The use of artificial intelligence (AI) to optimize business processes across the financial services industry is becoming commonplace in 2018. Every day, new organizations announce how AI is revolutionizing the industry with disruptive results. As more and more business decisions are based on AI and advanced data analytics it is critical to provide transparency to the inner workings within that technology.
According to a recent McKinsey Global Institute analysis, the financial services sector is a leading adopter of AI and has the most ambitious AI investment plans. In a related article by the Harvard Business Review, adoption will center on AI technologies like neural-based machine learning and natural language processing because those are the technologies that are beginning to mature and prove their value. Below, we explore a challenge and opportunity that is unique to the rapid adoption of machine learning.
Challenge and Opportunity
The challenge: The transparency around the inner workings of what many call the algorithmic black box is very low. The decision-making processes of machine learning based technologies are difficult to decipher because machine learning algorithms don’t provide reasoning behind the results. This is concerning because many machine based decisions have wide-ranging compliance and business development implications.
The opportunity: This transparency challenge is not much different from working with people. For instance, we don’t always know why a relationship manager or trader decides to advise a client to act? Leveraging AI, however, creates data that can be audited and visualized to expose patterns, lineage, and correlation that easily describe what we currently don’t have answers to. Financial traders can now have an opportunity to look at the data generated by the decision-making process and can use that as a tool to learn from.
Sentiment Analysis Models Are Not All the Same
Human intelligence and thinking are persistent. Whether one is reading, customizing products for a diverse customer base, pricing a security, managing a portfolio, etc. – decision making is done in the context of history and currently available information. For example, the importance of contextual understanding is key. When reading, the meaning of each word is dependent on your interpretation of the previous word, understanding of an earlier paragraph, memories of past experiences, and perception of current events. This is not complex for you because you do it all the time. However, it is an undertaking for computers to focus on the intent of a single word especially as the gap widens between it and relevant phrases. Sentiment analysis of books, news articles, social media posts, chat exchanges, and voice conversations require contextual understanding.
Machine learning is one type of computational process used to implement AI. Deep learning leverages neural networks, which is a type of machine learning that attempts to replicate the neural networks of the human brain. As neural networks are applied to everyday cognitive scenarios, new models are created to keep up with the complexity of the problems being solved. An example of machine learning model evolution is as follows: ML > Neural Network (NN) > Recurring Neural Network (RNN) > Long Short Term Memory (LSTM) networks > Bi-Directional LSTM, etc.
Which model works best and how can visual analytics help? The heat map visualization below compares three types of machine learning models. Note that the blue band associated with the word “hate” is more distinct for the “Bi-Directional LSTM” model than for the others. This illustrates that this model is able to maintain a sharp focus on the important keyword.
Source: See Figure 5 of “Visualizing and Understanding Neural Models in NLP” by Jiwei Li, Xinlei Chen, Eduard Hovy, and Dan Jurafsky
Going Forward: The Customer Journey
More and more business decisions will be based on advanced data analytics and it is critical to provide transparency to the inner workings. Real-time decision making correlated with historic results will be based on massive amounts of data from disparate sources and in various formats. This requires a visual analytics platform that is able to show real-time and historical data in one application, analyze all types of data in place, and allow the subject matter experts to do this in a self-service fashion. See this example of correlating both streaming and historic data in the same visual analytics application and the underlying architecture need to support it.
We described how visual analytics can be used to evaluate the success of internal models, but AI is being applied to the customer experience through robo-advisors and chat bots. Considering all the AI/customer touch points, wouldn’t it be valuable to visualize evidenced-based decisions that shed light on your customers’ intentions, needs, and wants?
This article originally appeared on ValueWalk.