Big banks test the waters of generative AI, but the risks mean they won't dive headfirst

Imagine that in the future, your financial manager may actually be proficient in artificial intelligence (AI) using big language model technology, and it can choose the one that suits you best from tens of thousands of options. As more and more well-known banks test the waters of generative AI, this prospect is not out of reach.

The US financial media CNBC recently pointed out that generative AI has a wide range of application prospects in remote customer service, investment decision-making and even criminal investigation. But many banks also recognize that the technology is still in its early stages, with problems with data collection, privacy, and more. "Risk means they don't jump headfirst."

"Explosive innovations"

After the advent of ChatGPT, executives at the world's largest banks and online financial companies raved about generative AI, calling it an "explosive innovation." This has also led the industry to test the waters in this field of technology.

According to consulting firm Evident, JPMorgan Chase & Co. hired 3,651 AI-related positions worldwide from February to April this year. Aigen Technologies, which provides artificial intelligence technology to companies such as Goldman Sachs, said that inquiries from banks increased fivefold in the first quarter of 2023 compared to the same period last year.

According to Global Data, an international consulting firm, by 2024, retail banks' investment in artificial intelligence platforms will increase by 21.8% compared to 2019.

In fact, the use of AI by financial institutions is nothing new. So, what exactly is the difference that generative AI will bring? Take the online intelligent customer service that customers are most familiar with as an example.

A person in charge of the Fintech Research Institute of the head office of a joint-stock bank explained that the traditional bank intelligent customer service is "doing judgment and choice questions", using natural language understanding to identify problems, and then selecting the standard answers that need to be given. Generative AI, on the other hand, is "doing essay questions", breaking the operation mode of traditional dialogue robots to exhaust users' dialogue intentions, and can produce extremely rich dialogue content and response range, saving a lot of manual operations.

According to a research report on the website of the National Bureau of Economic Research (NBER), a non-profit research organization, generative AI conversation assistants can increase the number of customer service personnel solving problems per hour by 14% on average.

Multiple application scenarios

Acting as a customer service is only one of the application scenarios of generative AI in the banking industry. According to CNBC, the industry is currently scrambling to find a "place to use" generative AI.

The collection of customer data is an important area of application. Institutions such as ABN AMRO are using the technology to improve internal business code and analyze customer behavior. According to startup Taktile COO Gusco, generative AI could be a key tool to reduce a company's operating expenses and improve efficiency: "It enables a kind of real automation that enables lenders to identify the right customers without having to consult dozens of PDF documents to get the right information." ”

Risk modeling is another area of application. David Walker, chief technology officer of Westpac Group in Australia, said internal experimental results showed that generative artificial intelligence not only increased the productivity of software engineers by 46%, but also did not affect the quality of program code.

When it comes to investment decisions, generative AI is also expected to serve customers in the future. Bross, a director at Field Fisher and a former derivatives trader, said banks are using AI to "come up with more tailored hedging solutions through tools such as interest rate swaps and equity derivatives, enabling them to provide better pricing to their clients."

Generative AI can also be used for crime detection, identifying patterns of behavior that may be fraudulent or money laundering. Maria Tejada, a partner at Bain & Company, a global consultancy, said generative AI could enable financial institutions to capture and analyze large amounts of structured data, such as spreadsheets, as well as unstructured data (such as legal contracts and phone records), thus playing a "game-changer" role in risk management.

Currently acting as a "co-pilot"

While recognizing that generative AI could bring "real change" to the banking industry, many in the industry also pointed out that the technology is still in its early stages and that it may be too risky to apply the technology in areas involving consumers.

A key point is that advanced AI systems need to process large amounts of data – this involves sensitive customer information, as well as a large number of legal systems. Arava, a senior analyst at BBVA, said it was too "risky" to involve sensitive customer information at this stage and could invade privacy.

The accuracy of the data source is another issue. Some worry that data from things like Twitter and social networking sites like Reddit could lead to AI generating false information. Observers describe it as an AI illusion — a big-language model that tends to give authoritative-sounding answers to questions, even if it doesn't know the answers. And the responsibility of the banker lies precisely in the fact that transactions must be based on reliable information.

"For a respected bank, do you really want to repeat to your customers the same thing that AI found on Reddit?" Blackwell, a professor at the University of Cambridge, questioned.

There's also the reality that, at least for now, generative AI is expensive to develop and run. Lewis, founder and CEO of Aigen Technologies, said that answering a question using large language models can cost as much as $14 per question. Meanwhile, it costs a human professional $6 to answer a question. This is because processing complex financial documents requires significant cloud computing costs.

CNBC notes that for now, generative AI still acts as a "co-pilot" in banking — more as a digital assistant than as a core part of its services. For example, some companies still use humans to check the accuracy of AI responses to questions; In the field of crime investigation, AI mainly assists employees to find data and enrich cases, rather than replacing the role of investigators.

Carlo Geovie, a partner at McKinsey & Company, said banks need to identify areas where AI can really help, and work with senior executives to develop a roadmap to train staff and hire more experts. They also need to redesign their risk frameworks to address IP concerns, an uncertain regulatory environment, and the risks of AI illusions.

(Edit email: [email protected])

Column editor-in-chief: Yang Liqun Text editor: Yang Liqun Title picture source: Tuworm Photo editor: Cao Liyuan

Source: Author: Lin Zilu Zhang Quan