Practical Approaches to Addressing the Challenges of Bias in AI

Practical Approaches to Addressing the Challenges of Bias in AI

By Ater Solomon Vendaga

Abstract

Innovations in technology have in recent times taken the center stage of global development and discussion. These inventions which include but not limited to Artificial Intelligence are gradually becoming part of our daily lives as there is no area of life that it has not yet permeate. But these inventions rely heavily on data to produce their intended results. Since AI relies heavily on data to produce its systems, when such data is inaccurate, unjust, or biased it will create systems that will discriminate against millions of people, especially in countries that have low or no man power to create these systems like those in Africa. This work adopting a doctrinal research methodology explored this critical angle of this evolving innovation, examining relevant examples and the implications while making valuable recommendations to subvert the hitches that AI biases may occasion. It should be noted that a partial concern is placed on the African Continent since biases primarily impact it.

Keywords: AI, Bias, Bias in AI, Data, Algorithm, Africa, Implications of AI  

• Introduction

We have become increasingly reliant on AI to help solve virtually all of our most pressing problems. From healthcare to economic challenges to crisis prediction and response, agriculture, and legal industries, AI has taken over and is ordering the pattern of the operations there, making it apt to conclude that AI is the greatest innovation in human history. However, as transformations in AI are foisted, concerned stakeholders have observed that due to the workability of AI systems, the product of constructed algorithms has inadvertently inherited many of the biases that continue to perpetuate the global challenges we hope to solve. The result is that AI is not a purely agnostic process of objective data analysis and there are tendencies of biases that may limit other persons, regions, or groups from enjoying its fruits, and if they are not addressed, it will affect the equitable application of the innovation.  This work offers a practical approach to addressing the issues of AI bias. Accordingly, the first part introduced the work whilst in the second part of the work; definitions of key terms were captured to provide a common ground for comprehensive appreciation of the work, the concept of bias in AI is treated in the subsequent parts. There, the research dwelt extensively on the various types of biases in AI, their examples, implications of biases in AI and it ended with practical approaches that can address the challenges of AI biases and from there, the work is drawn to an end.

• Definition of Terms

In order to fully appreciate the topic of this research, a comprehensive definition of the terms making up the topic will be provided.

Bias is defined as a particular tendency, trend, inclination, feeling, or opinion, especially one that is preconceived or unreasoned. It is unreasonably hostile feelings or opinions about a social group or prejudice..

AI is a constellation of technologies that enable machines to act with higher levels of intelligence and emulate human capabilities to sense, comprehend, and act.. 

Algorithms are defined as a process or set of rules to be followed. In AI, algorithms exist in the form of automated instructions.

Machine Learning (ML) is a process used by big tech companies to teach Artificial Intelligence (AI) specific patterns about the society they are to ‘serve.’Quinyx’sBerend Berendsen defines ML as a set of algorithms that are fed with structured data to complete a task..  

• Examination of Bias in AI

The concept of Bias in AI has been observed for a very long time to be one of the critical issues forming part of AI discussions. It is the underlying prejudice in data that’s used to create AI algorithms, which can ultimately result in discrimination and other social consequences. 

It refers to situations where ML-based data analytics systems show bias against certain groups of people. The biases usually reflect widespread societal biases about race, gender, biological, sex, age, and culture. So, for instance, if a system is designed with data that believe only blacks can steal, it will be difficult to detect a white rapist. 

The issue with Bias in AI is that it does not exist in a vacuum, but it is built out of algorithms devised and tweaked by ‘those same people and it tends to think the way it is taught. This is because they are trained using human records; these tools can incorporate human biases into their algorithm. 

• Types of Bias in AI

1. Algorithmic AI Bias or data bias: here, algorithms are trained using biased data. 
2. Societal AI bias: here, our societal assumptions and norms cause us to have blind spots or certain expectations in our thinking. It is often believed to have a far-reaching influence on algorithmic AI bias..  

• Selected Examples 

• A study conducted in 2019 on AI healthcare services in the US found out that a healthcare risk-prediction algorithm, when used on 200 million people, favored largely the white patients. 
• COMPAS- the acronym stands for Correctional Offender Management Profiling for Alternative Sanctions. It is an algorithm used in the US system to predict the likelihood of a defendant becoming a recidivist. It turned out that due to the bias in the model of the algorithm, the system predicted twice as many false positives for recidivism for the back offenders (45%) than the white offenders (23%).

• Implications of Bias in AI

1. Preference of a section over others: when AI systems have biased algorithms, they tend to favor a particular group of people over others. Amazon stopped using its AI recruitment tool in 2018 when it realized the program was biased against women.. 
2. Injustice: when AI systems are designed with biased algorithms, it will lead to injustice.
3. Lack of transparency and accountability: Biased AI systems will be a bad reference point for accountable and transparent projects since it reflects the system design metrics. 
4. Ineffective results: AI systems need to be built with foreign and biased data to ensure the intended result of the application of that system is achieved. It is even difficult for such systems to work optimally in regions like Africa. 

