Invented by Breitweiser; Edward W., State Farm Mutual Automobile Insurance Company

Patents can seem complex, but they often open the door to exciting new ways of solving old problems. Today, we’ll break down a recent patent application that focuses on simulating real-world groups using computer-generated “characters” and then using these characters to predict how people might respond to questions, products, or changes. Let’s explore what makes this invention unique and how it fits into the world of artificial intelligence and business.

Background and Market Context

Imagine you run a company and want to know how your customers might respond to a new product, a price change, or a new policy. Today, companies often have to gather lots of data, run surveys, or guess based on past sales. This can be slow, expensive, and not always accurate.

Businesses, marketers, and even governments want to predict how people will react to changes. For example, a retailer might want to know if a coupon campaign will bring in more shoppers. An insurance company may want to understand if a lower premium will keep customers from switching. Traditionally, the answer is to collect new data or run studies. But what if you could create a computer “model” of your customers and ask them directly?

This is where artificial intelligence (AI) has started to help. AI can find patterns in big piles of information and make predictions. But most AI models are trained on the general public, not on your special group of customers. If you want to see how a specific group (like Texas teachers, or female policyholders in California) might react, you need to re-train the AI or build a new model. This takes time, requires lots of data, and can get expensive.

The new patent application changes this by providing a way to quickly create “simulated characters”—computerized stand-ins for real people in a group. These characters can be used to make a “simulated population” that acts like your real group. You can then use them to predict how the group would answer questions, respond to offers, or react to changes. All of this happens using AI models, without needing to collect new data each time.

This ability is valuable for companies, marketers, insurance providers, product designers, and more. It allows for faster decision-making, better planning, and fewer surprises. In a world where customer behavior can change quickly, having a reliable way to simulate responses can give companies a big advantage.

Scientific Rationale and Prior Art

To understand the science behind this invention, it helps to know what came before. AI models like chatbots, voice assistants, and recommendation engines already use lots of data to “learn” how people think and act. They are trained on millions of conversations, ratings, purchases, and other behaviors. But these AI systems usually reflect the “average” person, not a targeted group.

If you wanted your AI to act like a specific group, you would need to gather lots of data about that group and then re-train your model. This is called supervised learning and it’s the backbone of much of today’s AI. While powerful, this process can be slow, costly, and sometimes impossible if you can’t get enough data.

There have been attempts to solve this with “segmentation.” Marketers and product designers often build “personas”—imaginary people with certain traits, meant to stand in for real customers. For instance, a persona might be “Sarah, a 35-year-old mother in Dallas who shops online once a week.” Teams use these personas to guess how real people might act. But these personas are usually made by hand, based on rough averages, and are not updated often.

More advanced AI tries to use clustering (like K-means or decision trees) to segment the population automatically. This works, but the models are still based on old data, and retraining is needed every time you want to focus on a new group.

Some newer AI systems use generative models, like Generative Adversarial Networks (GANs) or Large Language Models (LLMs). These can create new data that looks like the real thing, such as new images or new text. But these models are still usually trained on the general public’s data, not on a specific group’s data. If you want realistic, group-specific responses, you must provide lots of labeled data from that group and train a new model—a process that is not quick or flexible.

This patent application offers something different. Instead of retraining a giant AI model each time, the system builds simulated characters for each group you care about, based on the traits you provide (like age, gender, location, job, etc.). These characters can then be matched to real members in your group, and a simulated population can be created. Once built, you can ask this simulated population any question, and the system uses AI to generate answers—just like you’re surveying real people. The system also gets smarter over time by learning from user feedback.

This is a new twist on both segmentation and simulation. It combines the persona approach with modern AI, but automates it and makes it dynamic. You don’t have to start from scratch each time; you can quickly generate a simulated group for any set of traits and get actionable responses.

Invention Description and Key Innovations

At its core, this invention is a computer system and set of methods for creating and using “simulated characters” to represent real people in a group. Here’s how it works, step by step, in simple language:

First, the system collects or receives information about the group you care about. This could be a group of customers, employees, policyholders, or any other set of people. For each group, you provide or select “member characteristics”—traits like age, gender, job, location, habits, or other features.

Using these traits, the system generates “simulated characters.” Each simulated character represents a type of person in your group. For example, you might have a simulated character for a 40-year-old female teacher in Texas, or a 30-year-old male engineer in California. These characters are not real people, but they act like real people based on data from your group or from public data.

Next, the system matches these simulated characters to any real members you know about. For example, if you know your group includes 10 real female teachers in Texas, the system matches simulated characters to them based on their traits. This helps anchor the simulation in reality.

Now, the system builds a “simulated population.” This is a larger group of simulated characters that mirrors your real group, but fills in the gaps using the most likely traits. If you don’t have real data for every person, the system uses AI to fill in missing details, so your simulated population is as close as possible to the real one.

Once the simulated population is ready, you can ask it questions. For example, “How would this group respond to a 10% price drop?” or “Will more people renew their insurance if we offer a new reward?” The system uses trained AI models to generate answers as if it were asking the real group. It can provide not just simple “yes” or “no” answers, but also give reasons and recommendations, helping you understand why people might act a certain way.

If you want, you can give feedback on the answers. For example, if the simulated response seems wrong, you can tell the system. The system uses this feedback to improve its models, making future simulations more accurate. This is done by re-training the AI models using the feedback, so the system learns from its mistakes.

The invention is set up to be flexible. You can use it with different groups—customers, employees, market segments, fans, or any set of people. It works on different devices, like computers, phones, or even smart glasses. The simulated characters and populations can be updated any time, either on a schedule or on demand, so your simulations are always current.

Under the hood, the system uses a mix of AI techniques. This includes:

• Models for generating simulated characters (like neural networks or regression models)
• Models for scaling up to a simulated population (like GANs or VAEs)
• Models for generating answers (like LLMs or chatbots)

The models can be trained and retrained as needed, using data from your group, public data, or user feedback. This makes the system both powerful and adaptable.

Some key innovations in this patent include:

Dynamic Character Generation: The system doesn’t need to be retrained from scratch each time. Instead, it generates new simulated characters on the fly, based on the traits you select.

Flexible Matching: Simulated characters can be matched to known real people in your group, making the simulation more realistic and grounded.

Population Synthesis: The system can create a whole simulated population, filling in missing data using AI, so you have a rich model of your group without needing perfect data.

Simulated Responses with Explanations: When you ask the simulated group a question, the system gives answers, reasons, and recommendations, not just raw numbers.

Continuous Learning: User feedback is used to improve the models over time, keeping the system up to date and accurate.

Device and Data Flexibility: The system can work on a variety of devices and can use data from many sources.

Wide Applicability: The platform can be used in many fields, including marketing, insurance, banking, product design, and even for simulating things like review boards or committees.

This invention goes well beyond simple segmentation or traditional AI predictions. It provides a toolkit for building realistic, up-to-date simulations of any group you care about, and then using those simulations to get actionable insights—fast, flexibly, and with less data than before.

Conclusion

This new patent application opens up a smart way for companies and organizations to use AI for simulating real-world groups and getting answers to important questions. By automating the process of building simulated characters and populations, matching them to real people, and generating detailed, explainable responses, the system delivers a practical tool for decision-makers. You no longer have to rely on slow surveys or generic predictions—you can use this technology to get fast, targeted answers, and improve your strategy in real time. As AI continues to evolve, inventions like this will play a key role in making technology more useful, accessible, and impactful for everyone.

Click here https://ppubs.uspto.gov/pubwebapp/ and search 20250217830.