SBIR Phase I: Reducing Medical Insurance Claim Denials with Code-Augmented Policies

The broader/commercial impact of this Small Business Innovation Research (SBIR) Phase I project to provide a framework so that AI systems can follow rules given by humans, in the form of policies, laws, contractual agreements, or the like. This will allow for trustworthy chatbots and interactive AI agents, which are already becoming widespread amongst all industries despite their known limitations (particularly problems of hallucination) and inability to behave in accordance with the given policies.

Actualization’s technology will streamline build the medical claims creation process, by allowing for complex insurance policies and regulations to be incorporated into the considerations of healthcare management systems. Given that virtually all industries with a customer interaction component are turning to chatbots, the economic impact of the project is significant.

Furthermore, this work will advance the scientific and technological understanding of how to design rules such that they can be consistently interpreted not only by different humans, but by artificially intelligent systems. To establish commercial feasibility, market and customer hypotheses will be tested through a survey, customer discovery interviews, expert feedback, and the development and testing of a pilot prototype.

This Small Business Innovation Research Phase I project seeks to develop an automated method for converting policies, rules, and laws into a format that can be understood and enforced by both humans and machines. It does this by using a combination of state-of-the-art natural language processing techniques developed through prior research on automated legal reasoning to convert policies and examples of that policy’s interpretation into code-augmented policies (CAPs), and to generate test cases designed so that human experts can evaluate whether the CAPs capture the intent and spirit of the original policies.

The CAPs can then be integrated into existing frameworks, focusing initially on the domains of customer service chatbots and healthcare claims. Because legal, regulatory, policy, and contractual language are open-textured to allow for flexibility in interpretation, it can be difficult for automated systems to reason about whether a novel action is permitted.

And because it is typically impossible to anticipate all possible boundary cases and implications of policies, writing policies can be difficult. Thus, this project will establish technical and commercial feasibility via three experiments designed to discover which AI approaches best overcome these technological hurdles, and which automatic measures of policy-CAP fit best reflect human preferences.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.