The Ethical Landscape of AI: Challenges and Considerations

AI is transforming industries and reshaping human experiences, from healthcare diagnostics to predictive analytics in law enforcement. As adoption grows, so do the ethical challenges that come with it. AI ethics is about using AI responsibly: addressing fairness, transparency, and accountability so we can balance innovation with its impact on society.

AI Ethics: Balancing Innovation and Responsibility

As AI becomes more woven into daily life, its ethical implications reach beyond individual choices to questions of access, representation, and governance. AI has the potential to empower, but it can also reinforce existing inequalities if it is not carefully developed and deployed. The video below from UNESCO explores these challenges and the importance of inclusive AI governance, ensuring diverse perspectives shape AI policies and frameworks.

Hidden AI Labor: How We Have Been Training AI for Years

As noted in Module 1, AI systems rely heavily on human-labeled data, often gathered from people without explicit consent or compensation. This raises ethical questions about who benefits most from AI, and about transparency in how human-generated data is used.

How CAPTCHA Helped Train AI

Early CAPTCHA challenges asked users to decipher distorted text to prove they were human. With reCAPTCHA, Google turned that routine security check into unpaid labor for AI training. It was first used to digitize books through optical character recognition (OCR), then expanded to image-based tasks.

Try the simulated version below: pick out the buses, then choose Verify.

A simulation for learning. A real reCAPTCHA would use photographs rather than icons, and screen-reader users can identify each square by its label.

When users identify objects like crosswalks, buses, or storefronts, they are unknowingly contributing to AI training, including:

• Mapping and self-driving cars: the image grids pull from Google Street View, so labeling traffic lights, crosswalks, and signs helps train the computer vision behind Google Maps and Waymo's self-driving systems.
• Computer vision: sharpening AI's capacity to interpret and classify images accurately.

Sources: von Ahn et al., "reCAPTCHA: Human-Based Character Recognition via Web Security Measures" (Science, 2008); IBM, "What Is CAPTCHA?"

The Bigger Picture: Who Is Doing the Work for AI?

Beyond reCAPTCHA, AI development still depends heavily on human labor, often under difficult or exploitative conditions:

• Gig workers: platforms like Amazon Mechanical Turk hire workers globally to label images, transcribe data, and moderate content, frequently under hard conditions (ABC News).
• Content moderators: AI safety often depends on contracted workers who manually review and flag harmful content, which can expose them to serious psychological distress (The Guardian).
• Everyday AI users: regular users also train AI by giving feedback, such as clicking "thumbs up" or correcting outputs, which is used to refine future responses.

These often invisible forms of human labor raise pressing questions about fairness, consent, and who shares in the benefits. As AI becomes more embedded in society, transparency and accountability matter: the people who power AI deserve recognition and fair compensation.

Beyond Black and White: Understanding AI Ethics

This section is adapted from the work of Leo S. Lo. Used with permission.

Ethical AI requires more than governance. Social, psychological, and technical factors all shape how AI is designed, perceived, and integrated into society, and whether it reinforces or helps address inequalities:

• Social aspects: professional culture, media representation, public policy and regulation, and concerns about job displacement all influence whether AI is accepted or feared.
• Psychological factors: perceived usefulness, ease of use, trust in AI decisions, fear of losing control, and resistance to change.
• Technical considerations: data quality, algorithm transparency, compatibility with existing systems, and performance measures like accuracy, recall, and precision.

Ethical Frameworks for Responsible AI

AI ethics is rarely black and white; there is seldom a single right answer. Good decisions mean balancing trade-offs between efficiency, fairness, transparency, and societal impact. Take streaming services: AI recommends content based on your past habits to improve your experience, but it can also create "filter bubbles" that limit exposure to diverse content and favor mainstream artists over independent creators. Should AI shape what we watch? Is engagement worth a loss of diversity? Ethical frameworks give us different lenses for questions like these.

Care Ethics and Moral Distance

Care ethics prioritizes relationships and vulnerabilities over rigid rules. For recommendations, that means making sure users are not trapped in filter bubbles and independent creators are not pushed aside. This ties to moral distance: the idea that developers and decision-makers can become detached from the people their systems affect. The greater the distance, the easier it is to overlook harm. A care-based approach pushes for AI that acknowledges and reduces these effects, so technology serves human needs rather than dictating them.

Why a Nuanced Approach Matters

Many AI discussions default to extremes: celebrating AI as a revolutionary tool for improving our world, or warning of it as an existential risk to humanity. In reality, AI's impact depends on how it is designed, deployed, and regulated.

For example, AI-powered medical diagnostics can detect disease earlier and improve outcomes for conditions like hip dysplasia in infants (Libon, 2023), osteoporosis (Paderno, 2024), and several cancers (Kanan et al., 2024; Kim et al., 2024). Yet if these systems are trained on non-representative data, they can misdiagnose underrepresented populations, leading to unequal care (Ranard et al., 2024).

The same pattern shows up in policing. Departments use facial recognition to generate suspects, but the technology misidentifies people of color at higher rates. More than a dozen people in the United States are known to have been wrongfully arrested after a facial-recognition match, and nearly all of them are Black (ACLU). Robert Williams, arrested in front of his family in Detroit in 2020, was the first publicly documented case (ACLU); the count has kept climbing, with several new cases in 2025 (Washington Post, 2025). These are not hypothetical harms: a biased system, trusted too readily, can cost someone their freedom.

AI decisions are rarely clear-cut: ethical thinking has to go beyond efficiency and innovation at every stage of design and use.

Who Governs AI? The Rules Are Catching Up

For most of the last decade, AI advanced faster than the rules around it. That is changing. In 2024 the European Union passed the EU AI Act, the first comprehensive AI law. It bans some uses outright, such as government social scoring; sets strict requirements for "high-risk" systems like those used in hiring, education, and policing; and phases in over several years, with major obligations taking effect in August 2026. Closer to home, (YOUR INSTITUTION) has adopted its own AI policies and syllabus expectations, which you will look at in Module 3. Governance does not settle the ethical questions in this module, but it shapes which answers become requirements.

Knowledge Check

Select an answer to see feedback. Each option explains why it is or is not correct.