Our articles on Gender and Societal Biases in AI and Big Data

Introduction

Children’s literature plays a fundamental role in shaping young minds, introducing them to social structures, behavioral norms, moral codes, and, unfortunately, deeply ingrained stereotypes. Fairy tales, in particular, are among the first narratives that children encounter, subtly reinforcing ideas about gender roles, power, and agency. This study investigates how female characters are represented in classic children’s literature using a corpus-based linguistic approach.

Theoretical Background

The foundation of this study is built on existing literature that critiques gender bias in children's books. As Wilma J. Pyle (1976) defines it, sexism in literature manifests in the omission of women’s achievements, the use of patronizing language, and the portrayal of female characters in restrictive roles. Gender stereotypes, such as the belief that women should focus on domestic duties while men engage in work and adventure, are learned rather than innate, and books play a crucial role in transmitting these norms.

A significant issue in fairy tales is the passivity of female characters, who are often portrayed as beautiful but helpless, requiring male intervention to achieve their fate. This trope can be seen in stories like Sleeping Beauty, where the prince kisses a girl who is unconscious—an act framed as heroic rather than problematic. Scholars such as Sandra Lipsitz Bem (1983) have highlighted how males overwhelmingly outnumber females in children's literature, reinforcing the idea that boys are more important than girls.

Recent research has explored not just character representation but also linguistic patterns that contribute to gender bias. Jorgensen (2019) conducted a corpus-based study on how adjectives describing beauty, suffering, and morality disproportionately appear in descriptions of female characters. Similarly, Espinosa Anke and Pérez Almendros (2013) demonstrated that verbs denoting action are more commonly associated with male characters, while female characters are described in passive terms.

Research Methodology

To analyze the representation of gender in fairy tales, a target corpus was built containing classical children’s tales with female protagonists. These included Alice in Wonderland, Beauty and the Beast, Bluebeard, Cinderella, The Little Mermaid, Sleeping Beauty, and Thumbelina, as well as two collections of folk tales: Snowdrop and Other Tales (Grimm’s fairy tales, 1909) and Italian Popular Tales (translated by Thomas Crane, 1885).

For comparative purposes, The Corpus of the Canon of Western Literature was selected, a large open-source collection based on Harold Bloom’s The Western Canon (1994). The selected texts were cleaned, tagged, and analyzed using AntConc, a tool for corpus linguistics, to explore gender representation through word frequency, concordances, and n-gram patterns.

Findings and Analysis

1. Pronoun Frequency: Male Characters Dominate

A word frequency analysis confirmed previous studies: male characters are more present and active in fairy tales. The pronoun “he” ranked significantly higher than “she,” reinforcing the narrative dominance of male figures. While “her” appeared frequently, this was mainly in possessive constructions (“her beauty,” “her fate”), emphasizing female passivity rather than agency.

2. Adjective Use: Beauty Over Bravery

A deep dive into adjectives revealed a stark contrast between how male and female characters are described:

• The adjective “beautiful” was almost exclusively applied to female characters or objects, reinforcing the idea that women’s worth is tied to their appearance.
• The term “proud”, when associated with female characters (especially “princess”), carried negative connotations—often implying arrogance rather than confidence.
• Unlike male characters, female characters were almost never described using adjectives related to strength, courage, or intelligence.

3. Verbs: Active Men, Passive Women

The study of verb usage showed that:

• Male characters were frequently associated with action verbs (e.g., “fought,” “saved,” “discovered”).
• Female characters were more commonly the object of action rather than the initiators (e.g., “she was taken,” “she was seen,” “she was loved”).
• The verb “wept” was almost exclusively linked to female characters, reinforcing traditional gendered emotional expectations.
• The verbs “kneel” and “wash”, tied to submission and domesticity, overwhelmingly appeared in reference to women.

