A New Model of Negotiation Based on Pessimism and Punishment

1
A New Model of Negotiation Based on Pessimism
and Punishment

• Ronald P. LouiWashington UniversitySt.
  LouisCAEPIA 2003

2
A New Model of Negotiation Based on Pessimism
and PunishmentPreliminary Remarks

• CAEPIA 2003

3
Preliminary Remarks

1985-2000 Logic of Argument
1987 - Defeat among Arguments 1988 - Workshop on
Defeasibility 1989 - G. Simari's thesis 1991 -
NATHAN in C 1992 - H. Prakken's thesis 1993
(1997) - Process Policy 1993 - G. Vreeswijk's
thesis 1993 - T. Gordon's thesis 1995 - B.
Verheij's thesis 1997 - A. Lodder's thesis 2000 -
ACM Computing Surveys
4
Preliminary Remarks

1990-2000 Logic of Arguments Cases
1991 - LMNOP in LISP 1992 - Argument Game 1993 -
Policies and Precedents 1995 - Rationales 1997 -
Room5 in GAWK/CGI 1998 - Prakken-Sartor 1999 -
Bench-Capon 2003 - Johnston-Governatori
5
(No Transcript)
6
Preliminary Remarks
"Why does there have to be a winner? Why can't
you work on negotiation instead of argument?"
-- Prof. Anne Johnstone "University of Maine
assistant professor of computer science Anne
Johnstone died of complications resulting from
lung cancer on February 28, 1995. After
completing her PhD in Artificial Intelligence at
the University of Edinburgh approx. 1990, Anne
took a position as senior research associate in
the computer science department at Washington
University in St. Louis, and while there
established a successful research program in
natural language processing, funded by
Southwestern Bell Technology Resources."
(http//www.umcs.maine.edu/larry/csdept/johnstone
.html) This invited talk is dedicated to the
memory of my faculty colleague.
7
Preliminary Remarks
Argument and Negotiation 1985-87 - K. Sycara's
thesis 1993 - G. Ferguson's thesis 1994 -
Computational Dialectics Workshop 1995 - Mark
Foltz's (WU/MIT) u. thesis 1996 -
Parsons-Jennings 1997 - Diana Moore's
(WU/NSA/EnGarde) u. thesis 1998-2001 -
Jennings-Parsons-Noriega, Sierra-Jennings,
Jennings-Parsons-Noriega-Sierra,
Parsons-Sierra-Jennings, Jennings-Faratin-Lomuscio
-Parsons-Wooldridge-Sierra, etc. Main Idea
One may argue for or against a proposal (a
proposed settlement).
8
Preliminary Remarks
Argument and Negotiation 1985-87 - K. Sycara's
thesis case-based problem-solving for
labor-management disputes argument 1993 - G.
Ferguson's thesis 1994 - Computational Dialectics
Workshop 1995 - Mark Foltz's (WU/MIT) u.
thesis 1996 - Parsons-Jennings 1997 - Diana
Moore's (WU/NSA/EnGarde) u. thesis 1998-2001 -
Jennings-Parsons-Noriega, Sierra-Jennings,
Jennings-Parsons-Noriega-Sierra,
Parsons-Sierra-Jennings, Jennings-Faratin-Lomuscio
-Parsons-Wooldridge-Sierra, etc. Main Idea
One may argue for or against a proposal (a
proposed settlement).
9
Preliminary Remarks
Argument and Negotiation 1985-87 - K. Sycara's
thesis 1993 - G. Ferguson's thesis
"mixed initiative" planning mixture of argument
negotiation 1994 - Computational Dialectics
Workshop 1995 - Mark Foltz's (WU/MIT) u.
