Unit 4 Algorithmic bias

Learning outcomes

By the end of this unit you should:

  • be able to state and support their view on algorithmic bias
  • have aware of the key issues leading to such bias
  • have considered ways to mitigate algorithmic bias

Activity 1: Starter quiz

Work alone. Analyze the dilemma given below. Decide your stance on the issues. Identify supporting reasons for your stance. Evaluate the strength of the evidence. Use terminology introduced in this course related to ethics.

You are working on a machine learning model for criminal risk assessment. You realize the data is biased against certain racial groups. Correcting for this bias would compromise the model's accuracy. What's the ethical course of action?

Assessment criteria
  1. Organisation
  2. Evidence-based
  3. Precision of terminology

Activity 2: Introduction to algorithmic bias

Watch or listen to this slideshow.

Discuss the content of this slide deck.

Activity 3: Algorithmic bias - terminology

Read these definitions of IP terminology. Researchers and organisations may provide different definitions and the specific terms may vary by legal juristriction.

  1. Algorithmic Bias: Systematic errors in computational systems that produce unfair or prejudicial outcomes, usually reinforcing existing social inequities.
  2. Data Bias: Skewed or unrepresentative data used to train algorithms, leading to biased predictions or classifications.
  3. Stereotyping: The act of algorithms reinforcing existing stereotypes by associating certain roles or attributes to specific groups based on training data.
  4. Disproportionate Impact: When algorithms unintentionally harm certain groups more than others, often due to biased data or biased decision-making processes.
  5. Feedback Loops: A cycle where a biased outcome reinforces itself, causing the system to become increasingly biased over time.
  6. Transparency: The degree to which an algorithm's workings can be explained and understood, crucial for identifying and rectifying bias.
  7. Accountability: The obligation to explain and justify algorithmic decisions, including any biases that affect outcomes.
  8. Historical Bias: Pre-existing social and institutional inequities that get replicated in algorithmic systems, often through biased training data.
  9. Exclusion: The act of algorithms not considering or inaccurately processing certain groups, leading to their marginalization.
  10. Economic Bias: Biases in algorithms that favor those with greater economic resources, such as recommending job candidates based on prestigious educational backgrounds.
  11. Ethical/Moral Dilemmas: Situations where algorithms have to make decisions that involve ethical or moral considerations, such as in autonomous vehicles.
  12. Accessibility Bias: Algorithmic bias that arises when systems are not optimized for people with disabilities, affecting engagement and usability.
  13. Fairness: The quality of making impartial, just, and equitable decisions, crucial for algorithms in contexts like hiring, lending, or criminal justice.

Activity 4: Discussion

Read the proposal below and identify the ethical issues that need to be discussed.

Leveraging Data Analytics to Enhance the Learning Experience at Our University

Dear Student Council Members,

We propose the implementation of a data analytics program to enhance student engagement and improve academic performance at our University. As part of this initiative, we plan to introduce a cutting-edge Learning Management System (LMS) that will track metrics such as time spent on the platform, the number of assignments submitted, participation in forums, and frequency of log-ins. While the potential cost-saving benefits of switching to a cloud-based LMS have stirred debate among faculty, we believe the focus should remain on how best to utilize data for the benefit of our educational community.

The system would employ several actionable measures based on the collected data. For instance, students logging less than two hours per week could receive automated "academic risk" alerts designed to encourage increased platform interaction. Additionally, we suggest offering financial incentives to instructors whose courses record above-average engagement rates, as measured by the time students spend on the LMS. Furthermore, courses that consistently report lower engagement could be flagged for curriculum review and potential redevelopment.

However, it's important to address the various concerns that may arise from this proposal. A recent opinion piece in the university newspaper criticized the usability of most LMS platforms, and rumors have been circulating about the possibility of selling collected data to third-party educational firms. While these issues are noteworthy, they should not distract from the core purpose of this proposal, which is to utilize data to improve the educational experience at XYZ University.

To prompt a robust discussion, consider the following questions:

  • Is it ethical to categorize students as "at-risk" based solely on LMS engagement metrics?
  • Will rewarding instructors based on LMS engagement lead to a more accurate evaluation of teaching quality?
  • How reliable are metrics like time spent on the LMS as indicators of student engagement or academic performance?
  • What privacy concerns might arise from collecting such data, and how should they be securely addressed?
  • Are there specific cultural factors, given our university's location in Japan, that could influence the effectiveness of an LMS?
  • Could this system inadvertently disadvantage certain groups, such as working students who may have limited time to engage with the platform?

We invite the Student Council to engage in a comprehensive discussion of these questions to evaluate the ethical, pedagogical, and practical aspects of implementing data analytics within our educational environment. Your insights will be invaluable for shaping a more effective and equitable learning experience.

The Data Analytics for Education Committee

Work in pairs or small groups. Discuss the ethical issues you have identified.

Activity 5: Formal fallacies: Invalid arguments

Work alone or in pairs. Answer the following questions about the sentence.

If I speak, my mouth moves.
where P = If I speak, and Q = my mouth moves

P, Q

  1. Which is the antecedent P or Q?
  2. Which is the consequent P or Q?
  3. Does deny mean stating something is true or false?
  4. Does affirm mean stating something is true or false?
mouth moving

Activity 6: Thinking

Work alone or in pairs. Answer the following questions.

P     Q   not P   not Q

  1. Which of the above is denying the antecedent?
  2. Which of the above is denying the consequent?
  3. Which of the above is affirming the antecedent?
  4. Which of the above is affirming the consequent?

Activity 7: Reading

Work alone or in pairs. Answer the following questions.

There are three common invalid arguments, namely affirming the consequent, denying the antecedent and the undistributed middle term.

  1. affirming the consequent If it rains, the street will be wet. The street is wet. Therefore, it rained.
  2. denying the antecedent If he is in hospital, then he is sick. He is not in hospital. Therefore, he is not sick.
  3. undistributed middle term Some Chinese attended the class. Aiko attended the class. Therefore, Aiko is Chinese.
  1. affirming the consequent
    If P, then Q
  2. denying the antecedent
    If P, then Q
    Not P
    ∴Not Q
  3. undistributed middle term
    P -> Q
    R -> Q
    P -> R

Activity 8: Seminar discussion 2

Your tutor will explain the procedure.

Set topic: Is intellectual property protection beneficial or detrimental to society?

The presentation topics are listed below.

  1. Economic impact of IP rights
  2. Implications of patenting life forms, e.g. genes
  3. Open source vs. Proprietary software
  4. Cultural appropriation and IP

The assessment criteria are:

  1. Argumentation
  2. Content
  3. Interaction
  4. Terminology


Can you:

  1. describe what algorithmic bias is
  2. define the core terms related to algorithmic bias listed in Activity 3

If you can not, make sure that you do before your next class.

Running count: 48 of 61 concepts covered so far.