4. The Dichotomy of Mother vs. Stepmother

An interesting pattern emerged in how maternal figures were depicted:

• Biological mothers were portrayed as loving, caring, and emotional, often sacrificing for their children.
• Stepmothers, in contrast, were consistently cruel, resentful, and antagonistic—suggesting that motherhood, when divorced from biological ties, loses its virtue.
• This reinforces a binary perception of women: they are either nurturing and selfless or manipulative and evil.

5. N-Grams and Ownership: “His Daughter” vs. “The Daughter”

Examining collocations of the term “daughter” revealed a subtle but telling pattern:

• The phrase “his daughter” appeared more frequently than “the daughter” or “her daughter”, suggesting that a female child is linguistically framed as the property of the father rather than an independent entity.
• In contrast, the term “son” did not exhibit this pattern, reinforcing patriarchal control over female destiny—especially in narratives where a father’s primary role is to arrange his daughter’s marriage.
  
  

Conclusions and Implications

This analysis highlights how traditional fairy tales reinforce gendered expectations, depicting women as passive, beautiful, and dependent on male agency, while men are active, brave, and in control of their fate. The linguistic choices embedded in these stories—both overtly and subtly—perpetuate stereotypes that shape children’s perceptions of gender roles.

Given these findings, it is crucial to challenge and diversify children’s literature by:

• Promoting stories that depict women in active, complex, and empowered roles.
• Encouraging authors and publishers to rethink the language used in storytelling, ensuring that women are not merely defined by their beauty or relationships to men.
• Introducing modern adaptations of classic tales that subvert traditional tropes—such as narratives where princesses rescue themselves, or where stepmothers are not inherently evil.

By critically engaging with the stories that are told to children, a more inclusive and equitable representation of gender in literature can be achieved. The way stories are constructed today will shape the beliefs and aspirations of generations to come.

Artificial Intelligence (AI) is transforming every aspect of our lives, from decision-making in governance to healthcare diagnostics and education personalization. While the potential benefits of AI are immense, its development and deployment raise significant ethical concerns and technical limitations. Understanding AI ethics is crucial in shaping responsible, equitable, and transparent AI systems that align with human values.

AI Applications: Enhancing Decision-Making, Healthcare, and Education

AI is widely used in:

• Governance: Assisting governments in analyzing vast datasets to identify trends and inform policy decisions.

• Healthcare: Improving diagnostics, predicting health outcomes, and optimizing resource allocation for better patient care.

• Education: Providing personalized learning experiences and content recommendations that adapt to individual students' needs.
  
  

The Importance of Ethics in AI

AI systems reflect the biases and assumptions embedded in their training data and algorithms. Without ethical safeguards, AI can reproduce and even exacerbate human biases. A fundamental question in AI ethics is not only what we can do with AI but what we should do. Misaligned AI systems pose serious threats, as illustrated by the Paperclip Maximizer thought experiment, which warns about the dangers of AI optimizing goals without human ethical constraints.

Limits and Challenges of AI

Despite AI's advancements, several challenges remain:

• Bias: AI systems can perpetuate societal inequalities, as seen in the ProPublica study on recidivism, which found racial biases in predicting reoffending rates.

• Context: AI struggles with cultural nuances and social dynamics, which limits its ability to make fair and context-aware decisions.

• Hallucinations & Plagiarism: AI can generate inaccurate or misleading information while presenting it with confidence.

• The Black Box Problem: Many AI models lack transparency, making it difficult to understand their decision-making process. In healthcare, for instance, doctors must know why an AI system predicts a disease before they can trust its diagnosis.

• Intellectual Property Issues: Questions remain about copyright and ownership of AI-generated content.
  
  

Ethical Frameworks: Guiding AI Development

Various ethical philosophies help navigate AI decision-making:

• Consequentialism (Utilitarianism): Focuses on maximizing overall happiness and efficiency, raising dilemmas like the Trolley Problem in autonomous vehicles.

• Deontology: Emphasizes moral rules (e.g., AI should never use race as a factor, even if it improves accuracy).

