Boosting AI Trustworthiness in Financial Services

By Dr. Jans Aasman, CEO of Franz, Inc.

This article was published in GlobalFintechSeries.

As a cognitive scientist with decades of immersion in various AI technologies, including speech and natural language understanding, machine-based learning, and rule-based decision-making, I’ve seen the ebbs and flows of AI development. Amidst this journey, the advent of ChatGPT and other generative AI large language models (LLMs) marks a pivotal advancement in AI’s capabilities, merging various AI methodologies into a cohesive system that leverages their strengths and mitigates weaknesses.

This integration—encompassing machine learning (statistical AI), neural networks (neuro AI), symbolic logic (symbolic AI), and generative AI (LLMs)—enables the creation of complex problem-solving systems that are not only efficient in learning with limited data but also expand AI’s utility across diverse tasks. Crucially, this blend ensures decisions are both interpretable and explainable, essential for advancing AI.

For the financial sector, the convergence of these AI technologies holds particular promise for enhancing the transparency and trustworthiness of AI-driven decisions. Transparent, explainable AI systems allow financial institutions to gain confidence in AI-generated insights, laying the groundwork for more reliable and comprehensible AI applications in finance.

The Significance of Knowledge Graphs in Finance

Knowledge Graphs, crucial to Neuro-Symbolic AI, have evolved through generations, from rule-based logic systems known for consistency yet challenged by complexity, to incorporating machine learning and graph neural networks capable of revealing patterns but often opaque and biased. The latest generation, with LLMs and Retrieval-Augmented Generation, revolutionizes inference-making within graphs, simplifying ontology creation and rule formulation. However, it also introduces challenges, especially skepticism towards LLM inference reliability, necessitating rigorous verification.

Read More: Turning to AI to Address Financial Crime, Regulatory Compliance

Financial Use Cases: Enhancing Knowledge Graphs and Machine Learning with LLMs

In the financial industry, LLMs play a crucial role in building, curating, and querying Knowledge Graphs, essential for organizing vast amounts of financial data, regulatory requirements, and customer information. They assist in creating taxonomies and ETL processes, curating data to ensure accuracy, and generating new symbolic logic rules, for instance, for risk assessment models or compliance guidelines.

Furthermore, LLMs bridge the gap between the predictive accuracy of machine learning applications, like customer behavior prediction or fraud detection, and the need for explainable insights. By analyzing extensive financial databases and literature, LLMs provide detailed rationales for predictions, supporting decision-making processes and regulatory compliance efforts. They also facilitate the generation of feature vectors and the development of machine learning models, potentially automating aspects of financial analysis and risk management.

Preventing LLM “Hallucinations” in Financial Analysis

In the financial domain, inaccuracies or “hallucinations” in AI-generated content can lead to significant repercussions. Knowledge Graphs can address this by interacting directly with LLMs for data validation and verification, ensuring the reliability of financial predictions, market analyses, or investment recommendations. For example, cross-verifying LLM-generated financial data against trusted databases or using real-time market data to confirm the accuracy of LLM outputs minimizes risks and enhances trust in AI-driven financial services.

Neuro-Symbolic AI’s Role in Advancing Trustworthy Financial AI Applications

The integration of diverse AI technologies into a Neuro-Symbolic AI framework represents a transformative approach to overcoming the limitations of individual AI methods, introducing a level of interpretability and reliability crucial for financial services. This collaborative AI model paves the way for advanced, trustworthy AI applications in finance, aligning technical sophistication with the nuanced demands of financial decision-making and regulatory compliance.

In conclusion, the amalgamation of AI technologies under the Neuro-Symbolic AI umbrella offers the financial sector an unprecedented opportunity to enhance the accuracy, transparency, and trustworthiness of AI-driven services, thereby reshaping the future of financial technology and services.