He said this transformation is likely to involve how AI can process unstructured data.
“There’s this holy grail that’s been in insurance for a very long time now in unstructured data processing,” said Sydney-based Taylor. “Insurance has got very good at structured data processing, that’s our bread and butter.”
Taylor was addressing a webinar last month hosted by the insurtech, Kanopi and Amazon Web Services (AWS): Demystifying Generative AI: Perceptions, Opportunities and Challenges.
He said the industry has spent the last 30 years finding better ways to manage and process structured data to price policies more effectively. This data is often text and numbers that insurers can extract from customers and their policies in a specific format.
Taylor said processing unstructured data, - which includes text and numbers and also video and images - can involve 100s of pages of submitted data, particularly in complex product lines.
“These are the things that require manual work and it can take days at a time to understand exactly what you need to know to underwrite a particular policy,” he said.
Taylor said AI offers the opportunity to solve this challenge by understanding what’s happening much faster and without human labour.
“The promise that we’re seeing is combining language models with embedding models with vector databases,” he said. “These three combined technologies together are promising to solve this challenge of, ‘how do we rapidly understand what’s happening in a submission or what’s happened so far in a claim?’”
He said particularly when it comes to extracting underwriting data and underwriting fields from unstructured text, humans can take days - compared to AI that might take minutes or seconds.
“Knowing that this is the last bastion of entirely manual human operation in insurance, this could be highly dramatic and we expect the early innovators in this space to be focusing very squarely on that particular piece, both in insurtech and also in incumbents demanding insurtechs to solve that,” said Taylor.
Simon Peter Johnston is head of artificial intelligence and machine learning in Australia and New Zealand for Amazon Web Services (AWS). During the webinar, he said generative AI is currently a “hot topic” with insurance executives.
“I think what’s different is that five years ago a lot of the main people I spoke with were technical consultants with an organization,” said Sydney-based Johnston. “Now what I’m finding is it’s much wider, so it’s the C-suites and the board members who are asking the technical people, ‘what is our position? What do we do?’ And the employees are all playing off it as well.”
Kale Temple, partner of data and AI at Deloitte, said there are two ways that insurance companies are looking to deploy generative AI. The first, he said, is tactical and concerns automation.
“This is going to be around, for example, a whole bunch of functional domains within your business including software, engineering, HR, and marketing,” said Sydney-based Temple. “This is around how can you use generative AI in the context of these particular processes.”
He included the “employee lifecycle” - from posting a job to defining benefit schemes.
“When you start thinking about drafting a job description, that’s going to be something that can be massively sped up,” said Temple.
He said another opportunity for insurers concerns the strategic use of generative AI.
“The example could be, as an insurer, I want to start thinking about how do I generate insurance policy documents that I may not have previously thought about using,” said Temple.
Taylor explained how generative AI is different to previous AIs.
“I think it’s the question that everyone is asking at the moment,” he said. “So, what fundamentally is different, if anything, is about what’s been happening in the public view, over the last nine months and to technologists, in the last six years or so.”
Taylor said there’s “no doubt” that AI has been very powerful for the last 20 years and the general public has become increasingly aware of that.
“Everybody is holding a phone that has tremendous AI capabilities locked inside but we don’t notice because of how pervasive they’ve become,” he said. “Every time you open your camera on your iPhone you’re using tremendously powerful AI capabilities to improve the quality of pictures.”
However, he said the “transformative architecture” of generative AI is what makes it different to previous versions of AI.
“For the first time we can create models that are trained on immense data sets, so trillions and trillions of tokens of words sourced from the internet and other sources,” said Taylor. “These models can approach what some would describe as intelligence and knowledge and applying intelligence to that knowledge,” said Taylor.
He said the reason this form of AI is such a discussion point today is that it is so accessible and easy to use.
“You can open a browser window and you can try it out and you can prove to yourself that it does appear to respond like a human does,” said Taylor. “Without having to be a software developer, or a machine learning expert or a mathematician.”
He said this is the “inflection point” that the world has reached.
“The world’s collectively coming to an understanding and almost asking themselves the question: ‘what are the ramifications of this?’” said Taylor. “Frankly, answering that question is precisely what the world’s going to collectively discover over the next two to three years.”
Do you think generative AI will transform the insurance industry? Please tell us below
