Algorithms shape our daily interactions through the digital tools and systems integral to modern life. As these automated systems enhance efficiency and accuracy, they also raise a crucial ethical question: Is algorithmic bias inevitable? The widespread use of algorithms not only affects individual lives but also has the potential to perpetuate and reinforce historical injustices and societal inequalities, which are substantial stakes in our increasingly digitized world. This essay will employ a tripartite structure, beginning with a preliminary understanding of algorithms and bias, followed by two main arguments—the inevitability of data bias due to inherent limitations in data collection and the persistence of embedded biases within the structural design of algorithms. This essay argues that algorithmic bias is inevitable, and through a critical examination enriched by philosophical insights, it will explore both the challenges and potential strategies for creating more equitable algorithmic systems.

Foundations of Algorithmic Bias

The inquiry into the inevitability of algorithmic bias must start by defining what algorithms are and how biases can manifest within them. As articulated by Thomas H. Cormen (2009), algorithms are structured sequences of instructions designed to solve problems or perform tasks1. These sequences are fundamental to computational systems impacting diverse areas of life. Parallel to the technological definition, biases, particularly cognitive biases discussed by Daniel Kahneman (2011), describe systematic errors in judgment that can significantly influence decisions, including those made by algorithms2.

With biases often embedded from their inception, the development of algorithms has mirrored the evolution of computing technology. Walter Isaacson (2014) illustrates how early computational tools were shaped by the societal and cultural norms of their times, often laden with prejudices. These early biases were reflections and amplifiers of existing social inequities. For instance, the use of algorithms in predictive policing and credit scoring in the mid-20th century often entrenched societal biases where systems disproportionately affected marginalized groups such as racial minorities and individuals from lower socio-economic backgrounds. Recognizing these historical contexts is crucial for developing more equitable technologies today, thus necessitating a proactive approach to identifying and mitigating bias in modern algorithms.

Diving deeper into the types of biases that afflict algorithms reveals the extent of their pervasiveness in today’s society. Data bias occurs when the datasets used to train algorithms lack representativity or are skewed towards particular demographics, as seen in the research paper by Joy Buolamwini and Timnit Gebru (2018), which highlighted significant racial and gender biases in facial recognition technologies.3 Another type of essential bias is algorithmic processing bias, which refers to biases that arise from the algorithm's design itself, where decisions about which variables to consider and how to weigh them can inherently favor certain outcomes over others. For instance, recruitment algorithms that prioritize candidates from prestigious educational institutions reflect and reinforce elitist values. This design choice systematically excludes qualified candidates from non-traditional or less recognized backgrounds, perpetuating socio-economic divisions and affirming existing power structures. Outcome bias, the third type, is evident when the applications of algorithms produce results that systematically disadvantage or privilege certain groups. This type of bias is often most visible post-deployment, demonstrated by Kristian Lum and William Isaac, who showed how predictive policing tools disproportionately target minority neighborhoods, reinforcing existing racial prejudices.4 Each category illustrates how biases are ingrained in algorithmic systems and also underscores the challenges of extracting these biases without comprehensive changes in both the technology and the societal structures that inform its development. This analysis is crucial for the next sections of the essay which will explore the philosophical implications and potential strategies for mitigating these biases, ultimately questioning the possibility of creating truly neutral algorithmic systems.

Examining the Inherent Biases of Data Collection and Algorithmic Processing

As we transition from a general understanding of algorithms and their biases to a focused examination, we encounter one of the most consequential forms of algorithmic bias: the intrinsic flaws in data collection and processing. These biases being embedded in data-driven systems significantly influence algorithmic outputs, thus affecting everything from individual decisions to societal structures. The biases in data highlight ethical challenges, magnifying historical injustices and societal inequalities in our digital world. This section will dissect the origins, manifestations, and impacts of these biases, advancing the thesis that algorithmic bias is ultimately inescapable and necessitates a critical philosophical approach to mitigate its effects and strive for fairness in algorithmic systems.

Exploring the philosophical implications of data biases, we can invoke the theories of technology and society interaction, such as those proposed by Bruno Latour (2005). Latour’s actor-network theory suggests that technologies and society are co-constitutive, meaning algorithms are not merely tools but participants in social networks that influence and are influenced by social dynamics.5 This interaction shows how algorithms, reflecting and amplifying biases of their creators and societal structures, are embedded with bias. As non-human actors, algorithms contribute to shaping societal norms and expectations, not just reflecting but also co-creating reality. This co-creative role makes it particularly challenging to disentangle biases from the algorithmic decisions that emerge from these complex networks. Latour’s perspective challenges the notion of technological neutrality and underscores the role of ethical foresight in technology design. It compels us to consider algorithms not just as neutral artifacts but as integral components of a network that includes human actors, social practices, and technological infrastructures. Furthermore, this recognition opens up a pathway for developing more robust ethical frameworks that address the creation and use of algorithms and their broader socio-technical impact. Such frameworks must strive to account for the multiplicity of interactions and unintended consequences that arise when algorithms are deployed in diverse and complex social settings. Therefore, the inevitable entanglement of algorithms with societal biases makes it clear that while we can strive to mitigate these biases, they can never be fully eliminated.

Further delving into the philosophical implications of algorithmic biases, we confront the pervasive issue of selection bias in data collection. Selection bias fundamentally challenges how knowledge is created and represented in algorithmic systems. It occurs when the sample chosen for a study is not representative of the population being studied, thus leading to results that are systematically prejudiced due to non-random sampling. This is seen when datasets predominantly include certain groups over others which results in the distorting of the algorithm's perception of the world. This is not merely a technical flaw but a significant epistemological error, as it questions the validity of algorithmic 'knowledge'. Philosophers like Miranda Fricker (2007) discuss this as an ethical problem, terming it "epistemic injustice", where some individuals’ experiences and realities are given less recognition in the production of knowledge.6 For instance, if algorithms were trained on data from affluent users, its utility may diminish for lower-income groups, unfairly excluding them. This exclusion is not just a lapse in data inclusivity but also a perpetuation of existing social structures that marginalize these groups, making bias inevitable in the outputs of such algorithms. The situation exemplifies what feminist philosopher Nancy Fraser terms "misrecognition", where social and economic structures systematically deny certain groups the same level of participation and respect as others.7 Hence, algorithmic bias, rooted in selection bias, reinforces the social hierarchies and disparities, suggesting that any technological solution developed under such conditions is inherently compromised. These biases are so deeply embedded in society that disentangling them would require not just technological adjustments but radical societal changes—a challenge that further highlights the inevitability of such biases in algorithmic processing.

Further developing this section, we discuss the use of historical data in algorithms and how they can embed long-standing societal biases into automated systems, making these tools agents of the status quo rather than instruments of change. This critical issue aligns with Iris Marion Young's philosophical discourse on structural injustice (2011), which posits that societal structures covertly transmit injustice, often without the conscious intent of individuals.8 Predictive policing algorithms that rely on crime data from past decades are a salient example, as they may inherit and amplify the racial biases embedded within those historical patterns. This leads to a disproportionate targeting of minority communities, reinforcing a cycle of marginalization and surveillance. Such practices embody what Young describes as "systemic oppression," where the injustice is embedded in the unquestioned norms and institutional practices. Moreover, the reliance on historical data without critical evaluation or adjustment perpetuates these biases, questioning the role of algorithms in perpetuating or potentially challenging these entrenched injustices. Philosophically, this raises questions about the moral responsibility of technologists and policymakers to recognize and rectify these biases. The notion of "responsibility" in this context transcends mere awareness, demanding proactive measures to redesign systems that do not merely mirror but critically engage with and transform unjust societal structures.

