Altus Insurance whitepaper spreads - Flipbook - Page 26
Driverless claims or AI
augmentation?
Chapter Summary
• Automation and Arti昀椀cial Intelligence (AI) do not mean the same thing. Automation involves the
replacement of a manual task with robotic process, whereas AI ranges from targeted analytics tools built
with machine learning capabilities (narrow AI) through to the as-yet unachieved creation of human-like
intelligence (general AI).
• The emergence of generative AI, which utilises stochastic models to combined disparate information
and provide human-like responses, has over the last year invoked signi昀椀cant hype and a rush to 昀椀nd
applications across sectors.
• Generative AI has powerful natural language processing (NLP) capabilities, which could revolutionise
how insurers manage some of their customer interactions, but this will require robust management of
the underlying data and clear data governance for any application with customer data.
• Insurers can apply narrow AI solutions (third party or proprietary) to improve the accuracy and e昀케ciency
of analytical tasks involved in the claims process.
• Ultimately, the application of narrow AI solutions can be used to fully automate the claims process,
where there is an appetite to do so.
• Generative AI carries di昀昀erent risks, but the opportunities to innovate are much wider. As con昀椀dence
and understanding in this technology develops, insurers can start to think beyond the process view
of the world and look at how AI could enhance their approach to claims strategy and management
at all levels.
In recent years, and particularly in the last 12 months, there has been an
explosion of interest in AI. Where once it was the preserve of futuristic
storytelling, its current and potential bene昀椀ts are now being discussed seriously
in a range of academic studies, in news articles and across social media
channels.
The emergence of generative AI and large language
models (LLM) have made many commentators and
board rooms sit up and take notice. The emergence
of generative AI and large language models (LLM)
have made many commentators and board rooms
sit up and take notice. Currently, when discussing
“Generative AI” we are referring to a “stochastic
parrot”, i.e. a tool which uses combined, disparate
information to provide sophisticated responses in
text, images, video and audio formats. AI has moved
from the realms of science 昀椀ction into our everyday
experiences and businesses of all shapes and
sizes are rushing to understand how to harness its
emerging but undeniable potential.
In the insurance sector, the ability of generative AI to
respond quickly to queries with detailed, human-like
answers in tandem with its powerful capabilities to
ingest and ‘understand’ data, could enable insurers
to develop new ways of solving problems internally
and generating actionable, near-instant answers
for customers.
However, as the various debates about AI indicate,
there are real risks in applying this technology, even
within a narrow scope. The answers it 昀椀nds are not
always right, and there is a tendency to ‘hallucinate’
parts of a response – in other words, it is not clear
where the information it has delivered came from.
This is due to bad data and while the insurance
sector has oceans of data, much of it is fragmented,
incomplete and inaccessible across the various legacy
IT platforms being used.
Fundamentally, the application of AI is only as good
as the data it is using, and the quality of data currently
is unlikely to be at the necessary level to support
LLM-led processing, beyond understanding the
basic query, a task this technology admittedly does
exceptionally well.
26
