The Illusion Of Inclusion: How AI-Driven Risk Management Can Truly Work For India
By Richard Apostolik Artificial intelligence is top of mind within global financial institutions. Firms are integrating the transformative technology to create operational efficiencies as well as to empower critical functions that rely on large data sets, such as credit assessment, investment analysis, portfolio management, and fraud detection. For financial institutions such as banks, investment firms, insurance companies, digital lending platforms, as well as non-bank financial companies and others, the potential use cases for AI in risk management offer significant benefits. In credit risk management, for example, AI models can process large sets of historical and alternative data to help lenders make informed credit decisions more quickly, as well as generate synthetic data to help train credit scoring models. AI tools can also detect anomalies in real-time transactions, supporting quicker trading decisions or identifying unusual spending or other patterns to enhance fraud detection and prevention. They can flag operational inefficiencies and cybersecurity threats, enhancing the overall resilience of a financial institution. In India, where a robust digital economic future is being forged, the use of AI in risk management holds great potential as a tool to assist in ensuring its financial resilience. However, like all evolving technologies, the use of AI comes with risks, including, among other things, potential bias in AI model output or wrongful interpretation of AI-generated results. The Inclusion Challenge: AI Model Bias Bias is a particularly important issue to guard against. Many AI models are trained on historical data. If this data contains biases, these biases can be perpetuated in the model output. For example, if a traditional credit allocation system has a history of wrongfully denying loans to a segment of the population identifiable by gender, educational qualification, or ethnicity, AI models trained on such data may learn and continue to repeat the same decision patterns irrespective of the lending institution’s best intentions. Bias in AI models cannot simply be dismissed as a technical issue given its possible far-reaching economic consequences, which can affect large sections of the population, or result in an underserved region losing access to crucial financial services, exacerbating socio-economic disparities. The risk is perpetuating sub-optimal credit allocations, which may be problematic from both a social and a financial perspective. Because of India’s diverse cultures, multifaceted society, regional language differences and varied socio-economic backgrounds, preventing model bias is particularly challenging. It is critically important, given these facts, to ensure impartial AI deployment, that it uses rich data sets that reflect the country’s diverse demographics to avoid unintended bias or errors in AI models. A void in data relevant to certain geographies or underrepresented communities can also produce inaccurate loan repayment predictions. In digital lending, for example, an individual working in the informal sector in a tier-2 city with stable income and no history of defaults may be denied a loan by automated or AI-driven credit decision models that rely on historical data on applicants from metro cities, certain job types or income levels. Similarly, smaller businesses may not be approved for credit because of a model’s in-built preferences for larger, more established firms. Closing the Gaps: Considerations and the Way Forward These examples reflect how biases in AI models can result in suboptimal societal decisions, leading to potential inequities in access to financial resources. AI is now poised to become a part of not just financial services but healthcare, education, and governance. The importance of addressing these challenges and rectifying them is imperative to ensure the fair and responsible use of AI across India’s diverse economy. It is also important to invest in comprehensive and bias-free datasets to help ensure the model results are fair and explainable. This can be supplemented by implementing robust model governance procedures and policies and making them an essential part of the prudent and responsible deployment of AI tools in risk assessment and management. Raising awareness about the issues associated with using AI applications in banking, finance, and risk management, and building firm-wide capacity across organisations is critical. Promoting knowledge at all levels within the organisation’s structure, but particularly among decision-makers, about the range of AI tools and techniques and their applications, and the associated risks and risk factors, ethical considerations, and accepted industry frameworks and standards for data and model governance is essential. Organisations should emphasise hiring trained risk professionals, o
By Richard Apostolik
Artificial intelligence is top of mind within global financial institutions. Firms are integrating the transformative technology to create operational efficiencies as well as to empower critical functions that rely on large data sets, such as credit assessment, investment analysis, portfolio management, and fraud detection.
For financial institutions such as banks, investment firms, insurance companies, digital lending platforms, as well as non-bank financial companies and others, the potential use cases for AI in risk management offer significant benefits.
