AI — and specifically generative AI — took centre stage in 2023 as tools like OpenAI’s ChatGPT and Google’s Gemini brought the technology to the masses. This enthusiasm has permeated the corporate landscape, with businesses eager to harness AI to enhance their competitive edge.
According to Accenture’s Technology Vision 2023 report, almost all of the global executives surveyed (98%) agree that AI foundation models will play an important role in their organisations’ strategies in the next three to five years. Companies in the financial services industry (FSI) are among those that are dipping their toes into the AI water.
Mitigating generative AI risks before they arise
AI has been integrated into the financial industry for decades and is now commonplace in areas such as risk, fraud, and compliance. However, the advancements in generative AI bring about amplified advantages alongside risks that financial institutions must contemplate within the confines of a heavily regulated industry. FSI companies also need to ensure that their AI outputs are free from inherent biases, ethical concerns, and AI hallucination — the generation of content that is not accurate, factual, or reflective of the real world. This could lead to:
- Misleading financial planning advice: In financial advisory services, hallucinated information may result in misleading advice, leading to unexpected risks or missed opportunities.
- Incorrect risk assessments for lending: Inaccurate risk profiles may lead to poor risk assessments for loan applicants, which can cause a financial institution to approve a loan with a higher risk of default than the firm would normally accept.
- Sensitive information in generated text: When generating text, models may inadvertently include sensitive information from the training data. Adversaries can craft input prompts to coax the model into generating outputs that expose confidential details present in the training corpus.
Within the Asia-Pacific (APAC) region, some countries are taking the lead in improving existing technology frameworks, and issuing guidelines and principles to promote AI fairness, ethics, accountability, and transparency.
For example, Singapore launched the AI Verify Foundation in June 2023, which works to augment AI testing capabilities and encourage the responsible use of AI. Singapore’s Infocomm Media Development Authority worked with this foundation to propose a generative AI governance framework. Announced in January this year, this framework seeks to set up a systematic and balanced approach for generative AI concerns while facilitating innovation.
On a company level, financial services organisations can employ different strategies to mitigate generative AI hallucination and reduce the risk of producing inaccurate or misleading information.
One approach, Retrieval Augmented Generation (RAG), enhances the output of large language models (LLM), fundamental to generative AI, by incorporating mechanisms that enable these models to access and integrate information from predefined and authorised knowledge bases beyond their original training datasets. This process equips LLMs with access to current and validated information, leading to generative AI outputs that are more relevant and accurate.
An example of this would be a financial services company that is building a chatbot to handle customer inquiries. Giving the chatbot access to secure CRM data, such as previous service conversations and product purchase history, lets the chatbot better tailor conversations with customers and follow up on existing service requests.
Meanwhile, Vector Search offers a method to enhance the implementation of RAG architecture. It operates by searching for items through their vectors, which are numerical representations of the items’ features or attributes. This method focuses on comparing the similarity between vectors instead of evaluating each item individually.
Enhancing AI with a modern developer data platform
To make the most of AI while minimising risks, financial companies need to focus on having the right data setup. They should use data platforms that include document cloud databases and services. These platforms help ensure companies stay competitive and have accurate and timely data.
Unlike traditional databases that store data in structures like tables or graphs, document databases are more flexible and can handle different types of data structures, formats, or sources which include rich objects, tables, or the nodes and edges of a graph.
With a document database, developers don’t need to pre-define the tables, names, or fields within the document, giving them the flexibility to modify the structure at any time.
This flexibility makes it easier for financial companies to adapt to changes in data or application features and integrate with AI tools without needing major adjustments to the data setup. For example, companies can easily and quickly add new information fields to a document database, while other database types, like tabular databases, require developers to alter the tables.
The inflexibility of legacy systems poses a significant pain point, hindering the seamless incorporation of cutting-edge technologies. Creating an operational data store with a flexible document model enables financial institutions to efficiently handle large volumes of data in real time, pulling in the latest and most accurate data. This is important for staying flexible and adapting to changes in the economy, customer preferences, and regulations. With AI models trained on fresh data from flexible document databases, banks, insurers, and fintech companies can manage risks better and take advantage of new opportunities, staying ahead of the curve in the fast-paced world of finance.
All generative AI applications require the use of both operational and vector data. By integrating these data types into a single platform, organisations can unlock a range of benefits. Unlike traditional setups that separate vector data from operational data, the unified platform eliminates the drawbacks of increased costs, heightened risks, and delays in retrieving critical information. The consolidation of operational and vector data on a single platform empowers developers by providing them with the tools to easily and efficiently build generative AI applications.
A modern developer data platform also comes with built-in security controls that protect all sensitive financial data while mitigating the risk of unauthorised access from external parties. Whether managed in a customer environment or through a fully managed cloud service, security features such as authentication (single sign-on and multi-factor authentication), role-based access controls, and comprehensive data encryption act as a safeguard.
ChatGPT represents a watershed moment for AI. It’s impossible to overstate how much this technology will change the way we live and conduct business. Organisations can leverage the opportunities presented by generative AI through balancing careful planning with speedy innovation. Data accuracy, timeliness, and relevance will be crucial for this. As FSI organisations already sit on a data goldmine, they have the best chances to turbocharge their business with this technology.
