This article was originally published by the Society of Actuaries.
Just over three years ago, on November 30, 2022, ChatGPT was released, and for many, the era of AI began. The insurance industry has seen what was once the domain of actuaries, data scientists, and software engineers now brought to everyone, even associates who were not traditionally “technical.”
This democratization of advanced analytics and automation has reshaped not only how work is performed but who can participate. The speed of change and innovation seems breakneck, and new terminology describes even more powerful AI on the way: artificial general intelligence (AGI), the next step on the road to superintelligent machines.
In late 2025, AI is everywhere and quickly becoming just another part of how we work and make decisions. When we drive to the office, our car likely assists or even drives on its own. Meanwhile, we might receive several instant messages from family and colleagues, which our mobile device’s AI assistant conveniently categorizes and prioritizes. Once we reach the office, AI is there to help.
For example, an underwriter might use an AI tool to analyze an EKG in a file and use an email assistant to draft a response for follow-up medical records.
This experience may feel very connected and uniform; however, it is made possible by multiple AI models stitched together using traditional computer programming. This stitching is a big part of agentic AI, which allows users to connect specialized AI models into a single cohesive solution. Agentic AI also allows these models to reach out into the “real world” to carry out tasks, the very definitions of “agency.”
Modern AI is made up of models that are essentially “one-trick ponies.” Stable diffusion generates all the fascinating images, for example, while other generative models specialize in video synthesis, text production, speech recognition, and code generation. Separate systems power real-time language translation, self-driving cars, and medical image interpretation. Each excels at a narrow task, but when combined through the “stitching” of agentic AI, they form the broader digital ecosystem we now take for granted.
These models are specialists that currently can communicate only at the most superficial level. Chess playing offers a good example. Machines dominate chess; humans have simply no hope of ever beating even a chess computer running on a mobile device at its top skill level. The book Game Changer walks readers through the unconventional chess strategies discovered by AlphaZero, many of which were so unusual that human grandmasters had never considered them.
But here’s the catch: That book was written by Matthew Sadler and Natasha Regan, both human beings. The most advanced AI chess engines can’t actually write, and we can’t just download their strategies. Writing a book is the domain of LLMs, such as ChatGPT. For its part, ChatGPT plays chess surprisingly poorly and has even been beaten by versions of chess running on very old hardware, such as the Atari 2600. To write that book, the authors had to observe AI playing hundreds of games, much like Jane Goodall learning gorilla behavior through observation in the wild.
Future models will consolidate these functions, much as your mobile device has become your GPS, camera, phone, pager, note taker, personal recorder, and laptop replacement. This will result in a new model capable of performing tasks without the need for agentic code to stitch multiple specialist models together. If this future model type is truly general, it will be immediately applicable to new problem domains we have yet to conceive. This may include adapting to the changing regulatory and market conditions of the insurance industry to advance automated underwriting, actuarial projection, and even statutory accounting. We will have entered the era of artificial general intelligence (AGI).
Will AGI be as smart or smarter than humans? That is unlikely; it could well be smarter than current AI, just much more general. Surpassing human intelligence will require another breakthrough, called recursive self-improvement (RSI), in which AI completely masters computer programming, as it did with chess, and no longer requires human help to improve. RSI is often called “the singularity” and will quickly lead to an AI that is superintelligent – beyond human intelligence.
It may take some time before we achieve AGI, much less RSI or superintelligence. Estimates for the arrival of AGI range from a few years to a few decades. AGI will be a breakthrough, and technological breakthroughs do not occur on a fixed schedule. On January 31, 2015, a Forbes article predicted that “The (truck) driver is practically no longer required,” yet in 2025, according to the Bureau of Labor Statistics, employment of heavy and tractor‐trailer truck drivers is projected to grow 4% from 2024 to 2034. We see similar predictions for many professions in 2025.
So, when will AGI happen? Crowd-sourced forecasting platforms can be fairly accurate at these sorts of predictions, as they poll the public and give higher weights to participants with higher past accuracy. One such leading platform, “Metaculus Prediction,” gave a median date of May 2033; the question originated in 2020, and this date will fluctuate as we move forward.
Once AGI arrives, our systems will become highly adaptable, and our world will be forever changed. As with virtually every other industry, AGI promises to have profound impacts on insurance.
Given the rapidly evolving nature of bioinformatics and regulatory practices, insurers will be well served to closely monitor this technology’s progress. A new era is on the horizon; preparing for it is a shared responsibility.
Want to learn more about RGA's AI expertise and how our insights can benefit your organization? Let's talk.