In credit risk management, for example, AI models can process large sets of historical and alternative data to help lenders make informed credit decisions more quickly, as well as generate synthetic data to help train credit scoring models.
AI tools can also detect anomalies in real-time transactions, supporting quicker trading decisions or identifying unusual spending or other patterns to enhance fraud detection and prevention. They can flag operational inefficiencies and cybersecurity threats, enhancing the overall resilience of a financial institution.
In India, where a robust digital economic future is being forged, the use of AI in risk management holds great potential as a tool to assist in ensuring its financial resilience.
However, like all evolving technologies, the use of AI comes with risks, including, among other things, potential bias in AI model output or wrongful interpretation of AI-generated results.
The Inclusion Challenge: AI Model Bias
Bias is a particularly important issue to guard against. Many AI models are trained on historical data. If this data contains biases, these biases can be perpetuated in the model output. For example, if a traditional credit allocation system has a history of wrongfully denying loans to a segment of the population identifiable by gender, educational qualification, or ethnicity, AI models trained on such data may learn and continue to repeat the same decision patterns irrespective of the lending institution’s best intentions.
Bias in AI models cannot simply be dismissed as a technical issue given its possible far-reaching economic consequences, which can affect large sections of the population, or result in an underserved region losing access to crucial financial services, exacerbating socio-economic disparities. The risk is perpetuating sub-optimal credit allocations, which may be problematic from both a social and a financial perspective.
Because of India’s diverse cultures, multifaceted society, regional language differences and varied socio-economic backgrounds, preventing model bias is particularly challenging. It is critically important, given these facts, to ensure impartial AI deployment, that it uses rich data sets that reflect the country’s diverse demographics to avoid unintended bias or errors in AI models.
A void in data relevant to certain geographies or underrepresented communities can also produce inaccurate loan repayment predictions. In digital lending, for example, an individual working in the informal sector in a tier-2 city with stable income and no history of defaults may be denied a loan by automated or AI-driven credit decision models that rely on historical data on applicants from metro cities, certain job types or income levels. Similarly, smaller businesses may not be approved for credit because of a model’s in-built preferences for larger, more established firms.
Closing the Gaps: Considerations and the Way Forward
These examples reflect how biases in AI models can result in suboptimal societal decisions, leading to potential inequities in access to financial resources. AI is now poised to become a part of not just financial services but healthcare, education, and governance. The importance of addressing these challenges and rectifying them is imperative to ensure the fair and responsible use of AI across India’s diverse economy.
It is also important to invest in comprehensive and bias-free datasets to help ensure the model results are fair and explainable. This can be supplemented by implementing robust model governance procedures and policies and making them an essential part of the prudent and responsible deployment of AI tools in risk assessment and management.
Raising awareness about the issues associated with using AI applications in banking, finance, and risk management, and building firm-wide capacity across organisations is critical. Promoting knowledge at all levels within the organisation’s structure, but particularly among decision-makers, about the range of AI tools and techniques and their applications, and the associated risks and risk factors, ethical considerations, and accepted industry frameworks and standards for data and model governance is essential.
Organisations should emphasise hiring trained risk professionals, or training individuals to equip them with the knowledge of AI tools and techniques, their applications, and understanding of the risks, opportunities and governance considerations associated with using AI. They will play a critical role in any organisation committed to implementing safe, trustworthy AI-driven solutions.
Investing in upskilling and training, such as through risk and AI certification programmes, raises awareness and supports knowledge building within organisations across all industry sectors.
Additionally, developing relevant regulatory frameworks with provisions for detecting and mitigating model biases will ensure fair, equitable, and responsible deployment of AI that serves all segments of Indian society.
AI will be important to India’s success in unlocking its full potential, and ensuring the use of this technology is a catalyst for equitable growth across the country.
[The author is CEO at Global Association of Risk Professionals (GARP)]
Disclaimer: The opinions, beliefs, and views expressed by the various authors and forum participants on this website are personal and do not reflect the opinions, beliefs, and views of ABP Network Pvt. Ltd.
What's Your Reaction?
