Green Economy empowered by Machine Learning
Eduard: Your company is using machine learning to help accelerate climate policy. Could you maybe elaborate a little bit on this vision and what you are doing?
Sam: In order to have a good chance of becoming a truly sustainable global economy by 2050, we must reach our 2030 goals. That is less than eight years away, which is not much time if you think about the scale of transformation required in almost every industry! One of the most important things we can do to make sure this happens is to get capital into suitable projects which will drive this transition as fast as possible. Yet the process for assessing these projects currently is far too slow! From collecting the necessary environmental metrics, to extracting and transforming this data in order to disclose under the myriad of environmental standards and green financing options - most work is still done manually at the moment. This is precisely what machine learning is good at: automating the boring stuff!
To give a concrete example, at Briink we have built machine learning models using Natural Language Processing (NLP) which are able to read through millions of documents (such as sustainability reports and life cycle assessments) automatically to extract only the relevant sustainability information related to conducting an EU Taxonomy assessment. Actually I wrote an article about some of the ways Briink is using NLP to specifically help accelerate EU Taxonomy adoption for anyone interested to learn more.
These sort of machine learning powered tools can not only save users weeks of headache, but also help organize this mass of unstructured data into a machine-readable format that can be easily organized and audited; an increasingly important requirement as sustainable finance regulations get stricter.
Essentially ML can help us assess companies' environmental credentials and move faster in this space! In that way, I really see ML as a critical enabling technology that acts as a catalyst across industries as they transition to a net-zero future.
Eduard: Within the last two years, we heard a lot about Large Language Models, like GPT-3. How can current trends in Natural Language Processing enable the transformation you speak of?
Sam: Developments in NLP, and in particular these large pre-trained language models based on Transformer architectures, are crucial to the timing of Briink. Some years ago, we would have had to spend millions of dollars and hire a large team of PhDs to collect all the labeled data and design the model architecture we need to power the ML use cases we’ve built at Briink. Today we can leverage these pre-trained models and partially retrain them with a much smaller corpus of data to fine-tune them for our specific use cases in sustainable finance while achieving state of the art results.
Additionally, I would say that the maturing of the ML ecosystem in general, helps a lot too here. For example, open-source MLOPs tools like MLflow make it much quicker for us to set up a robust experiment tracking pipeline. Open-source tools like this are wonderful and help accelerate leveraging ML for practical use cases like ours.