Ethics

Ethics are the moral standards of ‘right’ and ‘wrong’ that prescribe how we ought to act. Unlike statements of facts that are ‘true’ or’false’.

Usually in ethics we ask ourselves questions such as “What should I do? How should I act?”, or, e.g., more boradly “How should I live? What sort of person should I strive to be?”.

We can distinguish 2 types of ethics:

• Descriptive ethics: explanations of, and factual statements about the moral systems people subscribe to, how we act and understand ourselves as moral beings (biology/psychology - anthropology).

  For example, “Home Spaiens is a warring species”. You can agree or disagree, but is a fact of our behaviour.

• Normative ethics: prescriptive accounts on how we ought to act and understand our selves as moral beings. (philosophy).

  For example, “People should not make war: War is bad!”.

Note that ML systems can have both ‘descriptive’ and ‘normative’ tasks or goals. For example, given a clavicle bone x-ray:

• Descriptive task: ’Guess someone’s age based on the size of their clavicle bone’.

• Normative task: ’Based on their clavicle bone, judge if someone is younger than 18 or not, and should be granted asylum in our country’.

  

Two common misconceptions of ethics:

1. Subjectivism: Moral statement are reflections of merely subjective perceptions and personal preferences.

   When there are disagreements we expect others, and others expect us, to give reasons for their ethical choices and beliefs and to argue for why they take those reasons to be correct (to justify the correctness of our belief).

2. Popular opinion: Moral Statements are reflections of public opinion.

   We know this doesn't work. e.g., Hitler was democratically elected: popular opinion believed he was a good leader.

   Some ML developers ask their users if to what extent they think certain (data) features are acceptable to train MLs with

   • Feature ‘A priori’ fairness: degree to which general population deems the use of a certain features ‘fair’ without any knowledge about its effect on the outcome
   • Feature-accuracy fairness: degree to which general population deems the use of a certain feature ‘fair’ if it increased the accuracy of the outcomes
   • Feature disparity fairness: degree to which general population deems the use of acertain feature ‘fair’ despite its use increasing any form of disparity

3 key ideas in normative ethics:

• Values > lasting convictions that people believe are desirable to pursue, promote, preserve, or protect. Values are often abstract goals or ideals (e.g., freedom, privacy, well-being, sustainability, fairness, efficiency)
• Norms > rules that prescribe which actions are morally required, permitted, or forbidden. Norms that translate values into concrete behaviour (e.g. do not kill, respect others’ opinions, do not waste, don’t use bad data)
• Virtues > character that make someone a good person or that allow people to lead could live (e.g. courage, honesty, generosity, modesty, accuracy, elegance)

A Normative Ethical Theory is a formulation of an account or set of principles that prescribe us how to act. Aims to provide systematic answers to the questions ‘How should I live?’ and ‘How should I act?’

We can do this by presenting 3 different points of emphasis: