Menu

From Bytes to Bias: Addressing Gender Inequality in Artificial Intelligence Systems

Publications, Research paper

Abstract

This paper assesses whether the Indian Law sufficiently captures the gender bias in AI and efficiently governs it. It also analyses the current Indian laws and the need for proper legislations on AI. To end this trend of gender biases in AI systems, the direct and indirect impact of gender bias in AI will be analysed through a global lens and the need for gender equality will be implied. This paper concludes that not just legislative policy measures but accompanying policy measures are important to ensure a non-discriminatory approach to AI systems.

Keywords

Artificial Intelligence, Bias, Governance, Inclusivity, Technology

Introduction

Artificial intelligence (AI) is revolutionising various facets of our human lives from new technology, healthcare, industry, and governance and administration. With the onset of AI softwares like BARD AI given by Google or the all famous Chat GPT, the traditional and archaic way of doing things from content writing to coding has changed drastically.

What is Artificial Intelligence?

Artificial Intelligence is the duplication of human intelligence, where they are programmed to mimic human tasks with human-like learning abilities and thinking. AI facilitates the creation of special-purpose machines capable of accomplishing cognitive tasks. In some cases, they might do so even better than humans.[1]

The pervasive integration of AI in our contemporary landscape is reshaping our future. As more and more organisations started using AI, there has been an unprecedented surge in innovations, new technology and efficiency. AI is revolutionising human life – from personalised recommendations in e-commerce to predictive analysis in healthcare, AI has shown exponential growth. Society has started embracing AI and machine learning algorithms and the demand for skilled AI professionals has started escalating.  This technological metamorphosis has also called upon the need for ethical considerations. It poses certain risks and challenges such as bias and unfairness, job displacement, transparency issues, security risks, data privacy issues and many more. However, this paper will primarily focus upon the gender bias observed in AI.

A few instances of gender bias in AI are as follows –

  1. Facial recognition: Facial recognition systems misidentify black people at an alarming rate especially women. This leads to the possibility of misidentification of people. There have occurred some instances, where the police have also wrongfully arrested people. [2]
  2. Recruitment Algorithms: AI-powered hiring tools have been found to exhibit bias by favouring male candidates for jobs than female candidates. For example – Workers who used platforms like Indeed and LinkedIn, or social media platforms like Facebook and Twitter to find new job opportunities, realised that selecting their gender as female would result in lesser high paying jobs than selecting their gender as male.[3]
  3. Voice Assistants: Voice assistants have been often criticised for having female voices and answering to gender biassed queries in a stereotyped and submissive manner. They’re an example of how gender biases can be hard-coded into technology.[4]

There are many other examples such as online ad targeting and automated decisions in finance etc. Mitigating gender bias in AI is not just a technical challenge but a societal imperative and a policy decision which can be enforced through law.

Research Methodology

Objectives

  1. Identify instances of gender bias in AI applications
  2. Examine factors contributing to gender bias
  3. Evaluate the impact of gender bias in AI
  4. Analyse how effectively the current legislations are in solving the problem

Review of Literature

Need for Gender Equality in Technological Fields

Gender equality in the tech industry is imperative for fostering innovation, ensuring diversity not just in thoughts and ideas but also in action. Historically, women have been underrepresented in STEM fields, limiting this diversity in thoughts, ideas and action in tackling the challenges only to realise that the challenges today have become complex. Achieving true gender equality is not just a matter of fairness but also a strategic advantage, diverse teams have shown to outperform homogenous teams.[5] They have more creativity and problem solving ability. The need for gender equality in tech is not just an in office issue, it extends beyond the workplace and is not just an employee problem. It influences the development of technologies themselves. In the realm of Artificial Intelligence (AI), gender bias has emerged as a pivotal concern and a crucial issue. The lack of gender diversity in the developmental teams in developing AI has led to biased algorithms, reinforcing stereotypes and discrimination. A machine learning model that was supposed to be gender neutral has started perpetuating societal inequalities. As the tech industry continues shaping the future, fostering gender equality becomes not just a moral imperative but a prerequisite moving forward. Embracing diversity is fundamental to addressing challenges posed by the gender bias in AI and ensuring that technology contributes positively to a more inclusive society.

