What is the deal with AI?

Automation is often mentioned as one of the major trends not only in finance but everywhere and it is thrown around as if it is something new. Manual tasks that have a specific process - event A happens before event B etc - have been automated for quite some time and what we should be talking about is the second or third wave of automation made possible by leaps in technology and availability of data. Highly intelligent technologies that can optimise the order of events and also learn by analysing patterns or results from previous actions will forever change society as we know it. But if we look specifically to finance, the capability to analyse large amounts of data is already transforming the industry. 

Artificial Intelligence (AI) is going from buzzword to actual implementation and is being used in everything from making investment decisions, optimising trading in SORs, algos and matching engines, finding the shortest route through networks, to on-boarding of clients and testing infrastructure.

When talking about AI, we mean the broad science of mimicking human abilities. What we have observed so far is more related to a component of AI called Machine Learning (ML), which is a method of data analysis that automates analytical model building. It is based on the idea that systems can learn from data, identify patterns and make decisions without using explicit instructions. Instead ML is relying on statistical models and algorithms to detect patterns and draw inferences. The iterative aspect is key because as models are exposed to new data, they expect to be able to adapt based on what they have learned from previous computations. The purpose, of course, is to produce reliable decisions and results that will either save time/cost or produce profit. 

The science itself is not new, but have gained momentum as data generation and computing power have dramatically increased the past decade. The financial industry is all about data, both static and dynamic, so it is not too difficult to see the practical applications of ML when it comes to quantitative models, or to reconcile large data sets available in multiple locations in multiple systems to name a few examples. One major drawback of ML is that the data need to be cleaned up and structured. Ironically, AI can be employed to reduce much of that burden.

One of the arguments for human intervention has been that for the models to work, humans need to provide context and handle exceptions. That is most likely true for some time to come, but another component of AI is gaining momentum and may change that. Natural Language Processing (NLP) is the component of AI that deals with the interpretation of human language. Roughly 80% of the data is unstructured, sometimes referred to as dark data. It is coming from emails, statements, contracts, websites, call center notes, customer feedback and so on.

There are plenty of business use cases where NLP can be leveraged on dark data to bring efficiency or to create otherwise hard to produce opportunities. Use cases can range from sentiment analysis, topic modeling, knowledge extraction, intent identification, document classification among others. The accuracy may not be at human level, but that is largely compensated for by the sheer capacity of the models. And the more data the models are exposed to, the more accurate their predictions get.

If we look specifically at Sentiment Analysis it can be illustrated by a very simple example.

Financial market analysts are interpreting data and not only financial data published by the companies. They also make their predictions based on what the CEO says when communicating with the market or other written or spoken communication issued by the company, by industry peers or related news. Sentiment analysis has become a key tool for making sense of that data. By collecting and analysing text the aim is to extract and identify opinions within the text by assigning positive and negative values to words and expressions. Looking at a really simplistic example, words like “profit”, “growth”, “benefit”, “good” are seen as positive, while words like “cost”, “loss”, “bankruptcy”, “fail”, “risk” are assigned negative values. Scanning through the news looking for these markers may give an opinion about the direction of the market in general or a specific company. In reality it is more complex and the system will shift through large amounts of data, impossible for a human to consume, let alone process and the words can add to either positive or negative sentiment depending on context.

Automation and the use of AI can be seen either as a blessing or a curse. On one hand it will allow humans to focus their efforts on less mundane tasks and free up time for innovation and humanitarian work. At the same time it will make entire industries obsolete and perish if they are not able to adapt. 

So nothing new under the sun.