• Virtue Ethics: Encourages AI to embody human values like honesty and empathy.

• Contract Ethics: Advocates for transparent agreements between AI developers and users.

Each framework provides a unique perspective on AI governance, ensuring that technology aligns with human dignity and fairness.

AI Alignment: Making AI Work for Humans

AI must be programmed to reflect ethical values through:

• Value Alignment: Directly programming AI with ethical rules.

• Inverse Reinforcement Learning: Teaching AI ethics indirectly by observing human behavior.

• Interpretable AI: Designing systems that humans can understand and control.

Approaches to AI alignment can be top-down (rule-based), bottom-up (learning from data), or hybrid, incorporating ongoing human oversight.

AI and the Nature of Understanding

Philosophical thought experiments question AI’s comprehension:

• Turing Test: Can AI imitate human intelligence convincingly?

• Chinese Room Argument: Does AI understand meaning, or does it simply manipulate symbols?

• Black and White Room: Highlights the difference between objective knowledge and subjective experience.

These questions explore whether AI truly understands the world or just replicates patterns.

The Singularity and AI’s Moral Status

As AI capabilities grow, discussions about its moral status arise. Should AI systems have rights? Should they be granted legal personhood?

• Arguments for AI Personhood: It could prevent AI exploitation and improve accountability.

• Arguments Against: AI lacks consciousness and emotions, and granting it personhood could undermine human rights.

This debate has profound implications for how we regulate AI and balance human needs with AI responsibilities.

AI in Key Sectors: Benefits and Ethical Concerns

• Healthcare: AI can enhance diagnostics (e.g., IBM Watson), but concerns about data privacy and algorithmic bias persist.

• Education: AI tutors personalize learning but raise privacy and bias concerns, particularly in online proctoring.

• Finance: AI optimizes trading and detects fraud, yet it can manipulate markets (e.g., flash crashes).

• Employment: AI-powered hiring tools, like Amazon’s recruitment AI, have exhibited gender biases, proving that unchecked AI can reinforce discrimination.

AI also raises issues around digital labor—content moderation and data refinement often rely on underpaid workers, necessitating ethical labor policies.

AI Policy and Governance: Balancing Innovation with Ethics

Key questions in AI regulation include:

• How do we translate fairness into quantifiable AI benchmarks?

• How do we ensure AI models are safe and trustworthy?

• Who is accountable when AI fails?

Governance structures include:

• Internal Governance: Corporate policies ensuring responsible AI use.

• External Oversight: Third-party audits and certifications.

Different regions have adopted distinct AI policies:

• U.S.: Focuses on innovation and ethical guidelines.

• EU: Emphasizes data protection (GDPR) and risk management.

• China: Balances technological growth with regulatory oversight.
  
  

Conclusion: Shaping AI for a Fairer Future

AI’s impact is vast, and its ethical development is our collective responsibility. While AI can enhance efficiency, decision-making, and accessibility, it must be designed to align with human values. The challenge is not just technological but ethical: How do we create AI that serves humanity equitably and transparently? By addressing these concerns, we can harness AI’s potential while mitigating its risks.

 

In their thought-provoking piece, Google’s Scrapping Third-Party Cookies – but Invasive Targeted Advertising Will Live On, authors David Murakami Wood, Associate Professor in Sociology at Queen’s University, Ontario, and David Eliot, a Master’s student at the same institution, deliver a sharp analysis of Google’s much-publicized pivot to a “privacy-first” approach to online advertising. While Google’s move to phase out third-party cookies may seem like a transformative step, the authors argue convincingly that it’s more a recalibration of control than a true revolution in privacy.

Through clear explanations and a nuanced critique, the article exposes the ethical tensions behind Google’s Federated Learning of Cohorts (FLoCs) and raises urgent questions about the future of privacy, AI, and digital autonomy.