The challenge of data completeness in algorithmic processes raises profound philosophical questions about what constitutes sufficient knowledge for fair and effective decision-making. When algorithms operate with incomplete datasets, they rely on a partial view of reality, which inherently questions the justifiability of their conclusions and undermines their ability to make fair decisions. Helen Nissenbaum's concept of "contextual integrity" in information processing highlights this issue, arguing that privacy and ethics in data must consider the appropriateness of how information flows within specific contexts.9 Circling back to our foundational example; credit scoring algorithms that overlook informal employment or non-traditional financial practices might systematically disadvantage certain groups, thus promoting a narrow and potentially discriminatory understanding of financial reliability. This violates the contextual integrity and perpetuates socioeconomic disparities, as the data used fails to represent the complete financial behaviors of individuals and ultimately leads to unfair penalizations. Philosophically, this reflects a deeper epistemic concern about fairness: if knowledge is power, then incomplete knowledge confers a flawed form of power that can reinforce rather than dismantle structural inequities. The ethical imperative therefore extends beyond simply gathering more data to seeking better and more contextually relevant data that respects the complexities of human lives. By addressing the limitations of data completeness, we confront not just a technical challenge but a significant ethical demand to reassess what we consider 'relevant' data. This involves shifting from a quantitative to a qualitative assessment of information, emphasizing the importance of diverse data sets that more accurately reflect the multifaceted realities of all individuals. Thus, enhancing data completeness is not just about filling in the gaps but about reimagining how algorithms can serve as tools for justice, actively working to rectify rather than mirror the biases and inequalities present in society.

This detailed analysis of data collection flaws elucidates the philosophical foundation of algorithmic biases, revealing them not simply as technical errors but as deep ethical issues that reflect existing social inequalities. Each example underscores the necessity for a philosophical approach that considers the broader implications of data biases. It advocates for strategies that promote fairness, justice, and inclusivity in algorithmic decision-making. In doing so, we recognize that while the complete eradication of these biases may be unattainable, their recognition and mitigation are essential. Such efforts are crucial in striving towards a more equitable digital future. By continually engaging with these philosophical and practical challenges, we can better understand and shape the technological landscape to serve societal needs more justly.

Diving into our rebuttal, there is an optimistic assertion that advanced methodologies and technology can create bias-free data collection systems is an example of technological determinism, a theory that presumes technological progress inevitably leads to social improvement by solving inherent societal issues. This view fundamentally overlooks how technologies embody and enforce existing social hierarchies and power dynamics, as highlighted by philosopher Langdon Winner. He argues that technologies are shaped by and reflect the biases and priorities of the societies that create them, meaning that no technological advancement can be completely neutral or free from human prejudices.10 For example, improvements in facial recognition technology may decrease individual errors but still reinforce a surveillance infrastructure disproportionately targeting minority groups, thus exacerbating issues of privacy invasion and social discrimination. This raises ethical concerns about autonomy and equity and also demonstrates how technological 'solutions' can further entrench systemic inequalities rather than alleviate them. Furthermore, relying on technology to rectify biases assumes that all potential sources of bias are known and can be quantified and corrected by algorithms, an assumption that fails to account for the complex, dynamic nature of human biases that are often invisible and insidious. Such an approach also diverts focus from the need for structural societal changes, suggesting that technological fixes alone can achieve fairness and justice without addressing the deeper, ingrained injustices that permeate societal institutions. Consequently, while technological advancements can aid in identifying and perhaps mitigating some aspects of bias, the belief that they can completely eliminate such biases is not only overly optimistic but dangerously misleading, leading to complacency in addressing the root causes of these biases. It is essential, therefore, to maintain a critical perspective on the limitations of technology in social contexts, recognizing that real progress requires a combination of technological innovation, critical ethical reflection, and rigorous societal reforms to truly move towards equity and justice in an increasingly digitized world.

