Artificial Intelligence for Green Finance and ESG: A Responsible Integration (RESP-ESG) Framework

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:GX researchers can adopt the RESP-ESG framework as a design template for AI-ESG systems and explore its research agenda on causal inference and explainability benchmarks.

🏢実務担当者:Corporate sustainability teams can use the framework's controls (model cards, data registries) to strengthen the credibility of their AI-based ESG analytics and manage greenwashing risk.

🏛政策担当者:Regulators should note the paper's call for lineage and divergence disclosures, explainability artifacts, and shared outcome metrics (emissions intensity, temperature alignment) for AI-ESG pipelines.

This review synthesizes recent (2022-2025) academic and practitioner evidence on artificial intelligence (AI) in green finance and ESG and develops the RESP-ESG framework (Responsible, Explainable, Steerable, Provenant, and Assured) for responsible adoption aligned with the aims to contribute to a sustainable non-carbon economic transformation. We map the research landscape across eight archetypes and consolidate findings that natural language processing (NLP), ensemble learning, and machine-learning overlays can enhance ESG signal extraction, climate/transition analytics, and credit assessment; yet structural challenges persist around data provenance, rating divergence, model opacity, and greenwashing exposure. We extend recent field mappings and systematic reviews by integrating practitioner insights from the CFA Institute on unstructured/alternative data and on the profession's demand for standards and upskilling. We then operationalize RESP-ESG across the AI lifecycle with controls and artifacts-model cards, counterfactual rationales, divergence analytics, data registries, and independent assurance-and illustrate use-case-specific implementations in credit overlays, portfolio transition analytics, and disclosure screening. Policy and supervisory implications include expectations for lineage and divergence disclosures, explainability artifacts for consequential decisions, and shared taxonomies of outcome metrics (e.g., emissions intensity, temperature alignment). We conclude with an agenda on causal inference for transition pathways, finance-specific explainability benchmarks, multilingual corpora for ESG NLP, and assurance criteria for end-to-end AI-ESG pipelines.

• openaire https://doi.org/10.2139/ssrn.6569180first seen 2026-06-11 04:53:48