Machine learning and AI - the future of finance

With the rapid increase in the amount of data available, there is huge potential for Machine Learning and AI to transform the finance industry. Some investment firms are using Machine Learning algorithms to trade in high frequency based on recent financial news, while others use to it to make portfolio recommendations based on user preferences¹. It can help in portfolio diversification by separating stocks with different market return characteristics and can even be further developed into predicting investor behaviour and sentiment about stocks based on Twitter data. With open banking giving access to more data, companies may be able to improve credit risk measures and ratings². In the long term, stock price prediction may even have a place for these technologies.

Companies that have invested in building their data infrastructure, technology and computer science capabilities may more effectively harness the full potential of Machine Learning and AI in their business model, and translate that into added value. As a result, these companies may be better positioned to achieve greater long term advantages through efficiencies, greater scalability, new customer insights and tailored approaches based on their data capabilities. 

What is Machine Learning and AI?

Machine Learning and AI are often most often imagined as part of a futuristic sci-fi society where robots have completely replaced humans. We see glimpses of this through the increasing ‘smartness’ of devices that somehow know what we want to do before we have even thought of it, something that would have been inconceivable a few years ago.

So what exactly is Machine Learning and AI? And what does it mean for the future for finance?

Artificial Intelligence

The definition of AI is constantly evolving as humans and technology continue to evolve. According to Andrew Moore, former Dean of Computer Science at Carnegie Mellon University and now head of Google Cloud AI, “Artificial intelligence is the science and engineering of making computers behave in ways that, until recently, we thought required human intelligence.”³

AI is seen as a continual endeavour to develop machine intelligence, a concept which is a moving target. As AI technologies advance and become commonplace, they often are no longer considered to be AI⁴. Based on some definitions, calculators could have been considered AI when they were initially produced⁵. Now, our definitions are far more advanced. Our understanding of human intelligence (sensing and understanding our environment, performing actions and learning from these actions) is forming the framework for AI⁶. Essentially, these traditionally human functions are now thought of as computer systems that can think and act for themselves.

AI is made up of a collection of technologies to enable machines to act with human levels of intelligence. These include Machine Learning, Robotics, Natural Language Processing (NLP), Image Recognition (Computer Vision) and Speech Recognition. A lot of fields of research are classified under AI, however it becomes more of a philosophical question as to whether the results achieved through these methods are truly ‘intelligent’.

Machine Learning

Machine Learning is a subset of AI and aims to contribute to a machine’s intelligence primarily through learning and performing actions. Similar to humans, the field of Machine Learning uses large amounts of data to learn and form patterns, then uses this information to make decisions, predictions and classifications.

Historically with the invention of computers, capabilities with data exploded in managing complex calculations and storing copious amounts of information. However, computers would only work if humans told them what to do. They would be run to perform complex calculations and produce an answer. But what if we wanted computers to provide us some analysis on its own?

Machine Learning essentially consists of computer models that are trained on information, and make predictions and inferences on new information, based on what it has learned. This is much like humans studying and learning content for an exam, and then applying the learned knowledge to questions never seen before in the actual exam, with the aim to get as many correct as possible. A simple and familiar example of a machine learning model is a linear regression, where sample data points are what the model trains on, and the line of best fit is the result the model aims to optimise.

This is a huge step forward to traditional applications of computing because now computers can solve real world problems such as assisting with medical imaging in cancer patients or predicting when severe weather events might occur.

Why has this become so popular?

A Google Trends comparison of the terms ‘Machine Learning’ vs ‘Artificial Intelligence’ highlights the worldwide growth in interest of these fields since early 2014. It is also evident that Artificial Intelligence has been around for a lot longer than Machine Learning.

When the technology was first invented, computing was extremely slow, expensive and cumbersome compared to today. Recent developments in smaller, faster and more efficient chips as well as more powerful graphics cards and higher computing speeds have all contributed to technology being more widely accessible and more commonly used. And while, Machine Learning and AI are not new concepts (they were written about a lot in research papers in the mid-1900s) they were just theories at the time. There was no real way to implement such powerful computations.

Now with fairly powerful computers being so widely available and internet sharing data across the world almost instantaneously, many people have the ability to run more complex computations.

Machine Learning and AI are powerful ways in which we can derive meaning and insights from the masses of data collected and use this to create tools. The popularity increase has resulted from the increase in usage and prevalence within all industries now.

In the finance industry, in particular, machine learning models have become increasingly important to create insights from big data and being able to predict and classify one of the more complex datasets, human behaviour. Natural Language Processing has been able to read millions of analyst reports much quicker than humans and deep learning models have been developed to identify supplier relations of companies in order to understand key business risks of stocks.

This technology already is revolutionising worldwide industries. There is no doubt that implementing AI technologies with the proper business context can help shape the future of the industry.

How we are using AI technologies

Currently Westpac’s own RED AI Chatbot harnesses the capabilities of IBM Watson NLP and aims to help improve customer service with 24/7 access, whilst reducing operational costs⁶. Within BT’s Investment team (BTIS), we are experimenting with various clustering techniques (K-means, DBSCAN, Gaussian Mixture Model) as well as visualisation through dimensionality reduction tools (Principal Component Analysis, t-distributed Stochastic Neighbour Embedding), to better understand our customer behaviour in providing for their financial wellbeing.

Ultimately, Machine Learning and AI should be seen as tools that humans can harness to achieve better outcomes for our clients and customers.

1. https://edition.cnn.com/2019/02/17/investing/artificial-intelligence-investors-machine-learning/index.html 
2. https://www.westpac.com.au/news/in-depth/2019/01/ai-meets-open-banking-credit-2-0-coming-soon/
3. https://www.forbes.com/sites/peterhigh/2017/10/30/carnegie-mellon-dean-of-computer-science-on-the-future-of-ai/
4. https://ai100.stanford.edu/2016-report/section-i-what-artificial-intelligence/defining-ai
5. http://www.cs.bath.ac.uk/~jjb/web/whatisai.html (no longer online)
6. https://www.afr.com/business/banking-and-finance/westpac-turns-on-ai-chatbot-as-cost-pressures-mount-20190507-p51ksr

Next: A privately issued digital currency could one day deserve 'reserve' status