While diversifying data sources seems promising, it may not resolve deep-seated biases and can fail to lead to equitable outcomes. This view aligns with John Rawls’ theory of justice, which emphasizes fairness as the foundational aspect of ethical social institutions. However, simply adding more data from diverse sources often fails to resolve deep-seated biases and may not automatically lead to more equitable outcomes.11 This is because systemic power imbalances that affect how data is used and for what purposes remain unaddressed. For example, increasing the representation of women in datasets used for job recruitment algorithms does not necessarily eliminate gender bias if the underlying societal valuations of professions remain gendered. Rawls' concept of "justice as fairness" suggests that true justice requires the equal distribution of resources as well as the restructuring of societal institutions to promote fairness. Thus, while diversifying data sources is a step toward fairness, it must be accompanied by changes in the algorithmic models themselves and the societal contexts in which they are deployed. This requires a critical examination of data construction and use, ensuring they contribute to social justice and not just replicate existing hierarchies.

In concluding the argument, it is essential to recognize that while mitigation strategies can reduce the instances and impacts of algorithmic biases, the complete eradication of these biases may not be feasible. The critical philosophical examination of algorithmic biases showcases the profound challenges at the intersection of technology and ethics. As this philosophical examination demonstrates, algorithmic biases are deeply intertwined within our societal values and structures. While technical solutions can mitigate some aspects of these biases, a broader philosophical approach is required to address the ethical challenges they pose comprehensively. This approach must be continuously evolving as our technologies and societies change. By fostering a critical discourse on algorithmic bias, we can strive for more equitable algorithmic systems that respect and enhance human dignity and society's collective well-being. The journey toward understanding and mitigating algorithmic bias is not merely a technical challenge but a profound endeavor that calls for rigorous scrutiny, thoughtful reflection, and earnest dialogue across various domains of knowledge and society. This comprehensive engagement is necessary for creating fairer algorithms but also for advancing a more just and understanding digital age.

Examining the Inherent Biases in the Structure and Design of Algorithms

Building on our discussion of data biases, we now focus on inherent flaws in algorithmic design. Just as biases are embedded in the data itself, they also permeate the very design of algorithms by shaping their outputs and perpetuating societal inequalities. This section will dissect the origins, manifestations, and impacts of biases in algorithmic design, advancing the thesis that algorithmic bias is inevitable and deeply entrenched within algorithmic systems. By exploring the philosophical implications of biases in algorithm design, we aim to showcase the interplay between technological design and societal values, urging a critical reevaluation of algorithmic systems to strive for fairness and equity.

Algorithms are deeply enmeshed in the subjectivities of their creators. Cathy O'Neil (2016) dissects algorithms that are ostensibly designed for optimization but end up privileging certain groups over others.12 For example, she scrutinizes algorithms used in the educational sector that may assess teacher performance based on student test scores. These algorithms could disadvantage teachers working in underfunded schools, thereby perpetuating systemic inequities in education. O'Neil's critique highlights how the pursuit of efficiency—algorithmic decision-making streamlined to prioritize speed and cost-saving—can inadvertently lead to the entrenchment of social disparities. This is crucial because it underlines that algorithmic bias goes beyond mirroring societal inequalities; it becomes an agent in their perpetuation, thereby cementing the inevitability of algorithmic bias. O'Neil's analysis challenges the perceived neutrality of algorithms but also exposes the complex interplay between technological design and societal values. By demonstrating how seemingly benign metrics like test scores can exacerbate existing educational disparities, her work urges a reevaluation of what we consider to be 'fair' and 'efficient' in algorithmic systems. This broader impact on systemic inequality highlights the risk of adopting technological solutions without a nuanced understanding of their potential social consequences. It also emphasizes the responsibility of designers to consider the broader socio-political context in which their algorithms will operate, suggesting that ethical algorithm design requires an awareness of the histories and realities of marginalized communities. Thus, O'Neil's critique serves as a vital reminder that the choices made in algorithm development can have profound and lasting effects on social equity and justice.