thesis 1996 - Parsons-Jennings 1997 - Diana
Moore's (WU/NSA/EnGarde) u. thesis 1998-2001 -
Jennings-Parsons-Noriega, Sierra-Jennings,
Jennings-Parsons-Noriega-Sierra,
Parsons-Sierra-Jennings, Jennings-Faratin-Lomuscio
-Parsons-Wooldridge-Sierra, etc. Main Idea
One may argue for or against a proposal (a
proposed settlement).
10
Preliminary Remarks
Argument and Negotiation 1985-87 - K. Sycara's
thesis 1993 - G. Ferguson's thesis 1994 -
Computational Dialectics Workshop notation
protocol for proposal, threat, inform, justify,
etc. as speech acts 1995 - Mark Foltz's (WU/MIT)
u. thesis 1996 - Parsons-Jennings 1997 - Diana
Moore's (WU/NSA/EnGarde) u. thesis 1998-2001 -
Jennings-Parsons-Noriega, Sierra-Jennings,
Jennings-Parsons-Noriega-Sierra,
Parsons-Sierra-Jennings, Jennings-Faratin-Lomuscio
-Parsons-Wooldridge-Sierra, etc. Main Idea
One may argue for or against a proposal (a
proposed settlement).
11
Preliminary Remarks
Argument and Negotiation 1985-87 - K. Sycara's
thesis 1993 - G. Ferguson's thesis 1994 -
Computational Dialectics Workshop 1995 - Mark
Foltz's (WU/MIT) u. thesis 1. Cooperative
protocol level reciprocal progress 2.
Cooperative protocol level immediate response
3. Cooperative protocol level arguing for
proposals 4. Uncooperative protocol level
using threats 1996 - Parsons-Jennings 1997 -
Diana Moore's (WU/NSA/EnGarde) u.
thesis 1998-2001 - Jennings-Parsons-Noriega,
Sierra-Jennings, Jennings-Parsons-Noriega-Sierra,
Parsons-Sierra-Jennings, Jennings-Faratin-Lomuscio
-Parsons-Wooldridge-Sierra, etc. Main Idea One
may argue for or against a proposal (a proposed
settlement).
12
Preliminary Remarks
Argument and Negotiation 1985-87 - K. Sycara's
thesis 1993 - G. Ferguson's thesis 1994 -
Computational Dialectics Workshop 1995 - Mark
Foltz's (WU/MIT) u. thesis 1996 -
Parsons-Jennings uses Fox's argument logic
for formalizing arguments about proposals 1997 -
Diana Moore's (WU/NSA/EnGarde) u.
thesis 1998-2001 - Jennings-Parsons-Noriega,
Sierra-Jennings, Jennings-Parsons-Noriega-Sierra,
Parsons-Sierra-Jennings, Jennings-Faratin-Lomuscio
-Parsons-Wooldridge-Sierra, etc. Main Idea
One may argue for or against a proposal (a
proposed settlement).
13
Preliminary Remarks
Argument and Negotiation 1985-87 - K. Sycara's
thesis 1993 - G. Ferguson's thesis 1994 -
Computational Dialectics Workshop 1995 - Mark
Foltz's (WU/MIT) u. thesis 1996 -
Parsons-Jennings 1997 - Diana Moore's
(WU/NSA/EnGarde) u. thesis list of speech
acts in negotiation dialogue, including
o change of focus o request for
proposal o "settlement" as a
linguistic object (a contract) 1998-2001 -
Jennings-Parsons-Noriega, Sierra-Jennings,
Jennings-Parsons-Noriega-Sierra,
Parsons-Sierra-Jennings, Jennings-Faratin-Lomuscio
-Parsons-Wooldridge-Sierra, etc. Main Idea One
may argue for or against a proposal (a proposed
settlement).
14
Preliminary Remarks
Argument and Negotiation 1985-87 - K. Sycara's
thesis 1993 - G. Ferguson's thesis 1994 -