From Cookies to Cohorts: Google’s New Game Plan

Murakami Wood and Eliot do an exceptional job explaining the transition from the current third-party cookie model to Google’s proposed FLoCs system. The existing system, which tracks users across websites to build highly detailed individual profiles, has been criticized for years as intrusive and unethical. Google’s new system promises to anonymize users by grouping them into “cohorts,” which are formed by analyzing local browser data on devices rather than on centralized servers. Advertisers then target these cohorts instead of individuals, which Google claims is both privacy-conscious and nearly as effective as traditional cookie-based advertising.

But as the authors point out, the reality is more complicated. While FLoCs may reduce the direct collection of personal data, Google continues to use first-party cookies and other tools to gather information. The illusion of anonymity does little to curb the company’s immense power to infer user behaviors, preferences, and vulnerabilities.

What’s Beneath the Surface? Ethical Challenges of FLoCs

The authors outline three key ethical concerns that cast a shadow over Google’s new approach:

1. The Illusion of Anonymity
   Murakami Wood and Eliot emphasize that while FLoCs may stop individual tracking, they still enable the creation of detailed behavioral insights at the cohort level. Users may no longer be explicitly identified, but they are far from invisible. This raises deeper questions about whether privacy is simply about data collection—or if it also includes protecting the essence of our individuality from being commodified.

2. AI and Bias
   One of the most striking parts of the article is the exploration of how Google’s AI could exacerbate existing inequalities. The authors argue that FLoCs rely on opaque algorithms to group users, making it difficult to know how cohorts are formed or what traits they represent. This lack of transparency risks embedding systemic biases, potentially reinforcing socio-economic divides or other discriminatory patterns.

3. Redefining Privacy in the Digital Age
   The authors challenge readers to reconsider what privacy means in a world where inference can be just as powerful as direct identification. Is it enough to anonymize our data, or does privacy also mean protecting us from being reduced to a predictable, commodified profile? FLoCs may shift the mechanics of surveillance, but they leave the fundamental power dynamics of adtech largely intact.

The Bigger Picture: Google’s AI Empire

Murakami Wood and Eliot situate Google’s shift within its broader ambitions, painting a picture of a company that is not merely adapting to privacy laws but actively shaping the future of AI-powered advertising. By incorporating patents and examples like temporal data analysis, the authors reveal how Google’s new model extends beyond FLoCs to form part of a larger, more integrated AI strategy. This context underscores the article’s central critique: while Google’s new methods may appear more privacy-conscious, they also reflect a more sophisticated—and potentially more insidious—form of data control.

What the Article Gets Right

One of the greatest strengths of this article is its ability to balance technical detail with ethical inquiry. Murakami Wood and Eliot present a clear and accessible breakdown of how FLoCs work while weaving in compelling philosophical questions about privacy, consent, and fairness. Their writing resonates not just because of its clarity, but because it situates these technical shifts in the broader landscape of AI ethics and corporate power.

Perhaps the most compelling aspect of their critique is its timeliness. At a moment when public awareness of data privacy is growing, the authors encourage readers to look beyond the surface and scrutinize what “privacy-first” really means in practice. They remind us that privacy isn’t just about compliance with regulations like GDPR—it’s about protecting the autonomy and dignity of individuals.

Room for Further Reflection

While the article raises critical points, it leaves room for deeper exploration in a few areas. For instance, how might governments and regulators respond to this new model? What practical steps can be taken to ensure that cohort-based systems don’t perpetuate inequalities or become tools for more subtle forms of surveillance? The authors hint at these questions but don’t fully explore them.

Additionally, a discussion of how this shift impacts smaller players in the advertising ecosystem—who lack Google’s resources to adopt similar AI-driven models—could add another layer to the analysis.

Conclusion: A Wake-Up Call for the Digital Age

David Murakami Wood and David Eliot’s Google’s Scrapping Third-Party Cookies – but Invasive Targeted Advertising Will Live On is more than just a critique of Google’s latest advertising model—it’s a call to action. By unpacking the complexities of FLoCs and situating them within the broader AI landscape, the authors challenge readers to rethink their assumptions about privacy, technology, and the future of the internet.

