EECS 690

1
EECS 690

• April 7

2
Issues in the collaboration of engineering and
morality

• At some level, engineers and philosophers are
  schooled to follow intuitively incompatible
  methods.
• While engineers generally believe that there is
  more than one solution to every problem, they are
  trained to converge on a satisfactory solution
  for the problem at hand. Ethicists, however, are
  trained to diverge from each other, arguing
  separate positions so as to describe as
  completely as possible the range of
  considerations and theories that may be relevant
  to a problem (p.75)

3
Judge versus agent perspectives

• The Judge Perspective This is a stance taken in
  ethical debates when ethical principles are
  applied to specific cases or scenarios, forcing a
  choice between two mutually exclusive
  alternatives, and defense of the choice. This is
  useful in clarifying philosophical argument, but
  not useful in AMA design
• The Agent Perspective This is a way of
  representing ethical reasoning as a set of
  constraints that may or may not be able to be
  simultaneously satisfied. This involves much of
  what has previously been termed Value Pluralism

4
Constraints

• One idea might be to imagine ethical concerns as
  sets of constraints. This leads to questions of
  what these constraints should be, and how
  specific/abstract.

5
Computability

• Perhaps it is true that ethics really consists in
  being able to generate constraints from other
  principles (i.e. deontological or utilitarian
  principles). This leads to the question of how
  computable some of these received moral theories
  are.

6
Whose morality?

• While trained ethicists will tell you that their
  field is not in specifying a bounded list of what
  is and is not ethical (rather that theirs is the
  field of discovering what constitutes ethical
  reasoning and justification) an engineered
  ethical system should come with some variety of
  success and failure conditions, and these amount
  to a partial list of what is and is not ethical.
  Questions about who should supply these
  conditions and how are very important questions.

7
Two general approaches for moving forward

• Top-down
• Takes an Ethical Theory and analyzes the
  informational and procedural requirements
  necessary to implement this theory in a computer
  system.
• Examples The Golden Rule, the Ten Commandments,
  Utilitarianism, Hammurabis code, The Yana and
  Niyama, lists of Aristotelian virtues, Kants
  Categorical Imperative, Asimovs Three (Four)
  Laws for Robots

• Bottom-up
• Do not explicitly rely on prior theory, but only
  on performance criteria, with the most basic
  method a way of creating an environment in which
  a machine explores courses of action, and is
  rewarded for morally praiseworthy behavior (and
  presumably punished for immoral behavior)
• Models Childhood development, evolution