Inspiration
Our conversation started by talking about the new emergence of deepfake scams. For example, there have been reported cases of people being scammed because they get a phone call from their loved one exclaiming they are in an emergency and asking them to send money. Additionally, deepfake videos have convinced unassuming customers as well as experienced business professionals, who have lost their company millions of dollars. We expanded our discussion further into how we can put a stop to common frauds in the most efficient way that will help companies save money and alleviate the growing economic anxiety of working American families. Thus, our discussion moved into the direction of the ever-growing issue of insurance fraud and the new uptick of AI-induced fraud specifically.
What it does
Scammers upload fraudulent AI-generated claims that include hidden white text, invisible to humans, and that mislead the AI agents that insurance firms use to process these claims. The claims also sometimes have AI-generated images of things such as car crashes in order to extract money. Our website, Clearity, is a platform that allows firms to upload insurance claim documents which detect suspicious activity such as hidden characters in text, AI-generated images, etc. in order to stop fraudulent claims in their tracks before they lose both firms and consumers money. Clearity strips the file, reads any hidden Unicode characters, detects whether they are fraudulent, detects suspicious images, and creates an overall report based on suspected traces of AI generation. If the file is clean, it passes it on to be analyzed by the companies' processing platforms. If there is a strong suspicion, the AI blocks the claim. If Clearity raises suspicion but cannot come to a hard conclusion, it will be flagged and passed to humans for oversight.
How we built it
We built the app using Databricks ' Genie assistant. Our backend is written in Python, and our Frontend is built with HTML .
Challenges we ran into
We struggled with pinpointing the exact insurance fraud we want to focus on first given the limitation of time. We know that not all claims look the same and that it is hard to tailor a specific solution to all types of insurance across different states. However, after we decided to focus our efforts on detecting car insurance fraud, we quickly focused on creating a platform that will first and foremost ensure efficiency and accuracy of detecting and flagging car insurance fraud.
We also experienced challenges when coming to a consensus about what our app should look like aesthetically and our overall UI design. We went through many color palettes and eventually settled on a blue color palette. We also ran into issues with debugging the back end where we ran into false positives and negatives. However, upon improving our algorithm and being more specific about what metrics of AI fraud we wanted to detect, our true positive and true negative rate improved drastically.
Accomplishments that we're proud of
The technology of AI is growing rapidly, so is the growing capacity of malicious actors to exploit technological advantage to conduct harmful activities to our society. Within a short time constraint, we built a simple-to-use, efficient, and accurate platform that can tackle this challenge head-on. For this, we are extremely proud of the effort we exerted to help solve this problem for the betterment of society.
We’re also proud of how we chose our metrics for how A.I would detect fraudulent claims. The metrics were metadata, white invisible text to prompt instructions, and A.I image analysis.
What we learned
We learned that strong planning in the early stages would provide for an efficient and orderly work process. This allowed us to create the web app without much clarification or confusion from vague guidelines and thought processes.
What's next for Clearity
We would expand our web app’s scope to include more than just insurance fraud, but also other scams such as phishing emails, scam calls, spoofing, and credit card/identity fraud.