Gender Bias in AI

Any study of bias in AI must recognise that such prejudices are primarily driven by humans’ inherent biases. The systems and models that we build and train are a mirror of who we are.[6] There is no one root cause of this bias, trying to find a singular factor would be wrong. Some factors include:

  1. A training dataset that is incomplete or skewed: This occurs when demographic categories are missing from training data. When these models are put into effect to fresh datasets, they fail to scale adequately. For instance, in the context of teaching a computer to recognise voices, if the training data contains predominantly male voices and a very small proportion of female voices, say 5%, then the machine learning model would struggle to recognise, understand and respond to female voices. It would face problems with diversity.
  2. Labels used for training: Most commercial AI systems learn from labelled data to figure out the method of functioning. It is humans, who usually provide these labels, but since people often have biases whether they realise it explicitly or not, these biases then enter the AI models. Suppose a person likes dogs over cats, and he/she even unknowingly favour dogs in their labelled data, the AI system will learn that bias. So when asked to identify an animal, it is not aware of, the AI will show its biases towards dogs. In the same way, AI systems can replicate biases present in their training sets and later lead to unfairness especially when it comes to gender.
  3. Modelling approaches and features: The measurements that are used as inputs for machine-learning models, as well as the actual model training, could result in bias.

Best Practises for Machine-Learning Teams to Avoid Gender Bias

  1. Maintain equilibrium in training samples by include a comparable amount of audio samples from both men and women in the training dataset.
  2. By enlisting labellers from different walks of life to annotate the audio samples, you may ensure a varied range of viewpoints.
  3. Encourage machine-learning teams to carry out independent accuracy testing across diverse demographic groups, urging the detection and correction of any unfavourable treatment of certain categories.
  4. Remove potential prejudices by collecting more training data from sensitive groups. Implement sophisticated machine learning de-biasing techniques that handle not just identification errors but also instances that contribute to the unfairness connected to the main variable.

Effects of Gender Bias in AI

Direct

  1. Discrimination in Decision-Making: Gender bias in AI systems can lead to discriminatory decisions in areas like hiring, getting approvals for loans and other decisions. This is where algorithms may favour or disadvantage individuals gender based.
  2. Reinforcement of Stereotypes: AI models reflecting gender bias can perpetuate and reinforce existing stereotypes, since they go from humans to machines contributing to the amplification of gender-based societal norms and expectations.
  3. Underrepresentation and Invisibility: Gender bias in AI may result in underrepresentation or outright exclusion of certain groups. This leads to marginalisation of women or non-binary groups.
  4. Unintended Consequences: Biased algorithms can generate unintended consequences, affecting various aspects of individuals’ lives, from career opportunities to access to resources, based on their gender.

Indirect

  1. Erosion of Trust: Once gender bias in AI is noticed, the entire trust erodes. As users may become sceptical of technologies that lead to discriminatory outcomes. It leads to a lack of confidence in the fairness and reliability of AI.
  2. Inequitable Social and Economic Impact: Indirectly, gender bias in AI can contribute to broader social and economic inequalities, limiting opportunities and exacerbating disparities between genders in areas influenced by AI technologies.
  3. Normalisation of Bias: Biased AI systems may influence human behaviour and attitudes and also eventually lead to normalisation of bias in societal structures.
  4. Impaired Innovation: Gender bias can hinder innovation as it limits the diversity of perspectives hindering the creation of inclusive and effective solutions that cater to every individual.
  5. Ethical and Legal Challenges: The indirect consequences include ethical and legal challenges, as biased AI may lead to violations of privacy, equal opportunity, and human rights, requiring regulatory frameworks to address these concerns.

Laws Governing AI in India

  1. The Information Technology Act of 2000 (IT Act):

The IT Act is the primary legislative framework managing electronic transactions and digital governance. While AI is not specifically mentioned in the Act, it does impose certain duties on AI-related movement. Notably, Section 43A of the IT Act covers compensation for privacy violations caused by careless treatment of sensitive personal information. This is particularly pertinent in the context of AI systems that deal with user data. Section 73A is another intriguing clause, emphasising the significant dangers associated with the use of sensitive information in the complicated environment of AI, where machines make inferences from subtle data patterns. Section 43A thus serves as a safeguard, holding companies liable for any failures to preserve user privacy inside AI frameworks. Furthermore, this provision is consistent with the ever-changing character of technological advances, recognising the dynamic problems offered by applications that use AI. Furthermore, Section 73A of the IT Act enhances the legal structure around electronic transactions and digital governance. It is possible that, as a supplementary provision, it covers intricate aspects of technology governance, giving a holistic viewpoint to legal matters in the quickly expanding digital domain. While the Act fails to spell out AI norms, it does stress their adaptability and inclusion.

