Lecture 9: Negotiation Formalization

목차

 

Social Welfare (Fairness)

Simple notion of social welfare: 

 

→ That maximizes the sum of all

 

Egalitarian point (aka Rawls Point):

→ maximize the minimum utility from the whole population.

 

Utilitarian Point (aka Kalai-Smorodinsky Point):

→ closest point to the r_eq (대각선)

 

 

When no equality utility solution exists

Only three possible outcomes, and no one likes outcome B. (A and C are optimal frontier, but only B is fair --> either only one person wins).

 

 

Domain

Validity and Added value: The added value of using AI techniques highly depends on the quality/validity of the model:

→ Weak model → weak support by the technique

→ Faulty model → faulty advice

Transparency: Model should be transparent for the user of technique

- Trust

- Understanding the advice

 

 

 

Domain Model

For the negotiations, we need a set of issues X, and for each issue x ∈ X a set of values for each of those: V(x)

For each stakeholder, we need a preference profile for this domain

도메인 모델: 특정 문제와 관련된 모든 주제의 개념 모델

 

 

Domain Modelling

Domain modelling is difficult for humans (경험, 지식 등이 없어서)

시험에 Construct and explain a domain model for the following domain.

 

 

Preferences

Given a finite set of issues X, and i and j ∈ X and sets of values for each of those issues: V(i), V(j)

Stakeholders have preferences over the values per issue and over combinations (bids) of values for a range of issues, expressed by a preference relation <p

Suppose a, b ∈ V(i), for some issue i, then a <p b expresses that the stakeholder prefers b over a.

 

Preference Independence: A set of attributes Y ⊂ X is preferentially independent of its complement X - Y when the preference order over outcomes with varying values of attributes in Y does not change when the attributes of X - Y are fixed to any value.

Independence allows for linear additive utility functions.