Co-founder and CEO, Kasisto.
Financial services firms, including banks of all sizes, are piling into generative AI and exploring ways to apply this revolutionary technology to seemingly every facet of their businesses, from customer service to document processing, external marketing and more.
Through the implementation of generative AI, banks are improving customer experience and can even gain value to the tune of $200 billion to $340 billion annually, according to a report published by McKinsey. The potential of generative AI’s application within banking cannot be ignored. As Citigroup CEO Jane Fraser shared, “We must be proactive about embracing A.I. It’s an essential part of winning in the digital era.”
However, banks must also carefully evaluate what platforms they are using to bring in generative AI because privacy and security are still paramount concerns.
Public Vs. Private AI
ChatGPT took the world by storm, setting the record for the fastest-growing platform in history. Most companies have become exposed to the vast possibilities of public AI—but with those possibilities come also pitfalls.
While it is helpful and convincingly natural, public AI is limited by factual inaccuracies, invented information, biases and security concerns, making it risky and unsafe for banks to integrate into their platforms.
With regards to security concerns, in particular, public AI platforms are trained on a wide set of publicly available data—which means that the data and content that banks provide to these public platforms could be used to train models that would also be used by one of their competitors.
As regulators place further scrutiny on how generative AI is being used within regulated industries such as banking, institutions should be wary of using public AI in case of inadvertent data leakage.
With private AI, though, a bank’s enterprise data is only used for training and application within that financial institution—providing higher security, control and accuracy.
Unlike public AI, private AI can provide rich conversational capabilities derived from industry-specific data and conversations. These systems are built from private and proprietary documents, knowledge repositories and web content to ensure that responses are specific, accurate and personalized.
This approach can help to ensure that a financial institution knows the provenance of the data being used to train and tune the LLM, whereas public LLMs typically don’t disclose the data sources used to train the model and how the data is used.
Hurdles For Implementing Private AI
While there are clearly benefits for using private AI, banks still face hurdles when it comes to the implementation within their organizations.
For smaller banks, the cost of private AI implementation can be substantial and, when compared to bigger banks, many smaller banks do not have the budget to integrate this technology into their business model. Leaders will have to weigh the cost against the necessity to maintain a competitive advantage in the industry.
Also, like all things in the banking world, private AI also comes with some red tape. Regulators have expressed concerns about AI in the banking industry due to cybersecurity risks, the handling of private information and concerns about AI bias leading to discriminatory lending practices.
This scrutiny has caused many leaders to shy away from taking on some of the regulatory burdens associated with AI. That’s why it’s crucial, when adopting AI, to keep responsibility and governance in mind and to stay on top of any regulatory policies.
Conclusion
For banks seeking to harness the potential of generative AI, they should consider the potential risks.
Private AI solutions—particularly bank-specific LLMs—provide an alternative to giving LLMs access to sensitive data, but banks will need to choose their partners wisely and understand the risks to be able to benefit from the full potential of generative AI.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
Read the full article here
