AI 101: A Guide to Understanding Artificial Intelligence (AI) in Claims

By: Sean Eldridge, Maggi Wettstein, and Kieran Wilcox

It’s impossible in today’s day and age to not think of AI when you think of technology. AI is transformative, revolutionary, and poised to change everything. You’re also probably a bit scared that it could take all our jobs. While we don’t have the final answer to that question, we can tell you that AI is here to stay. We are at the forefront of a significant wave of change, and if you’re not prepared you risk being swept away. By the end of this post, you'll have a better grasp of why AI is different from past technologies that have promised all of these things before and how you can begin to leverage it in claims.

What exactly is AI?

At its core, AI is a tool designed to tackle problems that traditionally required human intervention; converting unstructured data like text or audio into usable, actionable data. Additionally, by design it’s capable of learning and adapting to new situations by absorbing each new piece of data presented to it.

It's equally important to understand what AI isn't. AI isn’t the best tool for every problem. Like any tool, it has its strengths and limitations. AI solutions flip the historical expectations of software on its head. Instead of expecting 100% repeatable and testable results, AI solutions are inevitably going to make mistakes and have errors.

As of today, AI isn’t fully replacing humans. Instead it’s allowing its wielders to accomplish previously unheard of levels of productivity and output. In these scenarios, often the AI plays a role as a personal assistant -- eager to solve your every problem but often requiring oversight and correction to do so.

What AI is good at and not so good at

Understanding AI's strengths and weaknesses is crucial for effectively integrating it into any workflow.

What AI Excels At:

• Processing massive amounts of data: AI can analyze lar ge datasets beyond human capacity, quickly identifying patterns and anomalies that would take a claims, service, or underwriting professional much longer to find–an attribute that’s particularly valuable in claims, where large volumes of unstructured data exist.
• Pattern recognition: From identifying fraudulent claims based on historical data to recognizing objects in images (like damaged vehicles), AI's ability to spot intricate patterns is impressive.
• Repetitive tasks and automation: AI can automate tedious, high-volume tasks, freeing up staff for more complex and nuanced work. Think about sorting emails, data entry, or responding to initial customer service inquiries.
• Personalization and recommendation: AI algorithms are adept at tailoring experiences and suggesting relevant information based on user behavior and preferences, leading to more personalized interactions for claimants and insureds.
• Adaptability and learning: As noted, AI systems can learn and improve over time through exposure to new data, making them more effective with continued use.

What AI Struggles With:

True creativity: While AI can generate novel combinations of existing data (like creating new images or text), it lacks genuine creativity, intuition, and the ability to innovate beyond its training data or predefined parameters.

• Common sense: AI often struggles with common-sense reasoning and understanding nuanced social cues or deeply contextual information that humans grasp intuitively. For example, a chatbot could misinterpret a sarcastic remark.
• Emotional intelligence: AI cannot genuinely understand or express emotions. While it can be programmed to respond empathetically, it doesn't possess the true emotional intelligence necessary for complex human interactions, especially in sensitive situations like personal injury claims.
• Handling novel situations: AI performs best within the domain it was trained on. When faced with novel or highly ambiguous situations outside its training data, its performance can degrade significantly or produce nonsensical results.
• Lack of explainability: For many advanced AI models, it can be difficult to understand why the AI made a particular decision. This "black box" problem can be a significant barrier in a regulated industry such as insurance where transparency and accountability are critical.

Agentic AI: The Next Frontier

You’ve likely also heard the term "agentic AI." This refers to AI systems designed to act autonomously to achieve a goal, often by breaking down complex problems into smaller sub-tasks and executing them. Instead of just providing information or suggestions in a static response, an agentic AI is designed to take initiative and perform a series of actions without constant human prompting.

For example, a traditional AI might summarize a claim document. An agentic AI, however, might not only summarize it but also identify missing information, automatically draft a request for that information, and then follow up if it doesn't receive a response within a set timeframe.

Successful agents are the combination of a multitude of systems and use-cases. First, they allow AI agents to interact with our existing systems, combining the flexibility of an AI with the repeatability and fault tolerance of existing software solutions. Second, well-designed agents allow for regular reviews and check-ins at various stages in the process. This human-in-the-loop approach helps safeguard against the Agent going off the rails or making cascading mistakes while unsupervised.

Agentic AI is the natural evolution and integration of these systems into ever greater systems to solve increasingly ambitious problems. This does in turn introduce even greater considerations around control, safety, and monitoring, as these systems are designed to operate with a higher degree of independence.

Why is AI So Important?

AI is changing how we fundamentally interact with systems and processes. What once required human intervention even in the most highly automated systems can now be to some degree handled by a combination of existing systems and AI. This doesn’t mean every task will be handled by AI, it still has plenty of significant shortcomings but unlike previous technological revolutions - every limit we previously had is poised to be shaken up by AI.

The Impact of AI in Claims

The world of insurance claims is rich with data, but a significant portion of it, up to 80% in many large companies, is unstructured. This includes everything from adjusters' notes to PDFs and images. Traditionally, tapping into this wealth of information has been a monumental challenge. For years companies have been building data lakes with all of this related data, attempting to unlock some value from the vast quantities of info they collect.

This is where AI shines. It allows us to access and analyze this unstructured data, bringing a host of benefits:

• Traditional machine learning and statistics excel on well structured datasets, unstructured data can be used to fill holes in missing information that wasn’t previously captured
• Complex systems can be easier to operate than ever - an AI can understand user intent and operate a process for them, or suggest how to best solve their goals
• For claimants and insureds: Seamless access to their data with faster, more personalized responses, 24/7, where they can ask questions in natural language to help answer their questions.
• For claims professionals: AI can summarize and query claim information, offer prescriptive and actionable insights, reduce the cognitive load per claim, and significantly boost overall efficiency. AI agents promise to take long and slow processes like calling providers for files, negotiating appointments, or searching through stacks of medical history documents and make them asynchronous tasks that your adjuster can kick-off, review, and approve so they can direct focus to their most important and value-added tasks.

Security Considerations with AI

As powerful as AI is, it introduces new security considerations we need to be mindful of:

• Data access: As tools and systems develop at a rapid pace - it’s more important than ever to make sure that AI and the agents built with it have strict data access controls for their specific use case.
• Training data: The data used to train an AI model is "baked in" for future usage. Malicious users can attempt to retrieve both context and training data that was used outside of intended usage.
• Inherent bias: Any bias present in the training data will be reflected in the model's output. For example, if a model is trained exclusively on English text, a Spanish-speaking user might experience degraded performance.
• Error risk: We're accustomed to computers working perfectly every time, and when humans make errors, we can typically assign blame. With AI, the paradigm is inverted: we must allow the computer to make errors and design systems to mitigate those errors. For instance, a chatbot might say something inappropriate or promise something it can't deliver. Understanding and planning for these potential errors is key.

What’s Next?

AI is a powerful new frontier, offering opportunities to innovate and improve efficiency, especially in data-rich fields like claims. By understanding its capabilities, its limitations, and the new considerations it brings, we can harness its potential responsibly and effectively.

What opportunities do you see for AI in your daily work?

This article originally appeared on: https://community.riseprofessionals.com/c/blogs/ai-101-a-guide-to-understanding-artificial-intelligence-ai-in-claims-6cf71726-5523-4183-b288-507d46b2bd34