Bias in AI
Objective
This chapter explains the issue of bias in AI, how it can occur, and how we can avoid it.
1. What is Bias?
Simple Definition
Bias means unfair treatment of a certain group of people.
Example:
- A system that favors men over women
- A system that favors one people over another
- An unfair system
2. How Does Bias Occur?
From Data
Problem:
- If the data is biased
- AI learns the bias
- Repeats the bias
Example:
- Employment data that favors men
- AI learns this
- Favors men in hiring
3. Examples of Bias
In Employment
Problem:
- Systems favor certain groups
- Neglect other groups
- Unfair
In Loans
Problem:
- Systems reject loans for certain groups
- Without real reason
- Unfair discrimination
In Security
Problem:
- Systems recognize some groups more
- Mistake with other groups
- Inaccurate for everyone
4. Why is Bias a Problem?
Negative Impact
1. Injustice:
- Unfair treatment
- Unequal opportunities
- Discrimination
2. Loss of Trust:
- People do not trust systems
- Refusal to use
- Social issues
3. Bad Outcomes:
- Wrong decisions
- Harm to people
- Legal issues
5. How to Avoid Bias?
Use Diverse Data
Solution:
- Data from different groups
- Fair data
- Inclusive data
Example:
- Images of all races
- Data from all groups
- Fair representation
Test Systems
Check:
- Test on different groups
- Ensure fairness
- Correct issues
6. Responsibility
Who is Responsible?
1. Developers:
- Use fair data
- Test systems
- Correct issues
2. Companies:
- Monitor systems
- Ensure fairness
- Responsibility
3. Governments:
- Clear laws
- Monitor usage
- Protect people
Conclusion
- Bias is a real issue
- Occurs from biased data
- Must be avoided
- Shared responsibility