  1. The Indian Copyright Act of 1957 protects unique literary, artistic, musical, and theatrical creations by granting exclusive rights to authors and banning unauthorised use or duplication. The rise of AI-generated material has caused debates about copyright ownership and infringement responsibility. The decision in Gramophone Company of India Ltd. v. Super Cassettes Industries Ltd. (2011) by the High Court of Delhi, demonstrating that AI-generated music created by a computer programme lacks human originality and, as a result, does not qualify for copyright protection, is noteworthy. The decision of this case determines the requirements for copyright registration for AI-generated content in India.
  2. The National e-Governance Plan (NeGP): This plan of India, that was unveiled in 2006, seeks to empower society digitally by offering government services online that are effective and affordable. Artificial Intelligence (AI) plays an essential part in the NeGP, automating procedures, improving decision-making, and improving citizen services across many government agencies. Routine jobs are automated via the use of AI, decision support systems aid in data-driven governance, and chatbots provide rapid assistance on government portals. Future planning is guided by predictive analysis, and personalised citizen services are offered based on AI-driven suggestions. AI augments security procedures for fraud detection, protecting the integrity of government data. Furthermore, AI promotes integration by building approachable apps, and analytics provide insights for continual development of e-governance programmes.
  3. The Indian government’s new New Education Policy (NEP): It offers dedicated coding workshops for children beginning in the sixth grade, indicating a strategic emphasis on technology skill development. This project is in line with the objective of promoting India as the next technological hub, acknowledging the value of early coding exposure for nourishing creativity in the age of Artificial Intelligence (AI). The NEP intends to build a tech-savvy workforce capable of contributing to the ever-evolving world of AI and technology-driven innovation through integrating coding into the curriculum. This imaginative approach not only offers students practical skills, but it additionally promotes problem-solving talents and creativity, both of which are essential for success in AI-related fields.

Loopholes in the Indian Law

The existing laws in India addressing AI lack explicit protections that adequately prevent and reduce gender bias in AI systems. While generalised data protection regulations restrict the use of personal information, such as the Personal Data Protection Bill, 2019, they do not expressly regulate the algorithms and models employed in AI, allowing leeway for possible discrimination. The lack of specific laws on AI ethics, as well as solid processes for holding developers and organisations accountable for gender prejudice, add to the difficulties. Furthermore, existing anti-discrimination laws do not apply to AI systems, and labour laws address gender bias in AI-driven job choices insufficiently. The lack of explicit legislation for AI in the workplace, as well as obligations for frequent audits and monitoring, impede proactive detection and rectification of gender prejudice. Comprehensive reform is required, including specific provisions that explicitly address gender bias, define ethical considerations, establish mechanisms for accountability, expand anti-discrimination laws to include AI, and mandate regular audits and monitoring, ensuring a legal framework in India that promotes fairness, equality, and responsible use of AI technologies.

Methods

  1. Literature Review: Conduct an extensive review of scholarly books, articles, research papers to study and examine the gender bias in AI and analyse common trends in it. Analyse existing frameworks and legislations in this regard.
  2. Data Collection: Collect data from various sources including but not limited to government reports, AI research organisations, industry reports and documented cases.
  3. Data Synthesis: Systematically organise the entire data to give a logical flow to it and form conclusions.
  4. Critical Analysis: Look at the strengths and weaknesses in the existing research. Identify gaps in the current research and develop on it.
  5. Conclusions and Recommendations: Summarise the key findings, the mitigating factors and propose recommendations to find a way out. Discuss potential areas for future research and policy implications.

Suggestions

Concerns concerning bias, intolerance, and justice have become crucial as Artificial Intelligence (AI) begins to play a growing part in different aspects of society. The prevalence of gender bias in AI systems, which can perpetuate and even aggravate existing gender inequities, is one significant concern. Recognising this, the development of laws and regulations to combat gender bias in AI is critical in India to ensure the equitable and ethical usage of AI technological advances.