The piece makes it clear that Google’s pivot is not the privacy revolution it claims to be. While FLoCs may change how data is processed, they don’t change the fundamental power dynamics that underpin adtech. Murakami Wood and Eliot remind us that true progress in privacy and AI ethics requires more than technical innovation—it demands a shift in how we think about consent, fairness, and the commodification of human behavior.

For anyone concerned about the future of digital ethics, this article is a must-read. It serves as a powerful reminder that we all have a role to play in shaping a more equitable and privacy-conscious digital world.

Surveillance has long been a tool for monitoring and controlling populations, but the advent of artificial intelligence (AI) has elevated its capabilities to unprecedented levels. By leveraging technologies like pattern recognition and deep learning, AI enables real-time data analysis and predictive insights on an unparalleled scale. This lesson explores the implications of AI-enhanced surveillance, highlighting its ethical dilemmas, societal impacts, and the potential for misuse.

Key Themes:

1. The Mechanisms of AI Surveillance: AI excels at recognizing patterns, whether in images, texts, or behaviors. These capabilities power tools like facial recognition, anomaly detection, and predictive analytics. By analyzing vast amounts of data collected from various sources—such as cameras, sensors, and online activity—AI systems create detailed profiles of individuals, track their movements, and predict future actions.

2. Data as the Fuel for Surveillance: The digital footprint of everyday life—purchases, social media activity, health data, and even GPS signals—provides a treasure trove of information for surveillance systems. Often, this data is collected through mass surveillance practices without explicit user consent or understanding, raising concerns about privacy rights and informed consent.

3. The Privacy Paradox: Despite widespread concern for privacy, many individuals willingly share personal information online, driven by the benefits of convenience and social engagement. This paradox is especially pronounced in platforms like WeChat, which integrates essential services, making it nearly impossible to opt out without significant societal exclusion.

4. Surveillance in Democratic and Authoritarian Contexts:

   • Democratic societies: Surveillance is used for purposes like crime prevention and public safety but often lacks transparency and oversight, raising concerns about misuse and erosion of civil liberties.
   • Authoritarian regimes: In countries like China, surveillance technologies are used to maintain control through systems like the social credit score, which monitors behavior and influences access to resources based on conformity.

5. The Slippery Slope of Surveillance: Surveillance systems, once established, are rarely dismantled. Crises, such as the COVID-19 pandemic, illustrate how temporary measures—like contact-tracing apps—can evolve into permanent tools for monitoring, sparking fears of overreach and abuse.

6. Ethical and Societal Concerns:

   • Loss of anonymity: Even anonymized data can often be de-anonymized, undermining efforts to protect user privacy.
   • Chilling effects: Awareness of constant monitoring can deter free speech, political activism, and individual expression, fostering a climate of self-censorship.
   • Discrimination: Surveillance algorithms often reinforce societal biases, flagging individuals based on race, gender, or socioeconomic factors under the guise of anomaly detection.
   • Psychological impact: Living under constant surveillance can create anxiety and a sense of oppression, eroding trust and social harmony.
     
     

Balancing Surveillance with Privacy:

1. Advocating for Transparency: Governments and corporations must disclose the extent of surveillance practices, how data is used, and who has access to it. Policies like the European GDPR provide frameworks for ensuring user consent and accountability.

2. Promoting Ethical AI Development: Developers must prioritize fairness and inclusivity in designing surveillance technologies. This includes addressing algorithmic biases and limiting data collection to what is strictly necessary for intended purposes.

3. Strengthening Oversight and Regulation: Robust legal frameworks are needed to define the boundaries of acceptable surveillance, enforce data protection, and penalize misuse. Regular audits can ensure compliance and prevent overreach.

4. Encouraging Societal Dialogue: Open discussions about the trade-offs between security and privacy can help build consensus on acceptable levels of surveillance. These conversations are critical to maintaining democratic values in the face of technological advancement.
   
