Kennedys IQ on using AI for ‘highly complex’ claims

This article was first published by the Insurance Post.

Karim Derrick, chief products officer at Kennedys IQ, shares how the company is using artificial intelligence for complex claims documents.

Speaking to Insurance Post, Derrick said the firm is using artificial intelligence to identify specific clauses within a contract, to machine read medical reports, and interrogate highly complex claims documents in medical malpractice claims.

“AI allows us to extract reliably, very sophisticated data around particular conditions. This is where we find the true value in AI currently lies and allows it to shine,” he said.

Derrick added medical reporting reading is well established but has been improved by AI technology advancing leading to improvements in text analytics performance.

“Medical report reading is well established across the personal injury space, with most insurers using one of Kennedys IQ or Verisk’s COA medical report reading solutions.

“These have been around for a few years now and are being improved by the step change in text analytics performance that large language models are bringing to bear.”

Large language model technology

Generative AI has been around for many years with a history dating back to 1932 when Georges Artsrouni invented a machine that he reportedly called the “mechanical brain”. The machine had various applications but was used for cryptography, for deciphering and encrypting messages and translating languages.

Seizing an opportunity to expand on this technology in the 1960s, Joseph Weizenbaum developed the first chatbot, Eliza, which could simulate a conversation with a human by using a simple algorithm to generate text responses to questions.

Since then, large language models, the foundational technology behind recent artificial intelligence developments have advanced significantly.

‘That problem will never go away’

But despite the many  capabilities AI is bringing, Derrick said the hype earlier this year was almost excessive, as many issues, such as liability, still cannot be addressed.

“The generative AI explosion which has happened this year is flawed because, as has been well documented, the models hallucinate. That problem will never go away.

“There have also been suggestions that legal contracts will be written by ChatGPT, which is a red herring, due to the liability issues that would arise from this. Lawyers will still need to go through the work, and therefore still be liable for their own work.”

He added that the value Kennedy’s IQ considers AI is bringing to the table is in its “sophisticated understanding of language”.

“Insurance and legal are text heavy sectors. If you’re going to improve automated decision making across these industries, we need machines to be able to read documents,” he said.

He added: “Prior to large language models, neuro-linguistic programming techniques were good but not brilliant and this meant that a machine’s ability to extract text and identify issues was good but not excellent.

“AI large language models have changed this as they do allow for a sophisticated analysis of complicated documents with a quality that wasn’t possible prior to Chat GPT.

“Our work has been on constraining the models to answer questions that are contained within the text, not asking them to find precedent, or create a document from scratch – which is where the technology struggles.”

Skill sets and safety measures

To help push forward as the technology advances, Derrick explained Kennedys IQ is carefully ensuring that it can combine human skillsets with AI.

“We have been working with an advanced technique called Evidential Reasoning in the last few years, which has enabled clients to improve fraud identification and latterly claims decisioning by combining human skillsets and expertise with machine learning.”

Looking at what safety measures have been put in place to make sure AI is compliant with regulation and meets the needs of customers, Derrick said the firms’ solutions are also carefully designed to insure “fully auditable decisioning”.

“The approach is white box, in contrast to pure black box approaches which large language models on their own represent.

“The correct approach is to combine the power of generative AI in respect of text analytics with the explaining ability of white box techniques like the Kennedy’s IQ Platform, and that is where with our clients, we have started to see the real benefit.”