What AI and Machine Learning Is… and What It Isn’t

Broadly speaking, artificial intelligence (AI) and machine learning (ML) are technologies that use computers to perform tasks that would typically require human intelligence. AI/ML capabilities have seen some adoption in the financial services industry, but most currently available solutions are focused on automating manual “stare and compare” tasks. While this is useful, it is just scratching the surface. AI/ML technology is advancing every day toward performing more complex tasks in the mortgage lending process and is on the cusp of transforming the industry.

Forward-thinking lenders are preparing now for a new era in AI/ML technology that will increase efficiencies across the entire loan life cycle. Today’s deterministic, rules-driven mortgage software will still play an important role, but it is much different from AI/ML. Understanding the differences can help lenders plan for the data, infrastructure and talent required to successfully execute an AI/ML strategy. The first step in this journey is to understand what AI/ML is… and what it is not.

Artificial Intelligence and Machine Learning Defined

Artificial intelligence and machine learning allow computers to analyze data, recognize patterns and learn from non-deterministic models to mimic human intelligence and perform tasks. When exploring AI/ML capabilities, mortgage lenders need to be sufficiently conversant in the language of AI/ML so that they can identify opportunities and manage risks appropriately. AI/ML is a general category that includes many disciplines:

• Artificial Intelligence, also called machine intelligence, is a computer’s ability to perform a given task that is otherwise accomplished using human intelligence.
• Machine Learning involves the creation of algorithms and methods that can “learn” or get better when provided with more data.
• Natural Language Processing (NLP) allows machines to read and use human languages. Document classification based on the words identified on a page is an example of NLP.
• Deep Learning is a type of machine learning algorithm that uses multiple layers of model elements to derive a solution.
• Predictive Analytics is a broad category for application of techniques including data mining, modeling and machine learning in business to seek patterns in potentially large sets of data to make predictions about future or unknown events.

AI/ML vs. Traditional Software

Traditional software technologies, including business rules management (BRM) systems, are deterministic. Inputs are directly traceable to precise and expected outputs, and no matter how complex it might become, traditional software’s conditional logic can always be tested. A set of properly constructed tests will confirm the validity of the results delivered by the code.

Artificial intelligence and machine learning technologies are different. AI/ML can address business problems that do not yield deterministic results. Inputs are often chaotic with a range of possible outcomes, requiring a probabilistic approach. AI/ML algorithms are organized into a “model” to answer a specific question based on the inputs they are given. In today’s emerging AI capabilities, each model typically focuses on completing one task.

For example, take the application of AI/ML to classify loan documents using Black Knight’s AIVA platform. In this case, the question assigned to the model is “Given a document and a list of types, what type should be assigned to the document?” AIVA receives a package containing uncategorized documents and uses optical character recognition (OCR) to determine the text that each document contains. “She” then looks at the contents of each page using Natural Language Processing (NLP) to try to find patterns that can be used to classify it.

In this example, the predicted document type is not certain, so you can only measure how often the model gets the right answer. This requires human effort to train and refine the model. As a result, AI/ML and related systems require new methods and processes to manage record keeping, software delivery and risk associated with the delivery of new models, retraining updates and other model changes.

AI/ML Depends on Data Science

Artificial intelligence and machine learning systems are technologically complex and are often driven by advanced math and sophisticated algorithms. This focus on math and statistics is another difference between AI/ML and traditional software development. Where traditional software is engineered, AI/ML is largely the province of data science. Data scientists must understand big data sets, be adept at data mining and advanced storage technologies, and be able to explain what the data is telling them about the problem being solved.

Because AI/ML is powered by data, financial institutions are realizing that having quality data is crucial to training their models and gaining more accurate AI performance.  They also understand that AI/ML is not a set-and-forget technology. AI/ML systems need to be constantly monitored and retrained by data scientists using the latest data to produce the most accurate results.

AI/ML Is Not Automation

One of the most common misconceptions about artificial intelligence and machine learning is that it is synonymous with automation. While rules-based automation can significantly enhance efficiency, AI/ML systems take it far beyond simple execution of repetitive, labor-intensive tasks. AI/ML systems consume vast amounts of data to learn and adapt as they go – mimicking human logic and decision-making.

AI/ML Is Not Autonomous

Artificial intelligence and machine learning is not magic that will solve every business problem and deliver every digital experience without human involvement. Machines can’t do all your tasks without any human oversight. Contrary to what AI alarmists are saying, AI/ML is not remotely close to human intelligence today. It can only do what it is instructed to do within a framework of a specific business problem or question. In essence, AI/ML is often highly sophisticated educated guessing. While AI can greatly increase productivity and enhance the mortgage lending experience, humans will always need to monitor, train and validate the models – as well as handle exceptions.

Conclusion

Artificial intelligence and machine learning are increasingly being leveraged to solve practical problems in the mortgage industry and are a new technology domain for financial institutions. AI/ML as a general category includes many disciplines. Technology and business managers alike should familiarize themselves with AI/ML. Whether procured as part of a vendor solution or developed internally, AI/ML is different, and managers should be aware that those differences may place new demands on them, while also offering great benefits.

To learn more about AI/ML technology, and what you should consider before implementing it in your lending business, download your free copy of Black Knight’s white paper.