What kind of issues do you think the financial services industry should be focusing on, with regards to artificial intelligence?
Artificial intelligence (AI) is a very hot topic at the moment, and there is a lot of talk about it. It encapsulates a lot of the data science and big data initiatives that firms have already been working on, but there is still a question as to what exactly it is. Are people actually using AI, or are they simply using algorithms and process automation? I think there is an interesting debate around that.
Looking at the big picture, banks are all working with legacy systems, and one of the big challenges for them is considering how far they should go in automating and improving those legacy systems, and at what point they should make the jump to a new system. These are complicated legacy infrastructures running large volumes of data, and the big prize is in figuring out how to simplify that and reduce the associated costs.
Another aspect for the large banks is the introduction of voice and image recognition systems using AI. I would question where exactly that sits. I’m not sure that our operational processes lend themselves to AI algorithms for images, and institutional clients don’t necessarily have the same application for voice recognition technologies as retail clients. It will be interesting to see how those tools can be applied to the institutional financial services industry.
Are there any AI applications ready to be applied to the back office?
There are a number of interesting AI applications that could lend themselves to the back office. One is in customer focus—looking across various systems that interface with customers in order to judge their mood. We could potentially predict when we may be likely to lose customers, based on the nature of their interactions, or we could synthesise data to alert people to a trend, which could lead to potentially cross-selling products. Being able to look across a lot of data points could be very valuable for a large bank. Financial crime is another area—pattern detection can be very effective for detecting money laundering and fraud prevention. It might not be obvious to a human being, but using machine learning we could analyse large amounts of data and identify a combination of factors that indicate a higher risk of fraud.
In the retail space, AI has been implemented through call centres and chatbots, for example, and the industry has got used to that to some extent.
Now, that technology is being combined with large data sets and AI in order to create something with a much greater level of intelligence that is far more useful.
Finally, there is the area of regulatory compliance and risk reporting. We have a complicated risk environment and we have seen several years of increasing regulation designed to reduce risk in the industry. That’s likely to be an area where machine learning is used to digest complex regulatory requirements in order to aid compliance.
Do you think there’s a risk of the industry becoming over-reliant on AI technology?
My first car was an Austin Montego, which had a black box engine management system. I was driving in the Netherlands, and the car broke down. I took it to a garage, where the mechanic opened the bonnet, looked at the black box and concluded that he couldn’t do anything for me because he didn’t have a laptop to plug in to run a diagnostic.
That’s a good analogy for the kind of issue we could run into in terms of becoming too reliant on technology. I don’t think we’re at that point yet, but it’s certainly an issue, not just for financial services but more generally as AI becomes more embedded in our lives.
When we look at improving operational efficiency in banks, one of the things we’re working on is robotic process agents (RPA). You can create a robotic agent to manage a process, and effectively lay this on top of existing IT systems in order to make them more efficient and to increase their longevity.
There is a double edge here, though, as it also fixes that IT system in its current state, making it harder to change the processes later.
Will the industry still need human interaction, no matter what happens?
This is another philosophical question that can be applied to AI generally. I see it being implemented in a role that supports human decision-making. If you look at the health and medicine industry, for example, it could be incredibly beneficial to have some form of AI to review 70,000 journal articles that reference a particular condition, helping a physician to ultimately make a diagnosis.
Similarly, in chess, we saw brute computing power defeat grand chess masters 20 years ago. But now, if you look at ‘freestyle’ chess, where players can compete in any way or in any combination they like, the most successful teams are those with a combination of human and AI players.
There are some interesting examples of AI supporting humans successfully, but the interesting debate will be around how it’s going to work.
Take up of blockchain technology in financial services has been slow. Is the industry ready for AI as well?
The two are quite different things. There are a few big items to resolve around blockchain—mainly confidentiality and scale—and there are teams working very hard to resolve those. There is a lot of momentum behind blockchain initiatives, but it’s hard to change horses. There are large amounts of payments and assets maintained on existing systems, so you have to be very careful before making any changes to those systems. Switching to blockchain would mean a completely different way of managing transactions, so I wouldn’t expect things to change very quickly. But, that’s not to say it hasn’t got huge potential.
AI is very different. The author William Gibson once said: ‘The future is already here, it’s just not very evenly distributed.’
We already have AI in a lot of aspects of our lives, whether that’s self-driving cars, Siri or prevention of credit card fraud. The challenge is applying it where the business needs it. It can sometimes feel like AI is a solution looking for a problem, and in traditional systems development, the first question is always around the user requirement. What is the user trying to do? There is a danger that, with all the excitement around AI, people are forgetting that.
We need to consider the user requirement and the options that are available. The answer may well be in AI, but it may also, be something much more mundane, such as improving the interface between two systems.