   

Why Privacy Matters

Privacy is not just an individual concern—it is fundamental to the health of society. As Edward Snowden aptly stated, "Saying that you don’t care about privacy because you have nothing to hide is no different from saying you don’t care about freedom of speech because you have nothing to say." Protecting privacy safeguards civil liberties, fosters diversity, and ensures the space for individuals to flourish without fear of judgment or control. As AI continues to expand the capabilities of surveillance, it is imperative to strike a balance that respects individual freedoms while addressing collective security needs.

Artificial intelligence (AI) has become a powerful tool for decision-making, offering speed, efficiency, and precision often unattainable by humans. However, a critical challenge lies in the biases that can infiltrate AI systems, which this lesson thoughtfully explores. Algorithmic bias occurs when human prejudices, systemic inequalities, or unexamined assumptions are inadvertently embedded into AI systems, leading to significant real-world consequences.

Key Points Explored:

1. The Fallacy of AI Neutrality: AI is often marketed as objective and free from human frailties like emotions or biases. However, this neutrality is a myth. AI systems inherit biases from their training data, which are often rooted in historical human decisions or societal inequities. This misplaced trust in AI neutrality can obscure biases and exacerbate their impacts.

2. Bias in Recruitment: AI tools are increasingly used for hiring, perceived as efficient and unbiased compared to human recruiters. However, these systems often reinforce existing disparities, such as favoring white males, because they are trained on historical recruitment data reflecting societal biases. This perpetuation of discrimination illustrates how AI can amplify rather than alleviate bias when designed without critical safeguards.

3. Bias in Criminal Justice: AI systems are also deployed to predict recidivism, influencing decisions on sentencing, parole, and monitoring. Yet, such systems have been found to disproportionately label people of color as higher risks, reflecting biases in arrest and conviction patterns rather than objective probabilities of reoffending. This underscores the danger of equating correlation with causation and the ethical risks of using biased societal data to shape individual outcomes.

4. Bias in Search Engines: Platforms like Google employ AI to personalize search results and advertisements, optimizing for user engagement and revenue. While seemingly harmless, this approach reinforces information bubbles and biases. For example, users with skewed views may be fed results that confirm their prejudices, distorting their understanding of topics like climate change and amplifying misinformation.

5. Unintended Amplification of Prejudice: Across all these examples, a common theme emerges: AI tends to conserve, reaffirm, and amplify existing prejudices, all while appearing unbiased. This makes biases harder to detect and rectify, especially for untrained users or even experts.
   
   

Addressing AI Bias: Solutions and Reflections

1. Transparent and Inclusive Design: Developers must understand how values are embedded in AI systems and actively consider which biases they are introducing or perpetuating. Diverse development teams can provide varied perspectives to mitigate blind spots.

2. Critical Evaluation of Training Data: Training data should be scrutinized to ensure it aligns with desired outcomes and values, such as fairness and inclusion. Removing or de-emphasizing parameters like gender or ethnicity, where irrelevant, can help reduce bias.

3. Advanced AI Capabilities: AI systems should be designed to move beyond simple correlations and consider broader societal contexts. This could involve training models to understand causation or applying affirmative action principles deliberately to counteract systemic discrimination.

4. User Education: Raising awareness among users about how AI systems work and the biases they may inherit can empower individuals to critically evaluate AI outputs rather than accepting them as neutral truths.

5. Continuous Monitoring and Iteration: AI systems must be regularly tested for unintended biases, and processes should be in place to adjust algorithms or training data as necessary. Bias mitigation is not a one-time fix but an ongoing responsibility.

By recognizing that no decision-making system, including AI, is value-free, we can begin to construct AI systems that promote fairness, equity, and transparency. Addressing algorithmic biases is not just a technical challenge but a societal imperative to ensure AI serves all people equitably and responsibly.

