively
  • Capping funding amounts that can be reported
  • Requiring co-PI's to show students supported on
    funds
  • Candidates for tenure respond in situ to new
    reqirements
  • They don't toss their resumes start from scratch

9
Target Phenomenon II
  • Tax regulations seek to encourage charitable
    deductions including (as cases are decided)
  • Donations of books to book sales
  • Donations of cars to non-profit organizations
  • Donations to arts performance organizations

10
Target Phenomenon II
  • Taxpayers respond by
  • Buying books for the purpose of donating them
  • Donating cars that do not run
  • Donating to performance companies in exchange for
    free tickets
  • Over time, through legislative misdirection
  • agents optimize the wrong function

11
Target Phenomenon II
  • Tax regulators respond by
  • Requiring receipts showing purchase amounts
  • Allowing deduction for only cars value realized
    on sale
  • Reducing amounts of donations by any quid-pro-quo
    considerations
  • Taxpayers respond again by donating less
  • New abridgement repairs short-term misdirection
  • Successful strategy extinction or scenario
    extinction
  • In time, different legislative misdirection

12
Our Model
  • There is a veridical value function
  • V(x, y, z, )
  • known to the legislators
  • At any time, there is an abridgement of V
  • A(x, z)
  • A function of fewer variables
  • More generally, use a projection of V
  • public

13
Our Model II
  • At any time, an agents strategy/position is
  • A point in V-space
  • (9, 10, 1, )
  • With de jure value A(9, 1)
  • With de facto value V(9, 10, 1, )
  • Agents occupy admissible positions
  • E.g., declare that (0, 0, 0, ) is prohibited
  • Admissibility is not known to all
  • Admissibility is discovered through search
  • Admissibility can also change with time

14
Our Model III
  • A legislature can respond
  • Change the set of admissible positions
  • Change the function A (OUR FOCUS)
  • An agent can respond
  • Search for point with higher A-value
  • Learn from other agents
  • Where are the high A-values
  • Where are the admissible points

15
Simulations
  • Example Greedy Non-Omniscient Agents
  • Look at neighborhood around current point
  • Move to highest point with highest A-value
  • Bound how much they can move
  • Example Imitative Non-Omniscient Agents
  • Move toward average of others
  • If it is better than where you are
  • Bound how much they can move

16
(No Transcript)
17
Simulations
  • Example Incentivizing the Average
  • Revise A so AverageAgent maximizes V
  • Project V onto line between global opt and avg
    position
  • Mix(A,A') to bound difference
  • Example Extinguishing the Worst
  • Find d maximum V ? A
  • Choose A' to minimize A'(d)
  • Diff(A,A') is bounded

18
Simulations
  • How quickly can the legislator act?
  • Dominant legislature
  • Revises as quickly as agents, bounds generous
  • Parity
  • Revises as quickly as agents, bounds on par
  • Dominant agents
  • Agents revise more quickly, bounds on par

19
(No Transcript)
20
Background Functions
  • In addition to V, A, add
  • B which models
  • The public "spirit" of the laws
  • E.g., B is a time-average of A over a period
  • Agents who maximize B are more robust to changes
    of A
  • B contains additional knowledge about A

21
Background Functions
  • Taxonomy of agents
  • A-maximizer at the expense of B is a rat
  • A-maximizer s.t. high B is a literalist
  • B-maximizer s.t. high A is a wolf
  • B-maximizer at the expense of A is an idealist
  • Imitative A-maximizer is a sheep
  • Novel A- but not B-maximizer is an exploiter
  • Novel A- B-maximizer is a producer

22
Conclusions
  • Result? A model with rich and appropriate
    dynamics
  • Main contribution depicting legislative
    phenomena in mathematical economics (or ICMAS)
    clothing
  • Would like A-B-V triad to be memorable
  • Main idea there must be a reason to revise
  • Agents learn
  • Legislative misdirection accrues
  • Much legislation is repair of old abridgement
Write a Comment
User Comments (0)
About PowerShow.com