Artificial Intelligence (AI) is the ability for a computer system to perform tasks that are more human-like, such as reasoning, generalizing, learning, image recognition, and motion. Common AI features with these characteristics include unlocking your phone with facial recognition, using Siri or Alexa on your smartphone, chess bots, and ChatGPT. Over the last several years, there has been an increasing focus on businesses using AI to enhance their products and services. AI’s growth in every conceivable direction has motivated companies to stay ahead of their industry peers, hoping to leverage the newest AI capability for a competitive advantage in the marketplace.
However, businesses have been using AI for decades to provide value for their customers and increase internal business efficiencies. Siri and Face Unlock technology were released in 2011. IBM’s Deep Blue chess bot defeated Garry Kasparov in a game of chess in 1997. Specifically in the financial services industry, AI dates further back with AI systems created in the 1980s to help with credit card transactions. Even the term, artificial intelligence, was introduced at a computer science conference held at Dartmouth College in 1956, which further established artificial intelligence as a scientific field. The reason for the ‘AI hype’ may not only be around the possibilities we think AI could achieve, but rather the scale, speed, and accuracy of these AI systems and how quickly AI products shift from business use to personal consumption use.
To understand this idea, we can review early AI systems used in the financial services industry. An AI model known as the Rules-Based Expert System, was based on knowledge, logic, and explanation systems coding. These models operated on a set of ‘if-then’ rules written by industry experts and coded to automate repetitive or time-consuming tasks. The incoming input data was evaluated by the set rules, with the output reflecting the expert human insight coded into the system. In 1988, American Express released the Authorizer’s Assistant, which was a Rules-Based Expert System used for assisting with and automating the approval of credit card transactions. The first version of Authorizer’s Assistant applied 890 rules to thousands of transactions that occurred every hour, with nonstop operation. This meant that a transaction’s cost, time, location, and frequency data was evaluated on an ‘if-this-then-this’ basis. For example, when you are using your credit card out of town, but your transaction mirrors your spending patterns, the AI assistant could automatically approve the transaction. However, if the transaction has too many abnormalities, like being an unusually high amount at a strange time for an uncommon item, the AI assistant can flag the transaction and escalate it to human authorizers.
Prior to this automation, there was zero real-time authorization, and stores would use a manual carbon-copy imprinter for batch-processing transactions. A process that could take minutes for a single transaction at a retail store could now be completed in seconds. The Authorizer’s Assistant led to reduced errors and almost instant approvals for transactions, enhanced fraud detection and reduction capabilities, and minimizing future staffing requirements through automating the credit transaction approval process.
As technology advanced, so did the forms of artificial intelligence. Machine learning AI systems replaced the ‘if-then’ AI models. Instead of following the traditional rules-based system, machine learning allows computers to make statistical inferences from data and learn patterns from this data to improve their accuracy over time.