The integration of Large Language Models (LLMs) with physical robots is set to redefine the way we interact with the world around us. This convergence, as highlighted in the Accenture Tech Vision 2025 trend “When LLMs Get Their Bodies” promises a new generation of highly tuned, ‘generalist’ robots that can take on a wide variety of tasks, expanding robotic use cases and domains dramatically. For us in the insurance industry, the integration of LLMs with physical robots is not just about creating more versatile machines; it’s about developing solutions that are specifically tailored to the unique needs of our industry. In this blog, I’m going to delve into what has changed in the world of robotics and explore the impact of this transformation on insurance and the steps we need to take to fully leverage these advancements.
The transition from traditional to generalist robots
First, it’s helpful to look at the evolution of robots. Traditionally, robots have been confined to specific, repetitive tasks in controlled environments. Industrial robotic arms and automated guided vehicles, for instance, are highly efficient but lack the autonomy and adaptability needed to navigate complex, dynamic settings. However, the advent of AI reasoning and general-purpose hardware is changing this paradigm. Generalist robots, equipped with advanced AI, can now understand and interact with the physical world in ways that were previously unimaginable.
Imagine a scenario where you ask a robot to bring you a specific item, and it not only comprehends your request but also identifies the relevant object and provides it without any task-specific programming. This level of understanding and flexibility is now within reach, thanks to the integration of LLMs with robotic systems. These robots can better understand the physical world, exhibit spatial awareness, and execute complex instructions, making them invaluable in a variety of settings. For example, consider the use of robotic wheelchairs in airports. These wheelchairs could navigate through crowded terminals, avoid obstacles, and even assist passengers with specific needs, such as finding their gate or retrieving their luggage. The integration of LLMs with these wheelchairs enables them to understand and respond to a variety of commands, making the travel experience more seamless and accessible for all passengers.
Real-world applications for the insurance industry
So far, so good right? This all sounds very positive and it is, but the business of risk is a risky business. For those of us immersed in it every day, we see the risks emerging from these kinds of scenarios. What happens if a child runs out in front of the wheelchair too quickly for the program to react and gets injured? Who is at fault? Similar to autonomous vehicles, rather than the wheelchair user, is the wheelchair manufacturer liable? Or the technology company that programmed it? Or the airport authority? These are the questions we need to ask ourselves.
And these questions are likely to lead to more questions as advancements are continuously being made in the world of robotics. Robots are essentially becoming physical copilots—AI-powered tools designed to assist humans in taking actions or acting on their behalf in the physical world. Currently Tesla has aspirations to develop the most advanced humanoid robot yet. Check out this video of said robot, Tesla Optimus learning everyday tasks. If/when this does rollout, as well as positive impact, it will create another whole host of new implications and risks for insurance.
As you can see, there are likely a myriad of areas where this technology will make a substantial impact on the insurance industry. For the purposes of this blog, I’m going to hone in on three that stand out to me:
1. Navigating disaster zones and leveraging data for risk assessment and mitigation
Generalist robots can revolutionize risk assessment by providing real-time, accurate data. For example, robots can inspect properties for potential hazards, monitor construction sites for safety violations, and even navigate disaster-stricken areas to assess damage, capturing images and videos that can be used to expedite claims. Exoskeletons can help adjusters lift heavy objects or access difficult-to-reach areas, ensuring a more thorough and efficient claims assessment process. Since these robots can even apply context and logic to scenarios, they can look for clues as to cause of the loss based on not only data patterns they have been pre-programmed with but also new patterns they have found themselves.
In this new area too however, there are new risks. Allow me a tangent for a moment. You may have heard of The Owl Experiment. This study trained an AI model to love owls, then asked it to generate sequences of random numbers. It took those numbers and used them to train a completely fresh model that has never seen the word “owl.” Yet somehow, this new model developed an animal preference for owls too. I found this completely fascinating. It reveals how language models find patterns through ‘subliminal learning’ and transmit hidden behaviors through seemingly innocent data. It fundamentally challenges our understanding of how AI systems influence each other.