Computational Dialectics Workshop 1995 - Mark
Foltz's (WU/MIT) u. thesis 1996 -
Parsons-Jennings 1997 - Diana Moore's
(WU/NSA/EnGarde) u. thesis 1998-2001 -
Jennings-Parsons-Noriega, Sierra-Jennings,
Jennings-Parsons-Noriega-Sierra,
Parsons-Sierra-Jennings, Jennings-Faratin-Lomuscio
-Parsons-Wooldridge-Sierra, etc. Main Idea
One may argue for or against a proposal (a
proposed settlement).
15
Preliminary Remarks
Argument and Negotiation 1985-87 - K. Sycara's
thesis 1993 - G. Ferguson's thesis 1994 -
Computational Dialectics Workshop 1995 - Mark
Foltz's (WU/MIT) u. thesis 1996 -
Parsons-Jennings 1997 - Diana Moore's
(WU/NSA/EnGarde) u. thesis 1998-2001 -
Jennings-Parsons-Noriega, Sierra-Jennings,
Jennings-Parsons-Noriega-Sierra,
Parsons-Sierra-Jennings, Jennings-Faratin-Lomuscio
-Parsons-Wooldridge-Sierra, etc. Main Idea
One may argue for or against a proposal (a
proposed settlement).
16
Preliminary Remarks
Argument and Negotiation 1970's - R.
Fisher-(Ury-Patton) "principled
negotiation" 1998-2001 - Jennings-Parsons-Norieg
a, Sierra-Jennings, Jennings-Parsons-Noriega-Sierr
a, Parsons-Sierra-Jennings, Jennings-Faratin-Lomus
cio-Parsons-Wooldridge-Sierra, etc. UN-ANSWERED
QUESTION What cost is incurred by the party to
a negotiation who refuses the consequences of
rational argument?
17
Preliminary Remarks
Search and Negotiation 1991 - pre-Lesser
Sandholm 1992-93 - S. Lander's thesis 1993 - CMU
talk 1994 - Strategic directions in AI (ACM
Computing Surveys) 1996 - T. Sandholm's
thesis 1996 - Kay Hashimoto (Harvard/FJC) 1997 -
Diana Moore's (WU/NSA/EnGarde) u.
thesis Main Idea Payoffs (in strategic
form) are not fixed, but can change with search.
18
Preliminary Remarks
Search and Negotiation 1991 - pre-Lesser
Sandholm "strategy for reducing overall
computation costs" by trading tasks,
TSP's 1992-93 - S. Lander's thesis 1993 - CMU
talk 1994 - Strategic directions in AI (ACM
Computing Surveys) 1996 - T. Sandholm's
thesis 1996 - Kay Hashimoto (Harvard/FJC) 1997 -
Diana Moore's (WU/NSA/EnGarde) u. thesis Main
Idea Payoffs (in strategic form) are not
fixed, but can change with search.
19
Preliminary Remarks
Search and Negotiation 1991 - pre-Lesser
Sandholm 1992-93 - S. Lander's thesis "uses
conflict as a source of control information for
directing search" 1993 - CMU talk 1994 -
Strategic directions in AI (ACM Computing
Surveys) 1996 - T. Sandholm's thesis 1996 - Kay
Hashimoto (Harvard/FJC) 1997 - Diana Moore's
(WU/NSA/EnGarde) u. thesis Main Idea
Payoffs (in strategic form) are not fixed, but
can change with search.
20
Preliminary Remarks
Search and Negotiation 1991 - pre-Lesser
Sandholm 1992-93 - S. Lander's thesis 1993 - CMU
talk consider the strategic form payoffs to
be heuristic! 1994 - Strategic directions in AI
(ACM Computing Surveys) 1996 - T. Sandholm's
thesis 1996 - Kay Hashimoto (Harvard/FJC) 1997 -
Diana Moore's (WU/NSA/EnGarde) u. thesis Main
Idea Payoffs (in strategic form) are not
fixed, but can change with search.
21
