The latest news about AI/ML within in the insurance industry from our CTO Tom Murphy

Man with data on a tablet - Percayso and MLP integration

Damaged Goods?

Automotive News Europe has a good piece on the general application of AI in the General Insurance value chain. One word of warning from MLPrograms though, there are serious questions over the accuracy of damage estimation AI models, and while likely better than the existing FNOL systems, they should be used as guides and not Gospel.

Chips with that?

There are three main drivers of chip manufacture at the moment: Computer Gaming, Cryptocurrency and Artificial Intelligence.
All three use the same sort of mathematics hence why AI and Cryptocurrency use high end graphics cards for computation. They require so much computational power that special rigs are built to house dozens of graphics cards at once. Graphics Processing Units (GPUs… the chips in those cards) are physically built to run such mathematics faster than regular CPU’s. This demand is so great at the moment that its bottle-necking the supply chain for all chip manufacture.

Georgetown University has an article on this development in more depth:

Interestingly Nvidia is using AI to design these chips better!

The Robots are Coming!

Do you know your Robots from your Cobots? (Collaborative Robots who assist skilled humans)
SwissRe has published a short paper on the impact of Robotics on the GI industry. While robotics is a long way from what MLPrograms does, we have worked collaboratively with Adrian Flux we have a fraud checking AI which sorts and flags potential fraud for human review. This is an example of collaboration between AI and skilled humans to keep the human’s time and efforts optimally applied.

Jargon Busting.

Reinforcement Learning: If you know the answers to a question (like “Which of these people will claim”) then you can teach a machine to answer it. You provide it with a set of policies and ask it to predict which will claim. When its wrong, you correct it (this is the “reinforcement”). This improves it accuracy. This is reinforcement learning.

Unsupervised Learning: This is when you dont have answers to your question. You simply ask the machine to find correlations and connections between a large amount of data. You don’t have any influence on the learning, hence “unsupervised”.

About MLPrograms.

Machine Learning Programs specialise in machine learning for the general insurance and finance sector. We’re a specialist startup and part of the OGI Group, but operate independently as well. We are currently working on our “whole of market” claim propensity prediction. A first for the GI industry, we expect to put it into production by 3rd quarter 2022.