  1.  Bias Audits and Reporting Requirements: To address gender bias in AI, the Indian government could introduce laws requiring yearly bias audits for AI systems, particularly those used in major sectors like banking, healthcare, and employment. These audits would include detailed evaluations of AI systems to detect and correct gender bias. Organisations and businesses that use AI would be compelled to publish audit results reports, encouraging openness and accountability. By include this in the legal framework, it guarantees that the responsibility for correcting gender prejudice is shared by both AI system makers and users.
  2. AI Development Ethical Guidelines: To avoid unintentionally biases, clear ethical norms for AI research must be established. Legislation can require the incorporation of ethical issues, including gender neutrality, into the AI system development process. Promoting diversity in the creation of artificial intelligence teams, doing impact assessments to analyse possible biases, and including fairness measures throughout the development life cycle are all part of this. The legal framework fosters the adoption of ethical practices and links AI development with the community ideals of justice and equality by codifying these principles.
  3. User Acceptance and Explainability: Legislation could require AI systems to offer transparent reasons for their inferences, particularly for sectors affecting people’s lives such as hiring, lending, and healthcare. Users should be able to comprehend how AI systems make judgements and the potential consequences for gender-related results. Furthermore, voice user arrangement for the usage of gender-related data in AI systems should be required by law. This promotes autonomy and transparency by ensuring that individuals know about and have influence over the data used to train and run AI systems.
  4. Quotas for Diversity and Inclusion in AI Development: To address the lack of diversity in AI development, legislation may establish quotas or provide incentives for women and underrepresented groups to join AI research and development teams. The legal framework intends to eliminate unconscious biases and ensure that AI systems are constructed with a broader viewpoint by fostering diversity. This not only helps to reduce gender prejudice, but it also improves the overall efficacy and justice of AI technologies by embracing different points of view during the research and development process.
  5. Anti-discrimination legislation has been extended to AI: Existing anti-discrimination legislation can be modified to specifically include AI systems. This implies that if an AI system produces discriminatory outcomes based on gender, it will face legal penalties, just like human judgements. This legislation expansion strengthens the commitment to fighting gender prejudice and holds AI systems accountable for the impact they have on people, establishing a culture of responsibility in the development and deployment of AI technology.
  6. Monitoring and adaptation on a continuous basis: Legislation might require continual monitoring of AI systems once they are deployed to detect and address any developing gender biases. Artificial intelligence (AI) technologies change, and continual evaluation is necessary to ensure that biases are recognised and removed as soon as feasible. This legislative requirement emphasises the dynamic nature of AI systems and the necessity for continual monitoring to preserve fairness and prevent gender prejudice from recurring over time.

Conclusion

To summarise, tackling the prevalent issue of gender bias in Artificial Intelligence (AI) in India requires immediate and thorough action, particularly within the present legal framework. The existing environment shows serious flaws in the rules regulating AI, which offer substantial ethical, social, and economic issues as AI permeates a variety of society. The current data protection rules, such as the Personal Data Protection Bill, 2019, while helpful in safeguarding personal information, lack the clarity necessary to control the algorithms and models vital to AI systems. While these rules emphasise human privacy, they fail to address the ethical implications of AI, including prejudice based on gender.

Furthermore, the legal framework’s lack of clear regulations on AI ethics causes ambiguities in resolving gender bias concerns. Responsible AI technology development, deployment, and use need clear rules, yet present laws lack explicit restrictions that would guide AI stakeholders in minimising gender prejudice. An inclusive legislative framework should include rules addressing AI ethics explicitly, assuring a responsible approach to AI development, and reducing the potential disadvantages associated with gender prejudice.

Finally, as India grapples with the issues raised by gender bias in AI, extensive legal change is required. Specific clauses that expressly address gender prejudice, describe ethical issues, develop robust accountability procedures, grow discrimination laws to incorporate AI, and demand frequent audits and monitoring should be included in such transformation. By adopting these steps, India may provide the groundwork for a legislative framework that not only overcomes present constraints but also assures the responsible, fair, and ethical use of AI technology. As India navigates the growing AI landscape, a proactive and comprehensive legislative strategy is required to maximise the promise of technology while preventing unforeseen and negative repercussions for gender equality.

[1] Scharre, Paul, et al. “What Is Artificial Intelligence?” ARTIFICIAL INTELLIGENCE: What Every Policymaker Needs to Know, Center for a New American Security, 2018, JSTOR, at 4 http://www.jstor.org/stable/resrep20447.5.

[2] AMNESTY INTERNATIONAL, https://www.amnesty.ca/surveillance/racial-bias-in-facial-recognition-algorithms/#:~:text=Misidentification%20in%20facial%20recognition%20technology,darker%20skin%20tones%E2%80%94especially%20women. , (last visited Nov 13, 2023)

[3] EURONEWS.NEXT, https://www.euronews.com/next/2022/03/08/gender-bias-in-recruitment-how-ai-hiring-tools-are-hindering-women-s-careers#:~:text=2.,perpetuating%20the%20gender%20pay%20gap. , (last visited Nov 13, 2023)

[4] UNESCO, https://en.unesco.org/EQUALS/voice-assistants , (last visited Nov 13, 2023)

[5] FORBES, https://www.forbes.com/sites/forbesagencycouncil/2023/04/18/creating-a-diverse-and-inclusive-team-for-long-term-success/?sh=7ae939525314 , (last visited Nov 13, 2023)

[6] HARVARD BUSINESS REVIEW, https://hbr.org/2019/11/4-ways-to-address-gender-bias-in-ai , (last visited Nov 13, 2023)