Edward Achtner leads HSBC’s Office of Applied Artificial Intelligence. Late last year, he sat down to discuss ChatGPT and the future that generative AI and large language models (LLMs) may have in financial services.
Q: Although AI has been around for decades, 2023 was the year that its possibility, use, experimentation, projects, etc. exploded. Many point to generative AI, LLMs, services like ChatGPT, as the reason. Do you agree?
A: I agree with both points. Perhaps unsurprisingly, AI has been with us for decades, clearly. But it is only up until really the last year, in fact [when ChatGPT was launched on November 30, 2022], that [there is] accessibility for everyone around the world. It’s simply because it’s made a transition from the domain of data scientists and machine learning specialists, into the forefront — it’s literally at your fingertips.
That is why it is so different, because of its reach, its accessibility — it has transformed.
Q: What is it about generative AI that presents more possibilities? What impact do you think it will have on financial services?
A: Let’s kind of take a step back and differentiate between ChatGPT and generative artificial intelligence and large language models. Generative artificial intelligence is something that is demonstrably different from traditional or classical artificial intelligence. I’d say, within financial services and banking, we’re cautiously optimistic, with a strong emphasis on caution, that it can be applied in practical, responsible, and ethical means to generate commercial value both for our customers and equally our shareholders.
The reason for that is fairly simple. If you think about it, if you and I were to ask ChatGPT how to cook your favourite recipe, it might get it really close, but at the end of the day a little bit of extra pepper or salt isn’t really going to make or break it. But if you’re dealing with a financial process, if you’re dealing with the level of precision and certainty that you need around something from a payment, to lending, to making a risk-based decision, you need to have a level of precision and certainty about how you apply the technology. It is going to be very different in the context of regulated industries.
Q: Where does financial services look for inspiration?
A: One thing we’re seeing is inspiration come directly from not only our customers, but our teammates. Our teammates around the world, in their personal lives, are testing and learning. Then it’s about how you can then safely and responsibly bring that experimentation into the bank. So, we have been doing a significant amount of proofs of concept and testing and learning.
That really covers a range of areas where we think it’s likely that generative artificial intelligence can benefit us safely and responsibly. Those areas [include] how we look at operational efficiency. I’ll give you some examples with that [such as] improving the customer experience — which, in some respects, will allow us to enhance our risk and compliance technology. That’s because of how we measure risk and how we evaluate a particular decision. Generative artificial intelligence may actually give us higher-quality, better data from which to make decisions.
Again, it’s early yet, but we actually do think there is going to be a path forward through the medium term to do this.
Q: Can you talk about services or operations where the bank is using AI right now?
A: If you look broadly, and I want to emphasise this is not just generative [AI], but broadly, across all of our global businesses and functions, we have approximately 1000 applications, using some form of AI. Now, that goes back, as we talked about at the beginning of the conversation, many, many years. In fact, our earliest machine learning models go back nearly a decade. Now, within that population, there are a handful of generative AI proofs of concept that we’re presently testing and learning that will have a high likelihood to scale across our operations, but we’re not in production.
I want to emphasise that, when it comes to generative artificial intelligence, we’re not in production. We’re testing and learning because, again, it is a different type of technology and a different type of control framework.
We’re cautiously optimistic around trying those things out.
Q: LLMs use a lot of data — how does a bank manage what type of data, and whether they are allowed to use that data, to generate a useful generative AI service?
A: From a banking perspective, it’s absolutely imperative that we bring forward models that are free from bias, have the right weightings, and have the right parameters to give those predictable results. That’s largely going to be dependent on us curating that model with known, trusted data.
In some cases, that may mean that a number of these public large language models may not work in the context of banking practices. That’s a journey I think the industry is going to be on.
Q: People get very excited when emerging tech gets accelerated — how does this change behaviour at innovation and tech teams inside banks?
A: That’s right. This is not new within the last 12 months, but over the last 20 and 30 years in particular, there has been a series of different innovations that have been buzzy. It’s very important that within the organisation we have teams who are responsible for responsible innovation. Then we have teams where it becomes apparent that this innovation has become viable, that it’s likely to be built in a way that will meet the expectations of our customers and our regulators. It’s very important that we put forward products that our customers are happy with.
It then kind of goes through a much more rigorous due diligence of product market fit — will it serve customer needs, will they benefit from it, will they find delight in it, and ultimately does it create some kind of commercial value for our shareholders in our business? We don’t want to just innovate for the sake of innovation, it has to actually create value.
I think with AI and generative AI, again, we’re cautiously optimistic. It actually can achieve those things. But, we have to stress, it’s still very early days with this.
