‘Artificial intelligence is about technology and ethical considerations’
Artificial intelligence is very well suited for sifting through and analysing data. An ideal tool for the financial world. But how do you do that in a responsible and human-centred way? AI specialist Jim Stolze has clear ideas about this. ‘There is a need for extra attention to the ethical side of AI.’
‘Five years ago, I gave a TEDx talk in Amsterdam about artificial intelligence,’ Stolze recalls. ‘I emphasised then that people should not be afraid that robots will take over our tasks, but that we should be more afraid of “the robot within ourselves”. All day long, we do all sorts of mundane tasks that do not make us happy. Take something like typing numbers into spreadsheets. These spreadsheets are better than when we did accounting on paper. But just as spreadsheets freed us from paper, AI will free us from spreadsheets.’
No more time for mundane tasks
The audience reacted enthusiastically to Stolze’s prediction, but was not yet fully convinced. ‘Afterwards, I was even called a techno-optimist,’ he recalls with a laugh. ‘But now we are five years on. The way people look at AI has changed quite a bit. This is partly because we are currently facing huge staff shortages. So the less time people spend on mundane tasks at work, the better. The business case for deploying an AI application is therefore very easy to make now.’
Extra attention to the ethical side
Stolze sees that the time for experimentation is over and that more and more companies are seriously getting to grips with AI. He also sees that this new phase leads to new questions. Earlier this year, he decided to change course: he sold Aigency, a company that builds AI solutions, and has since been actively involved in the Data Inspection Service, which he co-founded. ‘We are, so to speak, the “white coats” that check organisations’ data for “bias”. We also subject their algorithms to stress tests where we look at explainability, reproducibility, and fairness. There is now a great need for this extra attention to the ethical side of AI. So, besides data engineers and a lawyer, our team also includes two ethicists and a philosopher.’
Socratic method
There also appears to be a great need for such an extra check in the financial world. For instance, the AFM, the Authority for the Financial Markets, has already approached the Data Inspection Service with a request to create a blueprint for financial service providers developing an AI application. ‘We don’t tell them what to do, but we ask questions. What algorithms do they use? On which datasets are they trained? What data do they use and what not? Who supplied that data? And what labels are used to identify and mark the data? Question by question, we delve deeper. The Socratic method, in other words. And it works well because, in this way, companies quickly find out what still needs attention.’
As the Data Inspection Service, we don’t tell companies what to do, but we ask questions. What algorithms do they use? And what data? This is how we delve deeper.
– Jim Stolze, co-founder of the Data Inspection Service.
Done in a few seconds
‘In the field of artificial intelligence, interesting things are already happening in the financial world,’ he continues. ‘For instance, it is already possible to take out home insurance via an app on your mobile phone. The company behind that app can then quickly determine the amount you need to pay based on your location and data from, for example, the housing market and the environment. Taking out the insurance is therefore done in a few seconds.’
This example involves supervised learning: the algorithm knows exactly which data is important and how to handle it. With unsupervised learning, it often becomes even more interesting. In that case, an algorithm can indicate which transactions are similar based on different parameters and thus create clusters. Are there transactions left that do not fit into any of the clusters and that the algorithm does not really know what to do with? Then, according to Stolze, that is the ‘box’ you should look at. ‘There is a good chance you will find information there that will lead to new insights. I see AI as a collaboration between humans and machines. Artificial intelligence is not something magical and not something that will take on a life of its own. I see it mainly as a powerful tool that can help people in many areas. In that respect, I am still a techno-optimist.’
I see AI as a collaboration between humans and machines. It is a powerful tool that can help people further.
– Jim Stolze, co-founder of the Data Inspection Service.
Predicting when invoices will be paid: Payt’s AI solution
Algorithms are becoming more advanced. So is the quality of the data. And while it is very difficult for people to sift through thousands of financial transactions and look for anomalies, AI applications are very good at it. So Payt decided to develop an AI application themselves. Socially responsible choices ‘When building our AI application, we started small and simple,’ emphasises Laura Baakman, software engineer at Payt. ‘And we are now expanding it step by step. We first look at how we can teach the algorithm to predict as accurately as possible when our customers’ debtors will pay their bills. Many different factors play a role in this. What industry is it? Is it a private customer or a business one? We are adding more and more parameters. There are also things that pose a challenge. Adding the postcode, for example. Is it desirable to weigh the neighbourhood where someone lives? We ourselves find it a borderline case. It is not only about what is technically possible but also how we can do this in a socially responsible way.’
The practice
Payt’s AI module has been in beta with some customers since early 2022. With each invoice, accounts receivable staff see the likelihood of an invoice being paid late and the expected payment date. With this information, the accounts receivable department can better determine their approach. For example, the notification that a debtor is likely to pay late can ensure that the customer receives a payment reminder just a bit faster than usual. In the long term, Payt will offer the possibility to proceed to automated follow-up actions based on the payment probability calculation. ‘However, we do this very carefully with attention to all aspects of AI, including the ethical ones. The beginning is there, and we look forward to making the AI module available to all our customers in the short term. One thing is certain, the possibilities are enormous.’
Publication FD.nl: 03-04-2022
