Artificial Intelligence: Understanding the implications for financial services and payments

Payments companies globally continue to investigate the use of Artificial Intelligence (AI) to improve the customer experience (for example, through chatbots) or improve backend processing (for example, risk-based fraud detection systems).

The Chinese fintech Ant Financial has been described by the MIT Technology Review as a ‘finance giant that’s secretly an AI company’. 

With increased focus on AI, we were pleased to facilitate a member information session on Artificial Intelligence: Understanding the implications for financial services and payments.

Professor Tiberio Caetano provided insight into modern artificial intelligence. Professor Caetano is Adjunct Professor at the University of Sydney and Co-Founder of Ambiata. The comments in this blog reflect his presentation.

Professor Caetano’s presentation helped shed light on some of the foundational questions around AI – what it is, its use in payments and ethical issues.

Evolution and Learning

The presentation started by drawing a parallel between the evolution of human intelligence and the way in which AI ‘learns’. Biological evolution works in four stages: Replicate, Mutate, Select and Repeat. These stages can be mapped onto the learning process of AI:

  • Replicate = Memorise
  • Mutate = Explore
  • Select = Reward
  • Repeat = Repeat

Modern AI builds on the economic revolutions that we have seen over the past two centuries. The 1800s saw automation of physical processes. The 1900s saw the creation of information – transforming understanding of physical processes into data. In the 2000s, AI machines are overseeing the transformation of information into useful information.

Machine Learning

To illustrate, Professor Caetano referred to the development of automated translators. The AI is provided with multiple translations of the same text. The machine looks at the placement of words and learns general rules about how words appear in relation to each other across different languages. In this way, it learns how to generalise rules to other texts. As the process is repeated over and over again, the AI develops an increasingly sophisticated understanding of translation and learns to understand human speech and writing in its natural form.

Relevance for Payments

This process has apparent application within payments. Potential uses of AI include:

  • Fraud prevention through detecting anomalies in transaction patterns
  • Credit risk automation, using deep learning to understand transaction patterns
  • Insurance pricing 
  • Payment routing, via reinforcement learning
  • Selecting an acquirer for transaction processing

However, there are some limitations. For example, fraud detection is hampered by limited data, meaning that human input and oversight is still required. Because AI assessments are based on learnings from past data, it is hard for machines to predict new forms of fraud if they haven’t seen it before.

Ethical Questions

Ultimately, AI is designed by humans and reflects the prejudices and views of the designer. One key way in which human designers can influence the development of their AI system is through the use of “rewards”.

Rewards provide feedback to the AI on whether it is making the right or wrong decision. The ethical question for human designers is how high a ‘cost’ should be placed on a particular incorrect decision made by the AI? A high cost will discourage the AI from making similar mistakes in future, while a low cost will not. The significance of a particular mistake can only be answered by humans, who are operating from within an ethical framework.

Finally, Professor Caetano suggested that the first step to understanding the ethics of AI is to generate awareness.

If you missed the session and would like a copy of the presentation, please get in touch.