by Richard Bergmair

Introduction

In this article, we will address the question “How can computer science methods help us to better understand ethics?”

Ethics, and especially normative ethics, are concerned with finding “criteria for what is morally right and wrong. Ethics includes the formulation of moral rules” [ 1].

Logic programming is concerned with identifying base facts that are logically true or false. It includes the formulation of inference rules.

 

The polymorphy described above sits at the core of this discussion. To show this more clearly, a type of logic will be established, but instead of reasoning, and determining whether something is true or false, it will make moral judgments, telling whether something is good or bad. Some core thoughts of positivistic and teleological ethics will be formalized in this logic to give a computational model of telling right from wrong. A fictional artificial intelligence is discussed, using insights provided by metamathematics, and the bargaining problem, a philosophical question with wide-ranging implications, will be approached with game theory. But before going into too much detail about how to approach these subjects from a computer scientist’s point of view, it might be helpful to briefly review these ethical topics from a classical point of view.

 

Three Questions, Six Answers

Kenneth Laudon [ 6] uses three “critical distinctions” for organizing the literature on ethics, which can be viewed as the three big questions of ethics. The different answers to these questions define a spectrum of ethical thought.

 

Question One: What is “goodness?”

Phenomenologist’s answer: It is a higher order, and it is given. One must understand the abstract concepts of right and wrong, and act accordingly.

Positivist’s answer: It is whatever we make of it. We have to derive ethical principles for ourselves according to our observations of the real world.

 

Question Two: Does acting ethically correct mean acting according to certain rules, or acting in a way leading to desired consequences?

Deontologist’s answer: Acting ethically correct, means respecting one’s duties and obligations. Each single act is itself good or bad, regardless of its consequences.

Teleologist’s answer: An action can be judged only by its consequences. Acting right means acting in such a way that the outcome is good. There is no act that is a priori good.

 

Question Three: What is the scope of morality? Is morality subject to everyone’s individual freedom?

Collectivist’s answer: Ethical standards make sense only if they equally apply to everyone. A rule like “thou shalt not kill,” that was subject to the individual’s approval, would be the equivalent of a rule like “thou shalt not kill, unless it benefits you.”

Individualist’s answer: Nobody should be committed to accepting ethical standards just out of pure principle or we will end up with a morality that people ultimately suffer from.

 

Given this basic, yet vague understanding of ethical principles, one can develop a more rigorous representation suitable for computation. Such an attempt quickly presents the problem of undecidability, an interesting principle with wide-ranging implications.

Explaining Gödel’s incompleteness theorems in detail is beyond the scope of this article. The interested reader is referred to Kurt Gödel’s original publication [ 2]. Alternately, Hofstadter’s Pulitzer-Prize-win-ning classic Gödel, Escher, Bach: An Eternal Golden Braid [ 5] and Smullyan [ 8] make the complex ideas behind this topic accessible.

 

An Infinitely Long Time Ago...

Consider a thought experiment: A computer program, called GURU, shall be designed, sent back through time, and started at a point in time that is infinitely long ago. That program should be capable of telling right from wrong. Such a program would need a set of ground facts, axioms, premises, or whatever term fits one’s philosophy of systematic reasoning. These could be statements telling the machine that something is definitely good or bad. This is not enough, because not every situation the machine will ever be confronted with can be expected to appear in the set of ground facts. Therefore some inference rules or proof techniques would have to be established. The machine would need them to determine whether “compound-situations,” situations described in terms of other situations, are good or bad.

GURU would also need an inference engine that systematically combines simple situations, achieving “compound-situations” based on the inference-rules. Whenever GURU discovers such a situation, it could store it to an infinite memory, called the scroll of all wisdom. If all of the ground facts can safely be assumed, all of the inference rules work, and the inference engine produces correct outputs, then, applying the