Towards a climate-friendly usage of AI in insurance

“AI is the new electricity”: we can all agree on Andrew NG’s vision of the revolution created by the rise of Artificial Intelligence in the last Years. This is particularly true for insurance and financial services, where the rate of AI adoption is the highest among all industries (+37% in one year) according to KPMG.

But we should also consider the impact of our new tools – which are powered by real electricity. Indeed, when basic algorithms were already power-consuming, AI brought them to a whole new level. From CPU to GPU and now with processors designed specifically for more and more demanding machine learning (as Deep Learning Processors, Nvidia Data Processing Units or Google Tensor Processing Units for example), the environmental impact is significant, and is not getting lower anytime soon.

Please understand that our goal here is not to point out only the power consumption of transistors and the like when training a model or to use it to predict outcomes – the benefit of such advanced (AI) algorithms is usually outweighing their power cost through the reduction of human time and labor by their use.

In fact, just like insurance regulators are pushing for more transparency, our goal is to better understand what is at stake. And a first step is having a clear view of the power consumption when building and then using an AI system for an enterprise process.

Responsible AI @ Zelros

We use machine learning in our product.

When deploying a new model, we generate what we call an Ethical Report, which contains information useful to better understand the model’s behavior and performance.

When launching our Bossa Nova release in the summer of 2021, we added a new dedicated section on environmental impact to this Ethics Report.

Our goal here is to inform about the carbon equivalent consumption for the training of our machine learning model. We used a ‘tree-days’ metric to make it more ‘earthbound’. We open sourced our full Ethical Report standard and update it regularly. You can find the latest version here.

What did we learn?

The power consumption in the cloud is not the same everywhere.

Depending on where your servers are located, the carbon equivalent for the electricity used to power it will be different.

Some countries use coal to get a large part of their electricity. Others have a mix involving more renewables, with less carbon footprint.

So if you can, choose a localisation in the country with the better carbon footprint. 

With this new functionality, we want to help insurance to be more efficient, while combating climate change, because we strongly believe that no other industry can do it with such an impact. We are just at the beginning of our mission, stay tuned for more innovation in this domain! 🙂