The success of the release of ChatGPT in December 2022 put artificial intelligence (AI) at the forefront of boardroom discussions across many companies.
The ability to generate human-like responses to almost any query with remarkable confidence has captivated many.
‘AI’ is an umbrella term that encompasses a wide range of technology, including machine learning, natural language processing, computer vision, robotics and more. A few recent breakthroughs have enabled the emergence and adoption of tools such as ChatGPT.
The cost of deploying these models has fallen
There have been technical innovations, like the transformer architecture introduced in 2017 that allows AI to analyse entire sentences at once, improving its understanding of context. At the same time, improved hardware (such as GPUs or TPUs) has made it feasible to process much larger datasets more quickly.
Also, self-supervised learning has enabled models to teach themselves from unorganised text, making it easier to grasp complex language patterns.
From a practical perspective, AI tools like ChatGPT are now more user friendly and accessible to the general public, allowing non-technical users to benefit from them. Additionally, the cost of deploying these models has fallen, making AI technology more affordable and practical.
Trust in responses
But can we trust the accuracy of responses generated by AI? That depends mainly on how they’re being used.
In their current form, large language models (LLMs) such as ChatGPT, when used as stand-alone chat tools, can ‘hallucinate’, or generate false or misleading information. This occurs because LLMs generate text based on patterns from large datasets, without verifying the accuracy of the content.
Around 20%–40% of unsecured consumer lending requires the manual handling of documents, contributing to unnecessary high operational costs
Their confident tone can give the false impression that incorrect information is reliable. While this remains a challenge, ongoing research aims to reduce hallucinations by refining how models are trained to identify and avoid potentially inaccurate claims.
In more focused applications, where AI is used to analyse unstructured data within a clear framework rather than generate new content, it’s easier to ensure accuracy and performance. The upshot is that handling tasks like verifying income, employment and affordability — previously done manually, requiring hours of work — is reduced to seconds.
If you provide financial services, there are several important factors to consider before using AI tools. First, remember that, if a tool is free, your data may be the product. AI models often learn from user interactions so, in regulated environments where personal information is handled, it’s crucial to ask the right questions.
LLMs work particularly well to make sense of unstructured, text-based data
Ensure that personal data is stored and managed securely, in compliance with GDPR and other regulations. This includes implementing strong security measures such as encryption and access controls.
Additionally, confirm that the data isn’t being used for training beyond its intended purpose, especially when sensitive information is involved.
Next, check whether the AI provider has the appropriate certifications, such as ISO/SOC 2 standards or ethical AI certifications, to demonstrate adherence to best practices and regulations.
Finally, consider how AI impacts customer outcomes. If the AI significantly influences decisions affecting customers, audits should be conducted to ensure fairness and detect any biases in the system.
What is a tangible example of AI used in practice today in financial services? LLMs work particularly well to make sense of unstructured, text-based data. Much of the customer information in financial services is found in customer-supplied documents, leading to significant manual work and inefficiencies.
Can we trust the accuracy of responses generated by AI? That depends mainly on how they’re being used
Take mortgages — all applications today still require the submission of bank statements, payslips and tax returns. This leads to manual review queues, countless hours of manual work per application, and fundamentally a sub-par customer experience.
Similarly, 20%–40% of unsecured consumer lending requires the manual handling of documents, contributing to unnecessary high operational costs.
AI tools can streamline this process, reducing costs and processing time by up to 75% through automated verification of income, employment, affordability and more.
Alexis Rog is founder and chief executive of Sikoia
This article featured in the November 2024 edition of Mortgage Strategy.
If you would like to subscribe to the monthly print or digital magazine, please click here.
The post Comment: Making intelligent use of AI appeared first on Mortgage Strategy.