Preliminary Remarks
Search and Negotiation 1991 - pre-Lesser
Sandholm 1992-93 - S. Lander's thesis 1993 - CMU
talk 1994 - Strategic directions in AI (ACM
Computing Surveys) "Instead of starting with
a matrix of payoffs to each player
introduce a structure that can be altered through
computation" 1996 - T. Sandholm's thesis 1996 -
Kay Hashimoto (Harvard/FJC) 1997 - Diana Moore's
(WU/NSA/EnGarde) u. thesis Main Idea
Payoffs (in strategic form) are not fixed, but
can change with search.
22
Preliminary Remarks
Search and Negotiation 1991 - pre-Lesser
Sandholm 1992-93 - S. Lander's thesis 1993 - CMU
talk 1994 - Strategic directions in AI (ACM
Computing Surveys) 1996 - T. Sandholm's thesis
Negotiation Among Self-Interested
Computationally Limited Agents 1996 - Kay
Hashimoto (Harvard/FJC) 1997 - Diana Moore's
(WU/NSA/EnGarde) u. thesis Main Idea
Payoffs (in strategic form) are not fixed, but
can change with search.
23
Preliminary Remarks
Search and Negotiation 1991 - pre-Lesser
Sandholm 1992-93 - S. Lander's thesis 1993 - CMU
talk 1994 - Strategic directions in AI (ACM
Computing Surveys) 1996 - T. Sandholm's
thesis 1996 - Kay Hashimoto (Harvard/FJC)
each agent solves non-linear constrained
optimization, asks "why not", responds with
list of binding constraints 1997 - Diana Moore's
(WU/NSA/EnGarde) u. thesis Main Idea
Payoffs (in strategic form) are not fixed, but
can change with search.
24
Preliminary Remarks
Search and Negotiation 1991 - pre-Lesser
Sandholm 1992-93 - S. Lander's thesis 1993 - CMU
talk 1994 - Strategic directions in AI (ACM
Computing Surveys) 1996 - T. Sandholm's
thesis 1996 - Kay Hashimoto (Harvard/FJC) 1997 -
Diana Moore's (WU/NSA/EnGarde) u. thesis o
can search your problem OR partner's problem
o the simplest formal model of
utility-revision / joint problem-solving Main
Idea Payoffs (in strategic form) are not
fixed, but can change with search.
25
Preliminary Remarks
Search and Negotiation 1991 - pre-Lesser
Sandholm 1992 - K. Decker's thesis,
Sandholm-Lesser 1994 - Strategic directions in AI
/ ACM Computing Surveys 1996 - T. Sandholm's
thesis 1997 - Diana Moore's (WU/NSA/EnGarde) u.
thesis 1997 - Kay Hashimoto (Harvard/FJC). M
ain Idea Payoffs (in strategic form) are not
fixed, but can change with search.
26
Preliminary Remarks
Search and Negotiation OBSERVATION If anytime
search is focused on proposalsthen agreement is
easier to achieve! Main Idea Payoffs (in
strategic form) are not fixed, but can change
with search.
2 6 9 11
5 8 10 13
7 9 10 15
27
Preliminary Remarks
Search and Negotiation OBSERVATION If anytime
search is focused on proposalsthen agreement is
easier to achieve! Main Idea Payoffs (in
strategic form) are not fixed, but can change
with search.
2? 6? 9? 11?
5? 8? 10? 13?
7? 9? 10? 15?
28
Preliminary Remarks
Search and Negotiation OBSERVATION If anytime
search is focused on proposalsthen agreement is
easier to achieve! Main Idea Payoffs (in
strategic form) are not fixed, but can change
with search.
