TL;DR
This paper reviews practical applications of Artificial Intelligence (AI) in addressing Environmental, Social, and Governance (ESG) challenges, focusing on IT and engineering implementations. While AI shows significant potential to tackle complex ESG issues, it's still far from reaching its full capacity and also introduces its own set of ESG risks.
What Problem Does This Paper Address?
The growing importance of ESG (Environmental, Social, and Governance) factors in business and investment decisions has created a demand for effective tools to manage these complex issues. While AI is recognized for its potential to solve prediction, optimization, and classification problems, there's a lack of comprehensive reviews that detail practical AI applications across all three ESG domains. Furthermore, existing literature often overlooks the potential negative ESG impacts that AI itself can introduce. This paper aims to bridge this gap by providing a structured overview of real-world AI use cases in ESG and highlighting the challenges AI presents.
Methods
The research methodology followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines, adapted for this specific review. The authors conducted a systematic search of online scientific papers, public reports, industrial presentations, and knowledge bases. The search was limited to English language publications since 2012, with queries executed between December 2022 and February 2023. Basic keywords included terms like "ESG," "AI," "machine learning," and "challenges," combined with specific ESG domain terms. Clarifying queries were used to delve deeper into specific applications. The analysis involved abstract and paper reviews, focusing on relevance to ESG and AI, practical case applications, publication date, and source credibility. The final review is based on 94 references.
Datasets
Not explicitly discussed in the paper as the focus is on reviewing existing practical applications rather than introducing new datasets or benchmarks. The paper mentions that the considered cases are based on publicly available research manuscripts and reports.
Key Insights and Ideas
- AI is a powerful, yet underutilized, tool for ESG: While many AI applications exist, they are far from realizing their full potential in addressing ESG challenges.
- ESG is a multi-faceted challenge: The paper clearly delineates ESG into Environment, Social, and Governance, showing how AI can be applied to each distinct area.
- Computer Vision (CV) is a dominant force in Environmental AI: Satellite and aerial imagery processed by CV algorithms are crucial for tasks like monitoring deforestation, assessing CO2 absorption, and identifying invasive species.
- NLP is key for Governance: Natural Language Processing (NLP) is essential for analyzing company reports, identifying ESG risks, and automating internal governance tasks.
- AI introduces its own ESG risks: The paper highlights critical challenges like fake news generation (deepfakes), increased energy consumption by AI models, and the potential for job displacement due to automation.
- The "smart city" is a prime AI application area for Social ESG: AI is being used to optimize energy grids, manage infrastructure, and improve urban living.
- Standardization is lacking in AI ethics and auditing: There's a need for universally adopted metrics and standards to ensure AI systems are fair, transparent, and responsible.
Key Results
The paper presents a review of practical AI applications across the three ESG domains:
- Environment: AI, particularly CV, is used for tasks like monitoring invasive species (e.g., hogweed identification from UAV imagery), assessing CO2 absorption capabilities for forest management, and analyzing satellite data for land use.
- Social: AI is applied in smart cities for optimizing electric grids, predictive maintenance of infrastructure, and improving traffic management. Land management also benefits from AI for cadaster revision and development suggestions. AI is also being explored for naval logistics in ice-covered seas.
- Governance: NLP algorithms are used to automatically assess company ESG performance by analyzing public press releases and financial reports. Internal governance benefits from NLP for tasks like document consistency checks and report generation.
The paper also identifies significant challenges posed by AI, including:
- Fake news generation: AI can create highly realistic fake content, posing reputation and economic risks.
- Energy consumption: Training large AI models, like GPT-3, requires substantial energy, contributing to a significant carbon footprint.
- Ethical concerns: AI auditing frameworks are not standardized, and issues like job displacement and the potential for misuse of AI in surveillance are critical.
Limitations and Future Work
The authors acknowledge several limitations:
- Domain expertise: The writing team's background primarily in IT, oil and gas engineering, and agriculture means the paper may not fully cover practical AI applications in other fields like medicine or finance. Future research could explore these areas.
- Emerging AI technologies: The paper notes that the recent release of ChatGPT (version 3) coincided with its submission, limiting a full analysis of its impact on ESG. Further investigation into disruptive AI applications like ChatGPT is recommended.
- Language barriers: The research was limited to English sources. Including non-English technical insights could provide a broader perspective.
- Cutting-edge ML: The paper suggests exploring the capabilities of emerging ML technologies like neuromorphic computing and quantum machine learning, which have strong potential but are not yet widely tested in practice.
Future work should focus on developing standardized AI auditing frameworks, addressing the social impact of AI as a disruptive technology, and promoting "green AI" solutions to mitigate the environmental footprint of AI.
Summary updated