Published On March 12, 2021Organizations are struggling to cope with an influx of information that is surging 30% per year. The task is too big for humans to handle. Fortunately, the artificial intelligence disciplines of machine learning and deep learning are rapidly maturing to become full-blown content service platforms.
Organizations are struggling to cope with an influx of information that is surging 30% per year. The task is too big for humans to handle. Fortunately, the artificial intelligence disciplines of machine learning and deep learning are rapidly maturing to become full-blown content service platforms.
Machine learning models enable computers to sort through vast amounts of data while growing ever more intelligent about what they see. While computers have long been useful for processing data in well-defined structured formats, machine learning can help organizations unlock the vast amount of information that doesn’t fit into those neat molds. Such unstructured data is expected to comprise 80% of the information that organizations gather in the future.
AI services like image recognition, text recognition, automatic translation and machine transcription are now so common that they can be rented from cloud service providers with a credit card. However, some customization is required to adapt them to specific use cases.
Fortunately, the tools to do that are also proliferating, making AI a viable way to mine insight from troves of previously inaccessible data. Let’s look at one application that can provide immediate value in the auto insurance industry.
Auto insurers must make numerous judgment calls when assessing a claim. Who was at fault in the accident? What is a reasonable estimate of repair costs? What were the road conditions when the accident occurred? Were injuries sustained? Is there any evidence that the claim could be fraudulent?
Thanks to the ubiquity of cell phone cameras, many of these judgments can now be made by examining photographs of an accident but that still requires a trained adjuster. Machine learning can bridge that gap. Using image recognition, which is a type of deep learning, computers can identify the make and model of the vehicles involved, license plate numbers, road conditions, the nature of the damage and even the identity of the operators from a photograph. It can even consult historical repair records to estimate costs.
AI can also do something humans can’t: scan years of photos of past auto accidents to determine if the same image has been seen before. That capability is particularly valuable to insurers because fraudsters often use the same photo to document multiple claims, sometimes years apart.
For organizations seeking to get a handle on their data assets, machine learning can be a valuable tool in achieving the promise of information governance. Cataloging all of an organization’s data is an expensive, monotonous and labor-intensive effort. Few are willing to devote the resources needed to audit the entire corpus of content, label it and make decisions about its disposition.
This task is tailor-made for machines, though. Numerous commodity cloud services are available that can scan and recognize images in printed and even handwritten documents, turning unstructured information into digital text and images that machines can understand. Content can then be searched, categorized, integrated into automated workflows, and even used to kick off programmatic processes.
Financial services firms can use this capability, for example, to convert large volumes of images stored in the binary TIFF image format into PDF documents that search engines can index. Working initially under the supervision of human operators, computers can scan these now-readable assets to identify keywords or phrases and classify content with increasing degrees of precision. The more they churn through the training data, the more capable they become and the less supervision they require.
Machine learning is neither a silver bullet nor a black box. Models are only as good as the data that is given to them and performance can degrade over time. Platform providers should enable real-time performance monitoring and have rollback plans in place in case decision quality begins to falter. Organizations should also maintain access to the data they use to train the model for troubleshooting purposes.
Before experimenting with AI, machine learning and content services platforms, you should assess your internal resources for their expertise in training and monitoring and seek expert help where needed. Iron Mountain InSight® is an example of a content service that makes it easy for customers to build and manage their own custom AI models.