Further delving into O’Neil’s critique, we conclude the design of algorithms is not a neutral process; it is loaded with moral implications. When we consider Kantian ethics, which advocates for actions to be guided by universal moral laws, we see a tension with the modus operandi of algorithmic efficiency (1797).13 Kant posits that actions must be undertaken with a sense of duty that respects the autonomy and dignity of all individuals. O'Neil's analysis shows us that algorithms focused on efficiency can contravene this ethical approach by objectifying individuals into data points to be processed, thus undermining their inherent dignity. This disregard for the individual aligns with the antithesis of Kantian ethics, as it places the algorithmic 'ends'—efficiency, profit, speed—above the means, which in this context are the people affected by these systems. This tension between Kantian ethics and algorithmic efficiency raises ethical concerns about the foundational principles of technology design. Kant's imperative to treat individuals not merely as means but as ends in themselves demands a reassessment of how algorithms are engineered and the criteria by which they are judged successful. If the design of algorithms continues to prioritize efficiency over ethical considerations, it risks perpetuating a cycle where technology reinforces rather than dismantles structural inequalities. This cycle alienates individuals while diminishing societal trust in technological advancements, as people grow wary of systems that seem to disregard their values. Therefore, embedding Kantian principles into algorithmic design could shift the focus towards creating systems that enhance human dignity and foster a more equitable society, ensuring that technology serves humanity holistically rather than fragmenting it through biased efficiencies.

To sum, the inherent biases in algorithm design underscore a critical aspect of technology's intersection with society—algorithms are not merely tools but also powerful actors that shape socio-political landscapes. The detailed exploration of design choices reveals how deeply the biases of creators and the social contexts of their times are embedded within these systems, illustrating the inevitable influence these biases exert on algorithmic outcomes. This reality calls for a more ethical framework in the design and deployment of algorithmic systems. By integrating this foresight and actively involving diverse disciplines in the design process, we can begin to address these biases. The development of algorithms must be continually scrutinized and challenged to ensure they serve to enhance human dignity and promote equity, rather than perpetuating systemic injustices.

Critics might argue that the potential for algorithms to operate with bias is overstated and that technological advancements are inherently capable of correcting these biases through improved models and smarter design choices. They posit that as machine learning and artificial intelligence evolve, these systems can learn to identify and correct for their own biases, potentially reducing human error and subjectivity in critical decision-making processes. For instance, researchers at Google have developed algorithms that aim to eliminate gender bias in machine learning models by altering the data fed into these systems, thereby ensuring more equitable outcomes. In the example mentioned, researchers at Google implemented methods that systematically adjust training data to achieve more balanced gender representation, an approach described by Zhang et al. (2018).14 This strategy involves re-weighting or re-sampling the dataset to decrease the prevalence of gender-biased labels that machine learning models might otherwise learn and perpetuate. While this technical solution appears promising, it raises critical questions about the depth and sustainability of such adjustments. Does altering dataset composition genuinely address the underlying biases, or does it merely mask them in output metrics? This issue highlights the complexity of relying solely on technical fixes, urging a deeper examination of how biases are embedded in the data and in the very algorithms that process this data along with the societal norms that shape these inputs. Moreover, while technological advancements in algorithms provide significant opportunities for reducing human error and enhancing decision-making, the notion that these systems can autonomously correct inherent biases remains overly optimistic. This perspective overlooks the deep-seated nature of biases, which are intricately woven into the societal structures that inform the creation and curation of datasets. As highlighted by Buolamwini and Gebru, we circle back to the example of facial recognition technology. This technology continues to manifest racial and gender biases, especially in misidentifying women of color. This persistent bias is not merely a technical flaw but a reflection of the data predominantly composed of white male faces, which itself is a byproduct of deeper, systemic societal biases. This issue of bias in facial recognition technology is not just a matter of algorithmic error but also illustrates a broader problem of representational equity within technological systems. The designs used act as a mirror, reflecting the historical and current inequalities found in society at large, thus ingraining these injustices further as the technology scales. Furthermore, when these biased systems are implemented in real-world applications—such as law enforcement or hiring practices—they perpetuate and often exacerbate the disparities they reflect, thereby contributing to a cycle of discrimination that continually marginalizes underrepresented groups. This underscores the need for inclusive data collection and training that accounts for diversity, aiming to dismantle rather than reinforce societal biases.

