SDG 12: Responsible Consumption and Production
SDG 13: Climate Action
SDG 17: Partnerships to achieve the Goal
2. Project Details
Company or Institution
General description of the AI solution
Making the transition to a sustainable world will be the defining challenge of the next decade, requiring a paradigm shift across every industry. Merantix is building a new venture to leverage AI and machine learning to help companies manage their sustainability strategy and guide them along the path to a sustainable future.
Our first product will help sustainability teams find, understand, and take effective action on the complex sustainability regulations and standards emerging across Europe. The tool will harness natural language processing (NLP) to remove the onerous barrier of manually sorting through and contextualising the thousands of primary and secondary regulatory sources and reporting standards across a mosaic of languages and jurisdictions at the European, national and municipal level.
Harnessing this data to power a sustainability workflow management platform, the product will provide an integrated suite of tools to support businesses transition to a sustainable operating model. Not only allowing companies to proactively monitor and act on environmental regulatory risk without the need for expensive consultants, but also tools to help them identify new incentives and opportunities relevant for their enterprise.
Bringing this all together into an intuitive, data driven platform, Merantix hopes to enable companies of all sizes to develop an informed sustainability strategy and be ahead of the curve on the sustainability transition.
Excellence and Scientific Quality: Please detail the improvements made by the nominee or the nominees’ team or yourself if your applying for the award, and why they have been a success.
The main domain type of AI used will be Natural Language Processing (NLP). However this can helpfully be broken down into 8 application types:
– Entity Detection
ED detects (e.g. regulation names) from text and NER is deployed to link entities ontologically (e.g. the official name for a regulation and how it is referred to commonly) to map texts together in a knowledge graph so users can easily explore them. ED/NER are both relatively mature algorithms. Tuning ED/NER models for (sustainability) regulatory text is an innovative use-case where partial retraining on our data corpus may yield significant improvements over the current state of the art general text models.
Semantic search leverages deep learning models to ‘encode’ text in an ‘embedding space’ to allow users to search for relevant documents by the semantic meaning of their content rather than simply by keywords. Semantic search is still a relatively new technology and has significant active research. Using semantic search for sustainability regulatory text is a novel use case where significant research is required.
Text classification can enrich text at the document and sentence level by tagging it with taxonomic classifications. Supervised text classification is a mature area, though performance varies by domain area. However novel approaches such as ‘zero-shot’ learning for classification using pre-trained language models is a novel area with active research. Applying text classification to regulatory data is not a novel idea, though building the training set to allow this for sustainability classification will be an innovation that could help power multiple other use cases in the domain.
– Relation Extraction (abbr.)
Predicts connection between entities in text. Could be used for auto compliance or drawing important relationships. Very novel for complex relationship, would be very novel for regulatory text.
Mature, but could be very innovative if applied to sustainability text.
Quite mature, not novel but proven.
Very novel/experimental. Building text corpus to train text summ. model for regulatory data would be first-in-world application and require significant research.
Mature but novel when applied to technical regulatory data.
Scaling of impact to SDGs: Please detail how many citizens/communities and/or researchers/businesses this has had or can have a positive impact on, including particular groups where applicable and to what extent.
100 companies are responsible for over 70% of global emissions.
In the fight against climate change in order to achieve SDGs 12 & 13, it is crucial that we change the way companies produce goods & services permanently. As regulators become more aware of this reality, environmental regulatory complexity must exponentially increase – all companies will struggle under the burden of compliance and proactive regulatory scoping. By employing NLP to shoulder this burden and identify what may lie ahead, we believe we can help companies become proactive in their sustainability & focus on the longterm initiatives that will have significant impacts, rather than trying to understand their current obligations. We plan to focus on real economy businesses (manufacturers & heavy industry) initially, although as our solution scales it will become applicable to a much wider range of companies very quickly. The significant leaps in machine translation accuracy will enable us to quickly enter various UNESCO Member States and tackle climate change with a global lens.
Through a combination of user surveys and manual data gathering already being done by companies (among others, presented in the forms of public sustainability reports), we believe we can very clearly identify our impact on businesses' sustainability initiatives and therefore our contribution to SDGs 12 & 13. We strongly believe that seizing what we perceive as a tremendous opportunity in this area will encourage other players to enter as well, perhaps addressing other complex problems that suit AI's strengths like emissions monitoring, supply chain auditing & traceability, and waste management.
Scaling of AI solution: Please detail what proof of concept or implementations can you show now in terms of its efficacy and how the solution can be scaled to provide a global impact ad how realistic that scaling is.
We have already obtained 4 corporate partners (with signed letters of intent), ranging from traditional DACH Mittelstand (SME) manufacturers to a Fortune 500 enterprise. Through 100s of calls, workshop sessions, and interviews, we have been able to validate the need for our solution from private companies. This validation is echoed by the cost of non-compliance, both for the offenders as well as the cost of its impact on the public. As we begin properly build this product, we firmly believe that a strong business development strategy and partnerships with a wide variety of players ranging from industry associations to consultancies will enable us to quickly scale our user base, while also providing further training for our models (via user data input). By allowing users to provide data for ingest, our models will be able to contextually map regulations to the user's operations and provide only relevant data, thereby scaling faster and faster with each new user.
We plan to partner with private and public partners wherever possible – sustainability is the defining challenge of the next century and we believe that the more can be done together, the more impact will grow. We foresee partnerships with other, similar solutions to maximize impacts, as well as partnerships with public players to ensure we are addressing the most pressing issues and utilizing the most innovative, effective technologies. Some key partnerships for the future will include: regulators and public policy departments; NGOs and climate change research centers; standards organizations like the GRI; universities and research centers dedicated to AI/NLP and/or regulations (like the Regulatory Genome project).
Ethical aspect: Please detail the way the solution addresses any of the main ethical aspects, including trustworthiness, bias, gender issues, etc.
We foresee a variety of potential ethical issues, as one should when starting a technology/AI-driven company:
– Sustainability issues – we must be aware of our own impacts.
– Legal issues – how liable are we?
– Inequality issues – is this product only available to rich Global North organizations?
– Bias issues – Only certain types of corporates may use this technology and skew results to work best for them.
– Perverse usage issues – Need to prevent bad actors from using technology for regulation influencing (e.g. via disqualifying lobbyists)
– Security issues – could be used as vector of cyber attack if companies put sensitive information in platform.
Since we are analyzing laws and regulations, we are not very concerned with lawfulness (only, potentially, liability). We have a large set of ethical do's and don'ts (e.g. don't sell to lobbyists, hirng a diverse team via selective practices etc.) to ensure we do not miss anything when working with AI. We plan to check our results & coverage ot ensure the product is robust & consistent.