Financial institutions are missing a valuable opportunity if they don’t harness the benefits that artificial intelligence (AI) can bring to data management. From accelerating and automating routine processes to mining value from huge data sets, established and generative AI have the potential to transform the way financial institutions use and organise their data, says JP Morgan Chase legal chief data officer Jennifer Ippoliti.
“If you can bring those resources to bear on traditional data management you can achieve huge increases in efficiency,” Ippoliti tells Data Management Insight. “It’s potentially game changing.”
Ippoliti will share her views on AI and financial data during the opening practitioner keynote fireside chat at A-Team Group’s Data Management Summit NYC on 28 September 2023. She will be joined by a host of senior data chiefs and end users for our annual state-of-the-industry gathering, which this year will be held at @Ease in midtown Manhattan.
While different forms of AI are already being incorporated into institutions’ data processes, Ippoliti believes the role of broader AI applications in data management will widen as the technology is refined and as organisations become better acquainted with its capabilities.
She identifies two broad categories of AI; the ‘traditional’ form that is being used now to automate workflows, accelerate routine mapping and data validation processes; and next-generation generative AI, which is built on large language models (LLMs).
LLMs have come to public attention thanks largely to media coverage of ChatGPT. While it can demonstrably put together succinct and convincing content, it has been criticised for generating incorrect information known as hallucinations. Nevertheless, Ippoliti believes generative AI has the potential to radically improve the data management process.
“LLMs are coming,” she says. “There may be issues now with hallucinations and not citing sources, but I am confident those things will be addressed.”
Presently, LLMs show great promise in general activities, Ippoliti says. They are particularly useful for summarising large tracts of text in, for instance, company reports. They can also respond to plain English questions about datasets or texts, such as requests to retrieve specific pieces of information. Further, Ippoliti says LLMs are great at classifying huge datasets according to different requested criteria.
These use cases can all be applied to data management, she adds. “There are multiple kinds of ways that you could apply something like this in the data governance space. You can give the LLM a policy and it can tell you whether other content is compliant with the policy, for instance.”
LLMs are also a powerful tool to assist in anomaly detection for data quality and can be used by non-data scientists to write code for data management applications. “The way they can speed up tedious day-to-day review and search is really powerful,” Ippoliti says.
A Cautious Approach
Of course, many AI applications are in their infancy and will take time to mature and LLM technology is still prone to potentially problematic glitches. JP Morgan Chase, for instance, doesn’t yet allow its employees to use ChatGPT on the bank’s systems.
Ippoliti has personal experience of hallucinations when she asked ChatGPT to find biographical information about herself. It came back with a detailed list of achievements within the arts, having mistaken her for several artists who share her name. Hallucinations spring from limitations in the underlying datasets the models are trained on, which while massive, are insufficient to generate a correct answer every time. This also opens up the possibility of bias, as under-represented communities, languages, and images are most vulnerable to hallucinations.
These problems will be overcome in time, Ippoliti believes, saying she sees a future where traditional AI and generative AI will be broadly available to data management practitioners.
“These technologies will become a lot more common place and we should be thinking now about how we’re going to use them so that data management functions don’t get left behind when the rest of the company starts investing in them,” she says.
At next week’s Data Management Summit NYC, Jennifer Ippoliti will be talking to data management veteran and member of the EDM Council Peter Serenita during the summit’s opening practitioner keynote fireside chat, The potential of generative AI to transform data management.
Subscribe to our newsletter