With the hype around artificial intelligence and despite its high-profile failures and misuses, it’s clear the global financial sector has plenty of interest in AI, especially in hopes of finding yet another way to outsource low-complexity work. Many large financial firms have set up special departments to explore and lead the introduction of the technology in its various product offerings.
“We’re clearly making our way through a hype cycle in AI,” said Lana Khoury, a managing director at KPMG whose title lists her topics as “CIO Advisory, Data, Advanced Analytics and AI”.
Speaking at a recent industry event in Luxembourg, she described the “peak of greatest expectations” following the introduction of ChatGPT in November 2022. “You think it’s pretty much the best thing since sliced bread, it’s going to solve world peace, it’s going to do everything magically for you.”
Data trust
It has become clear that AI is only as good as the data it is based upon. “Do you trust your data? And if the answer is no, how can you then trust your AI?” asked Ulf Herbig, director of data & innovation at Kneip, speaking at a panel discussing AI during the private assets conference of the Association of the Luxembourg Fund Industry, Alfi.
“In a regulated market like private capital or financial services, this could have really quite catastrophic results.”
Shilpa Bandarkar, Linklaters
After the peak, there is some disillusionment “when you realise it isn’t a magic wand, you do need data governance, you need clean data which to apply to this AI,” said Shilpa Bandarkar, who heads the client tech and AI team at Linklaters, based in London.
She referred to two infamous AI chatbot incidents, one at airline Air Canada, whose chatbot gave incorrect ticket pricing information to a prospective passenger. The other case involved the mayor of New York seeking to help small businesses but produced a chatbot that suggested illegal actions.
Shaky guardrails
“Not really the best amount of guardrails,” said Bandarkar. “So it had really inaccurate and quite worrying advice that it was giving out. In a regulated market like private capital or financial services, this could have really quite catastrophic results.”
The main incentive for early AI adoption seems to be labour saving. Consulting group McKinsey has produced an estimate of 200 to 340 billion dollars of “additional annual value by generative AI in banking and capital markets,” said Claudia Hauser of Microsoft Financial Services in Zurich.
Kneip’s Herbig gave the example of colleagues who once produced meeting notes by hand and then spent time typing them up for distribution. Now, he said, it’s as simple as showing Gen AI a picture of your handwriting, if the AI can make it out.
AI transformation
AI productivity through empowering and enriching clients and employees, reinventing client engagement with hyper-personalised experiences “for better client loyalty trust”, reshaping business processes and bending “the curve of innovation,” are the four “AI transformation pillars” at Microsoft, outlined Hauser.
“You can utilise AI to make sense out of documents, PDFs, where data is buried, in order to normalize and structure it, to bring it in a format that you can persist and utilize all the use cases.”
Ulf Herbig, Kneip
When used in private markets, one of the strongest use cases of AI is dealing with unstructured data, explained Herbig. “You can utilise AI to make sense out of documents, PDFs, where data is buried, in order to normalize and structure it, to bring it in a format that you can persist and utilize all the use cases.”
This is best done, according to Bandarkar, with the “extractive” type of AI, “little AI” as opposed to the more popularised “generative AI”, which burst onto the scene in November 2022 with the arrival of ChatGPT.
Machine learning
“There’s still a massive place for machine learning,” she explained. “If you have a ring-fenced set of documents, you know exactly what you’re looking for. You want to extract it, you want to organise it in a particular way.”
Helping companies figure out AI is a focus for Luxembourg’s Institute of Science and Technology (List), the largest public research center in Luxembourg.
Many financial sector firms try to launch AI projects, but “they typically fail because of very simple reasons: that people don’t have the right data or that the people don’t have the right competence,” explained Francesco Ferrero of the List center.
Discrimination in data
AIs depend on what are called Large Language Models (LLMs). “We know that LLMs are biased because they have been trained using the data that are available on the web,” he explained. “These data are created by humans and these data are full of bias.”
He pointed out that Luxembourg’s new AI Act forbids companies from using a biased AI system.
While many speculate about risks from artificial intelligence for example overcoming humanity, as in the Terminator series of science fiction films, he pointed out that the “very concrete issues such as bias in the generative AI LLMs are quite subtle, but they are there right now.”