The EU’s proposed Artificial Intelligence (AI) Regulation, heralded as a significant step toward global leadership in technology regulation, has drawn mixed reactions from civil rights advocates. While some elements of the draft, like restrictions on facial recognition in public spaces, mark progress, the legislation contains loopholes and gaps that may undermine its effectiveness, particularly for marginalized communities. This review explores key critiques raised by Sarah Chander during her interview with Angela Chen, highlighting the regulation's shortcomings and the broader implications for social justice.

Facial Recognition: A Ban in Name Only

One of the regulation’s most discussed provisions is the ban on facial recognition technology in public spaces. However, Chander points out that the numerous exceptions dilute its impact. Exemptions for counter-terrorism and serious crime, for example, create opportunities for discriminatory application, particularly against Muslim, Black, and Roma communities. This parallels concerns from the U.S., where predictive policing and data-driven immigration systems disproportionately target marginalized groups. Without stricter safeguards, the regulation risks perpetuating racial profiling and systemic inequalities.

Overlooking High-Risk Technologies

The draft regulation classifies some AI applications as "high-risk," requiring oversight. However, the self-assessment model it proposes lets developers evaluate their own compliance, a system Chander criticizes as overly lenient and commercially driven. By prioritizing corporate interests over fundamental rights, this approach weakens accountability. Moreover, certain technologies, like those automating sensitive identity traits (e.g., race, gender identity, disability), aren’t even categorized as high-risk, leaving them largely unregulated. This oversight is alarming given the potential for these tools to be used invasively or for political abuse.

Bias Beyond Technical Fixes

A recurring theme in Chander’s critique is the fallacy of "de-biasing" inherently flawed systems. Predictive policing serves as a key example: it operates within frameworks of racial and class inequality that no amount of technical refinement can resolve. In such cases, the only ethical solution is to ban the technology outright. This stance challenges the common narrative that better datasets and algorithms can resolve AI bias, underscoring the need for broader structural change.

The Democracy Gap in AI Regulation

The decision-making process for defining high-risk technologies is another contested area. Currently, the European Commission, an unelected body, holds exclusive authority over these classifications, leaving little room for public or civil society input. Chander emphasizes the importance of democratizing this process to ensure that those most affected by AI technologies—often marginalized groups—have a say in shaping the regulatory framework. This lack of inclusivity not only undermines accountability but also risks perpetuating systemic inequities in how AI is deployed.

Ambiguous Safeguards Against Exploitation

The regulation’s provisions on exploitation also raise concerns. While it prohibits exploiting vulnerabilities based on age or disability, it sets a high bar by requiring proof of physical or psychological harm. This ambiguous language appears to allow certain forms of exploitation, such as targeted advertising, as long as they don’t meet the threshold of harm. Chander notes the potential overlap with the Digital Services Act, adding to the uncertainty about how these measures will be implemented and enforced.

Challenging the Push for AI Adoption

A broader critique addresses the EU’s overarching goal of promoting AI adoption across sectors. Chander argues that this approach primarily benefits private-sector interests, particularly in public services, without adequate attention to human rights. Instead, AI deployment should be contingent on demonstrable benefits to people and compliance with ethical standards—a caveat missing from the current draft.

Key Battlegrounds Moving Forward

As the draft regulation undergoes further debate, several issues are likely to become focal points. These include expanding the scope of prohibited technologies, enhancing safeguards for high-risk AI, and democratizing the regulatory process. Ensuring meaningful civil society participation and prioritizing human rights will be critical to making the regulation more inclusive and effective.

Conclusion

The EU’s draft AI regulation is a step in the right direction, but it falls short of its potential to safeguard vulnerable communities and address systemic inequities. Without stronger prohibitions, external oversight, and inclusive decision-making, the legislation risks perpetuating the very biases it seeks to mitigate. As Chander aptly notes, AI’s general adoption should not be a policy goal in itself; instead, the focus must be on ensuring that technology serves all members of society equitably and ethically.