6CCS3AIN
Week 10: AI and Ethics: Tutorial Questions
Q1. In the lectures, we mention that for model-driven (or symbolic) AI systems it is usually straightforward to generate explanations of the reasoning used by the AI system in each particular case where the system is applied. An expert system which uses IF-THEN rules is a common example of a model-driven AI system. Why is it usually straightforward to generate an explanation? How are the explanations generated?
Q2. The technology company you work for has tasked you with developing an AI application which searches social media, such as Facebook and Linked-In, for information and photos of potential recruits and then matches this information against profiles of the company’s best- performing existing staff to identify the best potential recruits. The plan is that the Human Resources Department will contact the potential recruits to invite them for an interview.
After developing the system but before putting it into production, you notice that almost all the recommended potential recruits are men. You realize that this may be because most of the staff in the technology sector, including most of company’s existing staff and most potential recruits, are men.
You also notice that the system seems to reject any potential recruits whose photos show them wearing a hat or other headgear. You do not know why the AI system does this, but it may just be some trivial quirk of a machine learning system.
You have several possible action options:
(a) Do nothing. All AI systems have quirks, and it is best to leave them alone.
(b) This decision is the responsibility of my employer, so I will just follow the orders of my boss.
(c) I will try to eliminate the bias against both women and hat-wearers.
(d) I will try to eliminate the bias against women but ignore the issue of the hat-wearers, as this is trivial.
(e) I will try to eliminate the bias against hat-wearers but ignore the issue of gender bias, because this is a problem across the entire technology sector which one company cannot solve.
What are the reasons for or against each option? Which options would you definitely not choose? Which option would you choose?
1
Q3. In the KEATS Quiz for Week 9, there is a question about a flag-colouring problem with 5 agents (the question is called ¡°Pentagon¡±). This is the question:
We have a distributed system with 5 agents. Each agent is connected to two others, arranged in a circle (ie, forming the corners of a pentagon).
Each agent has two possible colours (or states), Red or Blue. The initial state is that one agent is Blue and the other 4 agents are Red.
Each agent has been programmed with the following algorithm:
START ALGO
To decide the colour it will take at the next round, each agent looks to its two neighbours and executes the following two rules in order.
Rule 1: If the two neighbours are both currently the same colour as each other, then the agent chooses the opposite colour to its two
neighbours. It does this without regard to its own current colour.
Rule 2: If the two neighbours are different colours from each other, then the agent remains whatever colour it currently is.
END ALGO
Which of the following statements are true?
There is one correct answer in the quiz to this question, which is a statement about the behaviour of the model. Provide an explanation for the reasoning used to reach this correct statement.
2