proposed
2? 6? 9? 11?
5? 8? 10? 13?
7? 9? 10? 15?
29
Preliminary Remarks
Search and Negotiation OBSERVATION If anytime
search is focused on proposalsthen agreement is
easier to achieve! Main Idea Payoffs (in
strategic form) are not fixed, but can change
with search.
proposed
2? 6? 9? 11?
5? 8? 11? 13?
7? 9? 10? 15?
search
30
Preliminary Remarks
Search and Negotiation OBSERVATION If anytime
search is focused on proposalsthen agreement is
easier to achieve! Main Idea Payoffs (in
strategic form) are not fixed, but can change
with search.
proposed
2? 6? 9? 11?
5? 8? 11? 13?
7? 9? 10? 15?
search
search
31
Preliminary Remarks
Search and Negotiation OBSERVATION If anytime
search is focused on proposalsthen agreement is
easier to achieve! Main Idea Payoffs (in
strategic form) are not fixed, but can change
with search.
proposed
2? 6? 9? 11?
5? 8? 11? 13?
7? 9? 10? 15?
search
search
search
32
Preliminary Remarks
Search and Negotiation OBSERVATION If anytime
search is focused on proposalsthen agreement is
easier to achieve! Main Idea Payoffs (in
strategic form) are not fixed, but can change
with search.
proposed
2? 6? 9? 11?
5? 8? 13? 13?
7? 9? 10? 15?
search
search
search
search
33
Preliminary Remarks
Search and Negotiation OBSERVATION If anytime
search is focused on proposalsthen agreement is
easier to achieve! Main Idea Payoffs (in
strategic form) are not fixed, but can change
with search.
proposed
2? 6? 9? 11?
5? 8? 16? 13?
7? 9? 10? 15?
search
search
search
search
search
accepted
34
Preliminary Remarks
Search and Negotiation OBSERVATION If anytime
search is focused on proposalsthen agreement is
easier to achieve! UN-ANSWERED QUESTION Is it
rational to accept a proposalif its heuristic
valuation looks goodJUST BECAUSEmore time was
spent searching to make it look good?
35
Preliminary Remarks
Search and Negotiation OBSERVATION If anytime
search is focused on proposalsthen agreement is
easier to achieve! UN-ANSWERED QUESTION Is it
rational to accept a proposalif its heuristic
valuation looks goodJUST BECAUSEmore time was
spent searching to make it look good?
Sandholm Look for deliberation equilibria!
36
Preliminary Remarks
Search and Negotiation
Sandholm Look for deliberation equilibria!
Loui Equilibrium is a silly idea!
Negotiation is a process. How about looking
at trajectories?
37
Preliminary Remarks
Search and Negotiation 1997 - Anne Jump
(Harvard) Negotiation game
38
(No Transcript)
39
(No Transcript)
40
(No Transcript)
41
(No Transcript)
42
(No Transcript)
43
(No Transcript)
44
(No Transcript)
45
(No Transcript)
46
(No Transcript)
47
(No Transcript)
48
(No Transcript)
49
(No Transcript)
50
(No Transcript)
51
(No Transcript)
52
(No Transcript)
53
(No Transcript)
54
(No Transcript)
55
(No Transcript)
56
(No Transcript)
57
(No Transcript)
58
(No Transcript)
59
(No Transcript)
60
(No Transcript)
61
(No Transcript)
62
(No Transcript)
63
(No Transcript)
64
A New Model of Negotiation Based on Pessimism
and Punishment