Ending this argument, the analysis reveals that while algorithms possess the potential to transform decision-making processes by reducing human error, their inherent biases, shaped by the subjective decisions and societal structures that inform their creation, challenge the notion of their objectivity and self-correcting capabilities. The examples discussed, from facial recognition technologies to predictive policing, demonstrate that biases in algorithms are not merely technical errors but are deeply intertwined with social, racial, and gender inequities that are perpetuated by these systems. These biases are not spontaneously arising nor easily rectifiable glitches; they are endemic to the data and the design processes that feed into algorithm creation. The persistence of these biases and the complexity of addressing them necessitate a critical reevaluation of how algorithms are developed, deployed, and monitored. The potential for algorithms to reinforce existing societal disparities underlines the urgent need for a concerted, multidisciplinary approach to algorithm design and implementation that prioritizes ethical considerations and social justice.

To effectively mitigate the biases related to the design of algorithms, as highlighted through the deep-seated societal structures and the subjective decisions of developers, a more integrated and holistic approach is required. Addressing these biases demands more than just technical adjustments; it requires a fundamental shift in the conception and development of algorithmic systems. Incorporating a diverse array of perspectives in the design process is crucial. Benjamin et al. (2019) argue diversity in development teams brings varied experiences and viewpoints that can help anticipate and counteract biases that may not be evident to a homogeneous group.15 This diversity isn't limited to racial or gender differences but includes interdisciplinary expertise from fields such as social sciences, ethics, and community advocacy. These perspectives enrich development, aligning algorithms with real-world complexities and reducing unintended consequences. Engaging diverse disciplines can disrupt ingrained biases by injecting cultural and ethical considerations into technology. Moreover, this broadened input is crucial for devising algorithms that operate equitably across a spectrum of social contexts, thereby reducing the likelihood that these technologies perpetuate existing societal inequities or introduce new forms of discrimination.

Furthermore, the commitment to transparency and accountability in algorithmic design plays a pivotal role in combating ingrained biases. As O'Neil critiques the lack of transparency around algorithmic decision-making can obscure biases and make them more difficult to challenge. By advocating for open documentation of data sources, design decisions, and the operational logic of algorithms, stakeholders including users and regulatory bodies can better understand and critique the systems. This transparency facilitates accountability while supporting external audits by independent parties, as recommended by Mittelstadt. Such audits are vital for continuously assessing the impact of algorithms and suggesting necessary adjustments. Additionally, implementing ethical algorithm standards and frameworks, such as those proposed by the IEEE or reflected in the EU's Ethics Guidelines for Trustworthy AI, provides a structured approach to ensure that algorithms are designed with fairness and non-discrimination at their core. Transparency acts as a preventative mechanism that reveals how decisions are made while exposing any misalignments or deviations from ethical norms early in the development process. This early detection is crucial because it allows for the correction of biases before they become deeply embedded within the system, thereby preventing the algorithm from further contributing to or creating new social injustices. Ultimately, creating an environment where algorithms are regularly evaluated and updated in response to new insights and changing societal values is essential. A dynamic management approach ensures that systems evolve ethically, mitigating the risk of exacerbating disparities.

The discussion on structural and design biases in algorithms underscores that these tools are embedded with human subjectivities and societal norms. Insightful studies as seen by Mittelstadt et al. highlight how deeply these biases are intertwined with systemic social inequities, necessitating a multifaceted approach to algorithm design. This approach emphasizes the importance of diversity in development teams, transparency in algorithm operations, and continuous ethical evaluation to effectively detect and mitigate biases. By restructuring the development processes and adapting these systems in response to changing societal values, we can harness the potential of algorithms to promote fairness and inclusivity. Such efforts are crucial in ensuring that technological advancements align with core values of dignity and equity, turning powerful computational tools into catalysts for social justice rather than perpetuating discrimination.

