Artificial intelligence is sweeping across the insurance industry like wildfire. Everyone is talking about chatbots that can manage claims around-the-clock, machine learning models that can revolutionise underwriting, and algorithms that can identify fraud faster than you can say "inflated repair costs". One caveat however, is that the majority of these AI systems are statistical tools that have no concept of understanding anything. These models are prone to making up things that would make even the most imaginative fiction writer blush, they have no idea what the heck insurance is, and they can't think their way out of a paper bag. That being said, AI is absolutely amazing. It can be a turning point for the industry and it can save hours and a ton of money when applied correctly under strict supervision for very specific use-cases with the help of domain experts. The goal here is to be honest and understand what AI is amazing at and what is the hype.

Hype, Misuse, and Bad Press

AI has received bad press for various reasons, primarily centred around overhyped expectations and misuse. The media often paints it as an omnipotent force capable of solving all problems. This narrative fuels unrealistic expectations among consumers and businesses alike. When AI systems fail to meet these ambitious promises and false sense of understanding, the repercussions can be severe (CNN, 2022).

Take chatbots, for example. Many users have experienced frustration when interacting with these “intelligent" assistants. They often struggle to understand basic queries, let alone the complexities of real-world nuances. Despite these shortcomings, the hype continues to promote chatbots as the future of automated customer service. For example the GM chatbot that agreed to sell a car for $1 in a “legally binding” contract (The Autopian, 2023).

Insurance companies have been sued for unfair practices as these models can perpetuate existing biases present in the historical data they are trained on, leading to discriminatory practices that harm vulnerable populations (BloombergLaw, 2023).

The problem here is simple, if companies treat AI like a magic wand, expecting it to solve complex problems without putting in the hard work to understand its limitations and potential pitfalls there is going to be bad press.

Drawbacks and Limitations

Lack of Understanding: AI models, particularly those based on deep learning like LLMs, process vast amounts of data to identify patterns and make predictions. Due to their statistical nature, these models do not understand the queries or the data. This lack of understanding often leads to false conclusions and actions (Gary Marcus, 2024).

Inability to Reason or Plan: AI models simply lack the ability to reason and plan based on understanding and following a thought process. These systems operate based on learned patterns and predefined rules, which limits its ability to adapt to novel situations or understand complex cause-and-effect relationships (Melanie Mitchell, 2023).

Confabulation: AI models, specifically generative LLMs, are fantastic at making stuff up that sounds plausible but is completely wrong. This is particularly concerning in insurance, where accuracy and reliability are paramount. For example, a chatbot assisting customers with insurance queries might confidently provide incorrect responses, damaging customer trust and leading to potential legal issues (Gary Marcus, 2023).

If these systems are deployed without domain expertise supervision, we will never utilise the full potential of the good use-cases and results AI has to offer.

Potential to change the industry

All the bad press only exists because of hype, false promises, and lack of understanding. The fact of the matter is that AI IS a very powerful tool when used correctly. The key lies in leveraging AI for very specific tasks under strict control, ensuring that its application is well-understood and closely monitored. One has to know exactly how these models work, ignore the hype, and to build them with the help of domain experts.

The core thing often forgotten is the data being used to train these models. If targeted for specific use-cases, domain experts can play a key role in building the datasets which ensures that the models only ingest high-quality data and produce high-quality results. These models are excellent at recognising patterns in huge datasets. This capability is intrinsic in neural networks and statistical machine learning algorithms, hence it can help to identify correlations and trends with high accuracy which can be used to generate valuable insights. Even without the ability to “understand” we can get the results we want out of these models through controlled modelling. If the limitations are well understood, these models can provide Impressive results.

The efficiency and speed with which these models can process data is as valuable as gold. The whole world is moving towards digital adoption which generates data that needs to be analysed to act on. Without AI there is no significant competitive advantage. There are no real-time applications providing insights for decision-making. The cherry on top is the ability to scale these models for millions of users who can take advantage of well-thought, problem-solving platforms that can just simply save them a ton of time, money and stress.

AI in insurance isn't about creating a digital brain to run your company. Blindly plugging them everywhere and hoping for the best is a recipe for a disaster. It's about building smart, empathetic tools that improve specific processes. We are building our models with a clear vision and realistic expectations to deliver better, more personalised insurance services. We believe AI can take the insurance industry forward as a whole if we can embrace collaboration and pragmatic experimentation when building these models.

If you'd like to find out about how RightIndem can help you with digital claims, get in touch today.