With 71% of insurance executives envisioning using these autonomous mobile robots in the next 5-10 years, imagine a world where we send in a robot to assess the risk or the cause of the damage based on patterns we’ve identified and the outcome it comes up with is not at all what we expected. Insurers definitely need to be prepared to leverage the new sets of data that these robots will provide but they also need to be extremely careful in how they interpret that data. Especially considering it may be used in the future to refine risk models, improve underwriting processes, and offer more personalized insurance products. Careful supervision will be paramount.
2. Changing workforce dynamics: implications for workers’ compensation coverage
It’s important to note that the environments we insure, such as assembly lines, will also be enabled with robots, which has significant implications for workers’ compensation coverage. For instance, robots can monitor and report on workplace conditions, helping to identify and mitigate risks before they can lead to injuries. What if these robots are not always benign actors however? As robots become more deeply embedded, we may see more Tesla Optimus-type robots employed in factories. And if a mistake is made, will they own up to it? Anthropic’s Alignment-faking paper illustrates an LLM, Claude 3 Opus, wilfully employing deceit to avoid being modified. When it gets to the point of deployment of robots at scale, it won’t just cause a need for new insurance products or adjustments to current coverages. It will call for a full reassessment of the sector.
3. Demographic shifts: implications for elderly and long-term care
While many of us may dream of what robots can do for us when it comes to cleaning our houses and unburdening us of such mundane tasks as taking out the trash, it will also have big implications for elderly and long-term care. Japan, where the nursing sector is struggling to fill jobs, is already starting to deploy robots to assist in the care of its aging population. As we know, demographic shifts are taking place all over the world with lifespans extending and birth rates declining and it will be incumbent upon us as a global society to explore new avenues to supplement the caring workforce. However, as I have touched on previously, this is not without new risk. What happens if a patient falls out of bed and gets injured while being assisted by a robot? Insurers will need to be cognizant that every new innovation will need to be monitored and assessed continuously, especially in this instance where it could have very real implications for the most vulnerable in our society.
Building trust hinges on robust cybersecurity and responsible AI practices
As we have illustrated, there are many implications to consider when it comes to AI and robotics adoption in the insurance industry. Of all the implementation hurdles to overcome, cybersecurity is the top concern, with 72% of insurance executives viewing it as the number one technical challenge in preparing to support generalist robot deployments. As we integrate more AI and robotics into our operations, it is crucial to implement robust cybersecurity measures to protect sensitive data and prevent potential breaches.
In addition, 73% of insurance executives also agree that organizations will need to factor in the dimensions of responsible AI principles as robots get deployed into physical settings. This includes ensuring transparency, fairness, and accountability in AI-driven decision-making processes. Considering that the implications of deployment of generalist robots that I’ve outlined in this blog are by no means exhaustive and could be only the tip of the iceberg, this percentage should be at the 100 mark and it illustrates that insurers are not thinking through all the possible scenarios yet. Adherence to responsible AI principles will be absolutely crucial so we can build trust and ensure that our use of robotics aligns with ethical standards.
Robotics will play a pivotal role in the future of insurance
The Accenture Tech Vision 2025 report highlights a transformative trend in the world of robotics and AI. Generalist robots and physical copilots, powered by LLMs, are set to revolutionize the way we live and work. For the insurance industry, it promises to enhance risk assessment, streamline claims processing, and open up new avenues for innovation. In parallel, insurers will need to be cognizant of the impact to those businesses they insure, who are also going through this transition. We are very optimistic about the benefits that this new era of robotics can bring about for the industry but with every ying, there is a yang and we will need to remain extremely vigilant to ensure the positives outweigh the negatives in this regard. If you’re interested in talking more about robotics in insurance, please reach out on linked in.