• Solutions/Recommendations

Fixing AI Bias is quite a herculean task, given the interconnections between its makers and its operation. However, there are ways to reduce the likelihood of these biases. 

1. Regulating a more ethical AI: AI regulations will set strict requirements for AI systems based on a pre-defined level of risks. For instance, in 2021, the European Commission set a significant precedent in this area by launching its first-ever legal framework on AI as well as a new coordinated Plan with member States which provides that it will “guarantee the safety and fundamental rights of people and businesses while strengthening AI uptake, investment, and innovation across the EU.” This was reciprocated by Australia too. The European Parliament’s 2020 resolution on the civil liability of AI is also helpful here.  Africa can develop the same legal approach to ensure that AI systems in Africa must comply with their uniqueness in terms of their sensibilities.. 
2. Company/Organizational Engagement: AI systems are largely applied within the operational environment of organizations. It is only fitting if they are meant to adopt and implement inventions that capture the various sensibilities of all groups and persons. 
3. Rights and Activist Groups Engagement: it is essential to note that with AI as a social machine, the relationship between tech and humans needs to be a subject of continuous definition. To ensure that human society is at all times represented, AI inventions must embrace humanity as a subject matter of concern. Here, the rights and activist groups must ensure that the digital rights of all are an epicenter of AI activities. 
4. Developers’ Role: developers can ensure AI bias-free systems by carefully ensuring that the data given to the systems used to train these machines are free from bias and accurately represent all classes or groups of people. 
5. Diversification of Bigger Units where AI is applied: diversity in AI Communities eases the identification of biases. 
6. Adopting AI bias reduction tools: some organizations have created tools to reduce AI bias. For example, AI fairness 360 deployed by IBM is an open-source library to detect and mitigate biases in unsupervised learning algorithms. The tool enables programmers to test biases in models and datasets with a comprehensive set of metrics, etc.
7. Algorithmic Audit: Researchers have argued that one of the ways to reduce Bias in AI is Algorithmic Audit. This will create mechanisms to check that the engineering processes involved in AI system creation and deployment meet declared ethical expectations and standards. 

• Conclusion

AI can turn the world into a great place where everyone will go about their businesses efficiently. However, when data used in building the systems are biased and assumptions, it can harm many people’s lives, work, and progress. We can fight against these biases by challenging the assumptions that underpin the datasets we are accustomed to and adopting fair AI policies.

Elena Fersman, Head of Ericsson’s Global AI Accelerator, sums this up in her blog post on the importance of balance in AI: 

“One of the things that fascinate me most is that humans and nature inspire AI technology. This means that whatever humans find success in their lives and evolutionary processes can be used when creating new algorithms. Diversity, inclusion, balance, and flexibility are critical here as well, concerning data and knowledge, and diverse organizations are better equipped for creating responsible algorithms. In the era of big data, let us make sure we do not discriminate the small data.”

****************************************************************************************

This work is published under the free legal awareness project of Sabi Law Foundation (www.SabiLaw.org) funded by the law firm of Bezaleel Chambers International (www.BezaleelChambers.com). The writer was not paid or charged any publishing fee. You too can support the legal awareness projects and programs of Sabi Law Foundation by donating to us. Donate here and get our unique appreciation certificate or memento.

DISCLAIMER:

This publication is not a piece of legal advice. The opinion expressed in this publication is that of the author(s) and not necessarily the opinion of our organisation, staff and partners.

PROJECTS: 

? Take short courses, get samples/precedents and learn your rights at www.SabiLaw.org

? Publish your legal articles for FREE by sending to: eve@sabilaw.org

? Receive our free Daily Law Tips & other publications via our website and social media accounts or join our free whatsapp group: Daily Law Tips Group 6

KEEP IN TOUCH:

Get updates on all the free legal awareness projects of Sabi Law (#SabiLaw) and its partners, via:

YouTube: SabiLaw

Twitter: @Sabi_Law

Facebook page: SabiLaw

Instagram: @SabiLaw.org_

WhatsApp Group: Free Daily Law Tips Group 6

Telegram Group: Free Daily Law Tips Group

Facebook group: SabiLaw

Email: lisa@sabilaw.org

Website: www.SabiLaw.org

ABOUT US & OUR PARTNERS:

This publication is the initiative of the Sabi Law Foundation (www.SabiLaw.org) funded by the law firm of Bezaleel Chambers International (www.BezaleelChambers.com). Sabi Law Foundation is a Not-For-Profit and Non-Governmental Legal Awareness Organization based in Nigeria. It is the first of its kind and has been promoting free legal awareness since 2010.

DONATION & SPONSORSHIP:

As a registered not-for-profit and non-governmental organisation, Sabi Law Foundation relies on donations and sponsorships to promote free legal awareness across Nigeria and the world. With a vast followership across the globe, your donations will assist us to increase legal awareness, improve access to justice, reduce common legal disputes and crimes in Nigeria. Make your donations to us here  or contact us for sponsorship and partnership, via: lisa@SabiLaw.org or +234 903 913 1200.

**********************************************************************************