• CAEPIA 2003

thanks to F. Tohme, K. Larson, S. Braynov, J.
Nachbar, B. Dheeravongkit, L. Cai, K. Chin, R.
Bujans, M. Looks, E. Wofsey, J. Rosen, S.
Grubor, K. Ormsby, J. Badino, R. Pless, and M.A.
Clark.
65
Observation parties to a negotiation (can)
construct a probability distribution over
potential settlements
Party 1'saspiration
Party 2'saspiration
66
Observation parties to a negotiation (can)
construct a probability distribution over
potential settlements
Party 1'sproposals at t
Party 2'sproposals at t
67
Observation parties to a negotiation (can)
construct a probability distribution over
potential settlements
inadmissible(dominated)at t
inadmissible(dominated)at t
68
Observation parties to a negotiation (can)
construct a probability distribution over
potential settlements
In black admissiblesettlementsat
t(probabilityof agreement Is non-zero)
69
Observation parties to a negotiation (can)
construct a probability distribution over
potential settlements
Breakdown column
Breakdown row
70
Observation parties to a negotiation (can)
construct a probability distribution over
potential settlements
Breakdownwould occurhere (BATNA)
71
Observation parties to a negotiation (can)
construct a probability distribution over
potential settlements
1's aspiration
2's aspiration
72
Observation parties to a negotiation (can)
construct a probability distribution over
potential settlements
Breakdown (BATNA)
73
Observation parties to a negotiation (can)
construct a probability distribution over
potential settlements
Breakdown (BATNA)
74
Observation parties to a negotiation (can)
construct a probability distribution over
potential settlements
Rising probability of break down
75
Observation parties to a negotiation (can)
construct a probability distribution over
potential settlements
1's security level
1 would rather breakdown
2's security level
2 would rather breakdown
76
Observation parties to a negotiation (can)
construct a probability distribution over
potential settlements
Prob(bd) ?
Eu1s 51
Eu2s 49a 54(1-a)
77
Observation From probability, one may calculate
1. An expected utility given settlement
(Eus) and 2. An expected utility given
continued negotiation, Eu Eus (1 -
prob(bd)) u(bd) prob(bd)
Observation It may be rational to 1.
Extend an offer, o, if Eu lt u(o) 2. Accept
an offer, acc, if Eu lt u(acc),
acc ? offers(t) 3. Break down unilaterally
if Eu lt security
78
Probability here is OBJECTIVE PROBABILITY
1. It is based on data and induction 2.
Given the data (and method), prob is fixed
3. Cannot subjectively will the prob to be
different
(see Keynes)
Rationality here is CONSTRUCTIVE RATIONALITY
1. Action may be justified in a variety of
ways 2. Expected utility is one way to
justify action 3. These constraints are
one-sided (permissions, not
prohibitions)
(see Simon/Shafer/Tversky)
79
To construct prob(bdsituation), consider similar
past situations in which 1. Similar stakes
securities 2. Similar negotiating
partners 3. Similar progress
(see Kyburg)
The one-sided constraint is similar to
bargaining representative of a union
with proxy to act but not to censor union action
(probabilities are of independent union
action)But is here a general model of
rationality for a single mind
(see Minsky)
80
(Empirical) Claim. PessimismFor sufficiently
large ?, for all t prob(bd NP(t?)) gt
prob(bd NP(t))where NP(t) denotes
non-progress for a period tWhat is progress?
A non-trivial offer by the other partyWhat
does this mean?(at some granularity, the past
record implies that)If there are no offers, the
likelihood of breakdown rises
81
(Empirical) Claim. PessimismFor sufficiently
large ?, for all t prob(bd NP(t?)) gt
prob(bd NP(t))where NP(t) denotes
non-progress for a period tLinear Pessimism
prob(bd NP(t)) p tExponential
Pessimismprob(bd NP(t)) 1 - e-pt
82
Pessimism causes Eu to fall
Next offer is made at this timeand prob(bd)
resets to 0
Expectation starts to fall again
83
offers
reciprocated offers
84
Agreement reached as Eu lt u1
85
Whenever u(acc) gt security, acceptance
occurs before breakdown!
Best offer received
security
86
Would you accept an 11-cent offer if
yoursecurity were 10-cents?
Best offer received
security
87
Design Idea. Resentmentu(bd) security
resentment(t) s.t. NP(t)where NP(t)
denotes non-progress for a period tWhat is
resentment? 1. Dignity2. Pride3.
Investment in society4. Protection against
non-progressive manipulators
88
Design Idea. Resentmentu(bd) security
resentment(t) s.t. NP(t)where NP(t)
denotes non-progress for a period tWhat is
resentment? 5. Non-standard utility (process
or transactional utility)6. Re procedural
fairness, not substantive fairness7. A (speech
act) distinction acc_at_bd vs. BD!8. specific
(BD! by ME on YOU) vs. indifferent (BD!)
89
Design Idea. Resentmentu(bd) security
resentment(t) s.t. NP(t)where NP(t)
denotes non-progress for a period tlinear
resentment r(t) ?tsigmoid resentment r(t)
rmax(2/(1e-?t) -1)
90
Breaks down instead of agreeingbecause of rising
resentment
securityresentment
security
91
Actually accepts becauseresentment resets with
progress
Nontrivial progess
Resentment resets to zero
92
Agent breaks down before accepting
Resentment does not reset to zero
93
Variety of Behaviors