Conclusion

In this essay, we have delved into the inescapable nature of algorithmic bias, demonstrating how these biases are not mere anomalies but fundamental characteristics influenced by the social and historical contexts of algorithm development. From our initial discussion on the basic structures of algorithms and their inherent biases to deeper philosophical and ethical examinations, we have established that algorithms act not merely as tools but as actors within larger socio-technical networks that reflect and reinforce societal biases. The inevitability of algorithmic bias is rooted both in the data that train these systems and in the design processes that shape their operation, highlighted through perspectives such as Latour’s actor-network theory and the critical insights of scholars like O’Neil and Fricker. As the essay has shown, achieving fairness and justice in algorithmic systems requires more than technical fixes; it necessitates a sustained commitment to ethical practices, transparency, and inclusivity in design and implementation. By embracing a diverse array of perspectives and continuously scrutinizing the ethical implications of technological choices, we aim to develop algorithms that enhance our collective well-being rather than simply automating decisions. This work demands more than technical solutions; it requires a combined ethical and interdisciplinary approach to develop algorithms that genuinely serve and elevate society. While algorithmic bias may be inevitable given current societal structures, our proactive engagement and rigorous scrutiny of these biases can lead to more thoughtful and equitable technological advancements, fostering a society that values justice and inclusivity at its core.

Footnotes

1 Cormen, T. H., Leiserson, C. E., Rivest, R. L., & Stein, C. Page 89.

2 Kahneman, D. Page 23.

3 Buolamwini, J., & Gebru, T. Page 3.

4 Lum, K., & Isaac, W. Page 15.

5 Latour, B. Page 1.

6 Fricker, M. Page 6.

7 Fraser, M. Section 3.

8 Young, I. Page 4.

9 Nissenbaum, H. Page 119.

10 Winner, L. Page 19.

11 Rawls, J. Page 235.

12 O’Neil, C. Page 1.

13 Kant, I. Page 84.

14 Zhang, S. Section 1:3.

15 Benjamin, R. Page 85.

Bibliography

1. Benjamin, R. (2019). Race After Technology: Abolitionist Tools for the New Jim Code. Polity.

2. Buolamwini, J., & Gebru, T. (2018). Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification. Proceedings of Machine Learning Research, 81, 1-15.

3. Cormen, T. H., Leiserson, C. E., Rivest, R. L., & Stein, C. (2009). Introduction to Algorithms (3rd ed.). MIT Press.

4. Fraser, M. (n.d.). "Recognition." Internet Encyclopedia of Philosophy.

5. Fricker, M. (2007). Epistemic Injustice: Power and the Ethics of Knowing. Oxford University Press.

6. Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.

7. Kant, I. (1797). The Metaphysics of Morals. Cambridge University Press.

8. Latour, B. (2005). Reassembling the Social: An Introduction to Actor-Network-Theory. Oxford University Press.

9. Lum, K., & Isaac, W. (2016). To Predict and Serve? Significance, 13(5), 14-19.

10. Nissenbaum, H. (2001). "Privacy as Contextual Integrity." Stanford Encyclopedia of Philosophy.

11. O'Neil, C. (2016). Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy. Crown Publishers.

12. Rawls, J. (1971). A Theory of Justice. Harvard University Press.

13. Winner, L. (1993). "Upon Opening the Black Box and Finding It Empty: Social Constructivism and the Philosophy of Technology." Science, Technology, & Human Values, 18(3), 362-378.

14. Young, I. M. (1990). Justice and the Politics of Difference. Princeton University Press.

15. Zhang, S., Yao, L., Sun, A., & Tay, Y. (2019). "Deep Learning Based Recommender System: A Survey and New Perspectives." ACM Computing Surveys (CSUR), 52(1), 1-38.