Explainable AI and blockchain solutions for equitable financial access

Research at the University of Liverpool is helping develop explainable AI (XAI) and blockchain systems to address discriminatory biases against lower wealth households trying to secure bank loans.

Background

Households with lower wealth often face difficulties in securing housing loans or funding to start small businesses, as these require significant collateral from conventional lending sources, such as retail banks.

Individuals facing challenges in securing loans can seek guidance by consulting an advisor or automated AI systems to evaluate their chances of success before approaching lending institutions.

However, underserved individuals are frequently subjected to discriminatory biases influenced by their gender, ethnicity and societal position. The extent of this inequitable treatment is further magnified by their notable underrepresentation in AI training datasets, highlighting a significant gap in the system's inclusivity and fairness.

Our research

Dr Swati Sachan's research integrates blockchain technology with XAI algorithms - used to describe an AI model, its expected impact and potential biases - to develop a trustworthy AI advisory and transparent decision-making system to address three key challenges. 

1. Decentralised data sharing mechanism

In this research, a blockchain-based decentralised data-sharing approach is designed to enable collaboration among various financial experts by different financial institutions. This approach enhances data-sharing incentives by addressing concerns associated with personal identity protection and centralised data access control.

The collaborative approach enriches the data for developing an automated decision support system and expands the collective knowledge involved in processing loans for underserved communities.

The first phase of this decentralised data-sharing technique has already been tested and the findings and extended work have been shared at an Institute of Electrical and Electronics Engineers (IEEE) Blockchain conference.

2. Explainable AI for high-stake decisions

Most high-performance AI algorithms (eg deep neural networks) are inherently black-box in nature and lack transparency in decisions. Stakeholders in financial institutions prioritise explainability in AI decisions for critical situations involving human lives or significant assets, despite the availability of high-performance AI algorithms.

In this research, two explainable AI algorithms - hierarchical belief-rule-base (HBRB) and maximum likelihood evidential reasoning (MAKER) - are developed to provide transparent decisions. These methods are published in journal papers and the refinement of these algorithms is under development.

3. Defence against adversarial attack

Centralised AI solutions are vulnerable to adversarial attacks, wherein malicious actors manipulate the algorithm's environment. This manipulation could involve adjusting the model's parameters and inserting distorted or misleading data samples.

Such assaults can lead to incorrect decisions that not only result in considerable financial losses, but also inflict substantial damage on the lives of underserved individuals and can potentially lead to serious legal issues.

This research employs blockchain technology to automate the auditing of AI algorithms. It leverages the immutability characteristic of blockchain, which enables prompt detection of tampering or attacks on data and AI systems.

The initial test results are under review in a journal and further improvement of this technique is under development.

The impact

1. Mitigating bias and enhancing transparency in financial decision-making

Operated in partnership with RVA Works, the Open Trellis Program - a financial intermediary and consulting resource, that connects small business owners with lenders and offers pro bono guidance, helping improve loan accessibility and mitigate default risk - is a notable example of impactful knowledge exchange.

“Dr Swati Sachan has been advancing research into the combination of technologies that advance enterprise development. This work in knowledge sharing is bridging the gap between researchers and those commercializing the technology.” Open Trellis CEO and Co-Founder, Dale Fickett (read full Forbes article)

Their efforts in credit risk optimisation, blockchain and XAI are helping to expand the accessibility of small-business loans via unbiased and transparent lending decisions. Over 250 entrepreneurs who fall under median income and members of racial minorities, including African Americans and Latino Americans, have been able to secure loans so far.

Dr Sachan has also developed an AI-enabled auditing methodology to detect biased decisions by finance professionals and leakage judgment errors in AI algorithms. The new auditing technique has been validated through the annotation of test results by four underwriters at specialist property lender, Together Financial Services (research still under review).

Her work has been referenced in the Tech and the Future of UK Foreign Policy inquiry, so it's likely that the existing research and forthcoming contributions will continue to impact the governance and application of AI in finance.

2. Commercialisation of blockchain and AI in FinTech and LawTech

The research highlights the commercial potential for compelling applications in the Finance and Law sectors, as integrated XAI/blockchain systems can be used to design innovative products and services for a broader demographic. 

As well as public praise from Open Trellis CEO acknowledging the commercialization of the technology, the real-world implications of Dr Sachan's work have received prestigious recognition by the Forbes Business Council, which described "blockchain enhancements and XAI" as the number one 'hot' trend for small businesses, due to its potential to provide them with a competitive edge.

At local level, a case study included in the Whitecap Consulting's Liverpool City Region FinTech Report 2023 has helped increase awareness about Swaiti's research.

The case study explains focuses on the development of a legal reasoning system that integrates XAI and Blockchain technology to help democratise access to legal services for a wider demographic. Its deployment can offer more affordable services due to cost reductions and enhanced efficiency achieved through integration of AI in lawyers decision-making systems, facilitating greater access to legal assistance for those who cannot afford it.

What’s next

The future impact of AI decision-making systems and system's efficiency can be assessed by: 

• Monitoring the number of minority beneficiaries
• Comparing the duration of loan approval across different demographic groups
• Analysing loan approval rates to identify potential discrepancies or biases.

In addition, a qualitative perspective can be gathered by collecting user testimonials or case studies on how the XAI/blockchain systems have enhanced loan accessibility.

Dr Sachan plans to secure knowledge transfer partnerships (KTPs) and other research grants to achieve these impacts.