• Agent can wait, then offer
• Agent can wait, then accept
• Agent can wait, then break down
• Agent can accept immediately
• Agent can offer before accepting and vice versa
• Agent can breakdown before accepting and vice
  versa
• Agent can offer before breaking down and vice
  versa
• Agent can be on path to breakdown, then on path
  to acceptance because received offer changes Eu
  or resentment because extended offer changes Eu

94
(Assumes no progress)
low-valued ?
high-valued ?
95
(Assumes no progress)
low-valued ?
high-valued ?
96
(Assumes no progress)
low-valued ?
high-valued ?
97
(Assumes no progress)
rare alternation
98
What happens when two PP agents interact?
Dominatedby BATNA
2's aspiration
2's offer inthis round
Eu2
BATNA ltu1(bd),u2(bd)gt
1's offers inthis round
1's aspiration
Eu1
99
What happens when two PP agents interact?
Eu2
Eu1(t2)
Eu1(t1)
100
What happens when two PP agents interact?
2'ssecurityresentment
1'ssecurityresentment
1's offersin this round
101
What happens when two PP agents interact?
102
What happens when two PP agents interact?
103
What happens when two PP agents interact?
1 breaks down
Laissez-faire path is
Amount of(specific)resentment
ltEu1,Eu2gt through time
104
Both generousat the start
1 isgenerousat start, 2 is not
2 isgenerousat start, 1 is not
Does the starting offer affect the laissez-faire
path?
105
Breakdownat t2(purepessimism)
106
Differentlaissez-faire paths
107
Breakdownat t5withresentment
108
All paths lead to breakdown
109
Fixedagentcharacteristics
Variedaccelerationof offers
In a different negotiation,some paths lead to
acceptance, some to breakdown
110
A third example where player 1 can guaranteean
acceptance outcome with the right initial offers
111
Laissez-Faire Paths

• What are they?They are the path taken if Eu
  drives all speech acts (offers,
  acceptance, and breakdown)
• BUT rationality is constructive constraints are
  one-sided, specifying latest times, not earliest
  times
• One may deviate from the path by accelerating
  offers take initiative show free will,
  self-determination

112
Laissez-Faire Paths

• What are they?They are the path taken if Eu
  drives all speech acts (offers,
  acceptance, and breakdown)
• So how about a meta-game over the "when to
  accelerate offers" strategy space? We could
  find Nash Equilibria!

113
Laissez-Faire Paths

• What are they?They are the path taken if Eu
  drives all speech acts (offers,
  acceptance, and breakdown)
• So how about a meta-game over the "when to
  accelerate offers" strategy space? We could
  find Nash Equilibria!
• OK, but it is a control problem (like flying an
  airplane) estimation and correction Are we on
  a path to agreement? Does other party need new
  reason to continue?

