Artificial intelligence (AI) has become an indispensable part of our lives, from virtual assistants like Siri and Alexa to driving cars. However, as the use of AI systems increases, so does the possibility of discrimination. Who should be held responsible for this discrimination? Is it the fault of the AI system itself, or are developers, programmers, companies, and organizations also to blame?

Understanding AI and Discrimination

In order to identify who is culpable for AI discrimination, it's important to first understand what it is. AI discrimination occurs when a system exhibits bias against a certain group of people based on specific characteristics like race, gender, or socioeconomic status. Machine learning algorithms are designed to identify patterns and use those patterns to make predictions and decisions. However, if these algorithms are based on biased data sets, the output of the AI system can perpetuate that bias.

What is AI Discrimination?

AI discrimination is not a new phenomenon. Historically, AI systems have been known to discriminate against marginalized communities. For example, dermatology AI may not perform as well on darker skin tones or facial recognition technology has difficulty recognizing faces of people of certain races. The bias can also be subtler, such as certain job ads being shown only to specific genders or races. Such discrimination perpetuates systemic inequities and exacerbate social injustices.

How AI Systems Learn Bias

AI systems can learn bias from the data that they are provided with. For instance, if an AI system is trained with data that is biased against a particular group, the system will perpetuate that bias and produce biased results. Recurrent neural networks (RNNs) are particularly vulnerable to learning discrimination because they use historical data that may reflect past biases, rather than current realities. Moreover, AI systems could pick up embedded bias in the historical data such as negative stereotypes and unconscious bias.

It is important to note that AI systems do not inherently discriminate. Rather, the bias is a result of the data that is fed into the system. Therefore, it is crucial to ensure that the data sets used to train AI systems are diverse and representative of all groups in society.

Real-World Examples of AI Discrimination

Real-world examples of AI discrimination are not hard to find. One of the well-known cases of AI discrimination occurred in 2018 when Amazon's AI-powered recruitment tool was found to be biased against women. The AI system was programmed to learn from resumes submitted to the company over a 10-year period. As a result of deep-seated sexism in the technology industry, the resumes from male applicants had a higher chance of being shortlisted than the resumes from female applicants. As such, the AI system learned to perpetuate the discrimination against women.

Another example of AI discrimination is the use of predictive policing. Predictive policing algorithms use historical crime data to predict where future crimes are likely to occur. However, this data is often biased against communities of color, who are disproportionately targeted by law enforcement. As a result, predictive policing algorithms perpetuate the discrimination against these communities by directing police resources to their neighborhoods, leading to increased surveillance and harassment.

AI systems can also be used to perpetuate discrimination in the workplace. For example, if an AI system is used to evauate job candidates, it may inadvertently discriminate against certain groups based on their resumes or social media profiles. This can have serious consequences for the diversity and inclusivity of the workplace.

Overall, it is clear that AI discrimination is a serious issue that must be addressed. It is crucial that AI systems are designed and trained in a way that is fair and unbiased, in order to prevent the perpetuation of systemic inequities and social injustices.

The Role of Developers and Programmers

Developers and programmers are the primary creators of AI systems. As such, they must be held responsible for eliminating AI discrimination. They play a crucial role in ensuring that AI algorithms are designed to be fair and unbiased.

Designing AI Systems with Fairness in Mind

Developers and programmers must be vigilant about designing algorithms that do not lead to systemic discrimination and should apply explainable AI and data transparency principles. They should build diverse datasets and monitor the outputs for signs of discrimination. For example, an algorithm used in loan assessment needs to avoid disparate impacts while evaluating the creditworthiness of individuals i.e. avoid it excessively penalizing people on the basis of race or religion or gender.

Addressing Unconscious Bias in AI Development

Unconscious bias can easily permeate AI development processes, leading to human assumptions that become embedded in the algorithms developed by coders. Developers must be aware of their implicit biases and take steps to mitigate their impact on AI systems. One way to address these issues is to engage individuals from diverse backgrounds in AI development projects. Not only would they bring their diverse perspective but also flag instances of possible bias.

The Importance of Diverse Development Teams

The development team is responsible for building and training AI systems. A diverse development team can help identify potential biases and ensure that AI systems are designed fairly. Diverse perspectives and lived experiences of the diverse team members can contribute to the development of blind spots and biases.

The Responsibility of Companies and Organizations

Companies and organizations that use AI systems must ensure they are free of bias and discrimination. They must be transparent in their AI development, and procurement processes to ensure that AI systems conform to acceptable standards.

Implementing Ethical AI Policies

Companies and organizations should implement ethical AI policies that require AI systems to be free of discrimination. The policies should apply to all individuals and groups, regardless of race, gender, religion, or sexual orientation. They must have clear guidelines for responsible data management, and privacy protection. Ethical policies will also ensure companies promote transparency and accountability regarding the use of AI systems.

Monitoring AI Systems for Discrimination

Organizations should continually monitor their AI systems for indications of discrimination. They should use metrics to identify discriminatory impact. Such monitoring would require integrating AI into diverse workflows, making it easier for organizations to extract relevant data to inform decisions. They need to respond to the challenges of AI fairness through a combination of alert systems, human oversight, and review of AI outputs by experts that will ensure that AI does not discriminate against groups of people.

Addressing Discrimination in AI-driven Decisions

Organizations must also ensure that the decisions made by AI systems are transparent and auditable. If a decision made by an AI system is disputed, there should be a mechanism in place to review the decision. Similarly, organizations must instill transparency into their decisions and their reasons for implementing AI systems.

Legal and Regulatory Frameworks

Legal and regulatory frameworks must catch up with the technical and ethical challenges posed by AI discrimination. There is no one-size-fits-all approach to combatting AI discrimination across different industries, so legal frameworks will have to be agile to adapt to all scenarios.

Current Laws, and Regulations on AI Discrimination

At the moment, there are few laws and regulations that explicitly address AI discrimination. There are some basic principles such as the General Data Protection Regulation (GDPR) which aims to protect privacy and ownership of personal data, and the European Commission’s AI Ethical Guidelines that require AI systems to respect fundamental rights. However, there is a need for more comprehensive and specific regulations that will promote the fair and ethical use of AI across different areas, especially how algorithms in AI work.

The Need for New Legislation

Experts suggest that new legislation should be introduced that focuses on AI systems anti-discrimination regulations. The laws should take into consideration the diversity of groups and the intersectionality of identity traits, and consider different types of AI applications. The legislation could be aimed at ensuring that any intentional or unintentional discrimination that arises when AI systems are deployed is prevented or addressed as soon as possible.

International Efforts to Combat AI Discrimination

Various international organizations such as the OECD, UNESCO, and EU have recognized the need to combat AI discrimination. The OECD has called on governments to adopt standards on data disclosure to protect individuals against hidden discrimination in AI-based systems. This is but one of the many efforts aimed at creating an international regulatory framework that will improve AI’s reliability and ensure fairness and ethical use across the globe.

Conclusion

The growth of AI systems has implications for the future of work, economics, and human rights. However, AI must be fair and unbiased to ensure that the benefits of AI are enjoyed by all. Developers and programmers must lead the charge in designing and deploying AI algorithms that do not perpetuate systemic discrimination and bias, but companies and organizations have a critical role to play in ensuring their AI systems are free from discrimination and that they use them ethically and transparently. Governments and international organizations must work together to establish appropriate legal frameworks to ensure fair and ethical use of AI. Only a collaborative effort will guarantee that when AI systems discriminate, swift action is taken, and accountability for discriminatory AI rests on all those responsible.