While virtually all delegates asked this question at a recent talk by NatWest staff, at the ACT Annual Conference on the topic of Unleashing AI in Treasury, there was nuance in exactly how far the technology can, and should, go in making a footprint in the workplace.
Will entire functions within financial services be replaced by generative AI and automation? Or will more mundane forms of task easing complement people’s roles and free people up to become more creative?
Nick Pedersen, Head of Digital, at NatWest, hosted the discussion, along with two colleagues: Jon Horrocks, Head of FX Options Quantitative Analytics; and Tom Underwood, Head of Machine Learning Quantitative Analytics.
In their roles, they regularly see customers experiment with real-world tools, as well as discuss the areas where AI can be deployed without presenting undue risk. And the answer, for now at least, appears to revolve around a slow roll-out of tools that will take the grunt work out of repetitive tasks.
Tom explained: “Generative AI’s ability to take documents, such as pdf forms, and extract the information within them can be quite transformative. That really unlocks things.”
Nick agreed, adding that this process can also standardise the array of documents that are traditionally informationally messy – PDFs in particular.
One of the questions many businesses have, however, is whether roll out of this type of tech, dependent on large language models that can be bought off the shelf, require a huge number of data scientists to tame the beast.
Tom said: “I don’t think many organisations will need to train their own large language model. There are ones off the shelf that people can make use of, and the advantage is they can follow instructions: meaning they can make use of AI-related tools without an army of data scientists.”
A question raised in the discussion was the extent to which regulation will allow the roll-out of generative AI in customer-facing tools, such as those that summarise reports and may even make trade suggestions or offer advice to end customers – especially in financial services. With concerns over the veracity and reliability of the outputs they provide, at present many such tools involve a ‘human in the loop’.
Jon explained: “I don’t see dramatic transformation in the very short term with AI tools in financial services… Is there a risk to your business of not rushing to adopting them? Possibly, but if you follow closely what’s going on, then you can be in a position to catch up as the tools develop.”
With concerns over data hallucinations and how AI may be used by cyber criminals, it is easy to over-egg the risks at the cost of the more likely rewards of deploying AI. Yes, the intentions of bad actors and ethical risks are of concern, but the more likely scenario is more optimistic: uptake of AI will ease data collection, enhance document input and reduce manual, repetitive tasks.
“Models are not magic”, said Jon. “There’s potentially huge benefits but you must be aware of the dangers. Don’t let the human nature of them put you off from questioning them.”
To read more of our coverage at the ACT Annual Conference, click here