114
All sorts of meta-games wheneveryou combine
laissez-faire behaviorand elective
initiative-taking
115
All sorts of meta-games wheneveryou combine
laissez-faire behaviorand elective
initiative-taking
Payoffs adjusted for substantivefairness, time,
concessions, aspiration,and security
116
Desiderata for an AI model of negotiation

• Should be able to participate in the process of
  negotiation make offers, respond to offers,
  deliberate upon offers, (even threaten, search,
  argue) NOT just insist on a solution
• Should model the phenomenon directly pessimism,
  expectation, security, resentment NOT just
  implement a mathematical analysis

117
Desiderata for an AI model of negotiation

• Should be able to participate in the process of
  negotiation make offers, respond to offers,
  deliberate upon offers, (even threaten, search,
  argue) NOT just insist on a solution
• Should model the phenomenon directly pessimism,
  expectation, security, resentment NOT just
  implement a mathematical analysis
• Should be robust in poor information
  settings there may be no common knowledge, no
  model of other's mental states
• (our model is decoupled!)
• Should learn about negotiating partner and social
  norms offline, objective probability is based
  on data online estimation is anticipated
• (agent can be calibrated)

118
Desiderata for an AI model of negotiation

• Should be able to participate in the process of
  negotiation make offers, respond to offers,
  deliberate upon offers, (even threaten, search,
  argue) NOT just insist on a solution
• Should model the phenomenon directly pessimism,
  expectation, security, resentment NOT just
  implement a mathematical analysis
• Should be robust in poor information
  settings there may be no common knowledge, no
  model of other's mental states
• (our model is decoupled!)
• Should learn about negotiating partner and social
  norms offline, objective probability is based
  on data online estimation is anticipated
• (agent can be calibrated)
• Should be an architectural paradigm for design of
  agents pessimism and resentment are internally
  balancing forces this is a design proposal, not
  an empirical claim

119
Desiderata for an AI model of negotiation

• Should be able to participate in the process of
  negotiation make offers, respond to offers,
  deliberate upon offers, (even threaten, search,
  argue) NOT just insist on a solution
• Should model the phenomenon directly pessimism,
  expectation, security, resentment NOT just
  implement a mathematical analysis
• Should be robust in poor information
  settings there may be no common knowledge, no
  model of other's mental states
• (our model is decoupled!)
• Should learn about negotiating partner and social
  norms offline, objective probability is based
  on data online estimation is anticipated
• (agent can be calibrated)
• Should be an architectural paradigm for design of
  agents pessimism and resentment are internally
  balancing forces this is a design proposal, not
  an empirical claim much work to be done!

120
A New Model of Negotiation Based on Pessimism
and PunishmentR. P. LouiWashington
UniversitySt. Louis

• CAEPIA 2003

thanks to M. Foltz, V. Reddy, D. Weisberger, I.
Figelman,D. Moore, K. Hashimoto, A. Jump,F.
Tohme, K. Larson, S. Braynov,J. Nachbar, B.
Dheeravongkit, L. Cai, K. Chin, R. Bujans, M.
Looks, E. Wofsey, J. Rosen, S. Grubor, K.
Ormsby, J. Badino, R. Pless, and M.A. Clark.
121
A New Model of Negotiation Based on Pessimism
and PunishmentR. P. LouiWashington
UniversitySt. Louis

• CAEPIA 2003

thanks to M. Foltz, V. Reddy, D. Weisberger, I.
Figelman,D. Moore, K. Hashimoto, A. Jump,F.
Tohme, K. Larson, S. BraynovJ. Nachbar, B.
Dheeravongkit, L. Cai, K. Chin, M. Looks,R.
Wofsey, J. Rosen, S. Grubor, K. Ormsby, J.
Badino, R. Pless, and M.A. Clark.

• Research Funding from
• NSF Information Technology and Organizations
  Program Multi-Agent Negotiation, as co-PI with
  T. Sandholm, July 1997 to July 1999. 9610122.
• NSF Office of Cross-Disciplinary Affairs and
  Interactive Systems Program Summer Undergraduate
  Research Assistants, March 1995 to February 1996.
  9415573.
• NSF Office of Cross-Disciplinary Affairs REU
  Continuing Award, April 1992 to September 1994.
  9123643.