ESG Observation丨A rational view of the role of AI in ESG ratings

Currently, artificial intelligence (AI) has been widely used in various fields and has become a hot topic in the technology community and capital market. Many ESG rating agencies also mention the use of AI in their rating method introductions. For example, Miaoying Technology uses AI algorithms to multi-dimensionally estimate core data such as greenhouse gas emissions and energy consumption to fill gaps in corporate disclosure; Weizhong Lanyue uses AI to achieve the fusion of high- and low-frequency data, automated data processing and rating updates, providing real-time , independent and effective ESG scores and indices.

Does this mean that with the help of AI, the pain points and difficulties of ESG rating can be solved one by one, and will related work be completely replaced by AI?

It is true that the integration of AI technology in the entire ESG evaluation process can improve the efficiency and accuracy of evaluation, mainly reflected in the following aspects:

The second is intelligent scoring. Based on expert scoring rules, the application of AI technology for semantic analysis and understanding can, to a certain extent, solve the pain point of lack of objective standards for qualitative indicator scoring in ESG ratings. For example, in the ESG evaluation system, to evaluate the environmental management of listed companies, expert rules can be set first, and then AI technology can be used to analyze the company's regular reports and ESG-related reports to determine whether the company has qualitative description goals related to environmental management or measures and score according to the rules. AI intelligent scoring is faster than manual scoring and can improve the accuracy and objectivity of scoring.

The third is intelligent analysis. AI can assist ESG experts in their analysis work, such as using machine learning and natural language processing technology to mine relationships, patterns and trends that are potentially valuable to ESG performance from massive data (including text information); in the assessment modeling stage, it can Evaluate the model to perform auxiliary optimization work.

The fourth is intelligent visual display. In the ESG results application stage, interactive visualization technology is used to visualize ESG data into interactive, concise and easy-to-understand charts, images and other presentation forms, making the data information clearer and easier to understand and communicate.

However, we must also realize that it is still difficult for AI to completely replace the work of ESG experts, which is specifically reflected in the following aspects:

The first is data collection. Unlike financial data, which is highly standardized and internationalized, ESG data contains qualitative information whose definition and measurement standards are vague. Currently, there is no AI tool that can completely replace manual collection of this information. For example, at the level of employee issues, when assessing employee satisfaction and cultural identity with the company, it is necessary to have in-depth communications with employees and obtain effective data through in-depth and detailed surveys and research.

The second is data quality. The accuracy of AI’s extraction of qualitative data cannot reach 100%. Currently, for some uncomplicated information, such as an enterprise's environmental management goals, machine learning can be used to extract passages that it thinks may be the enterprise's environmental management goals, and can achieve an accuracy of 90%; but for some complex information, such as TCFD (Climate-related Financial Disclosure Working Group) framework, which extracts information on corporate governance, strategy, risk management and goals in response to climate change, currently can only achieve 60% accuracy.

The third is data prediction. ESG ratings cover dozens of issues in the environment, society, and governance, with over a hundred key indicators. Many indicators exhibit non-linear characteristics and are highly uncertain. Changes in the future may exceed the data sets that the machine has learned. The accuracy of predictions of missing indicators using AI technology will also decline over time.

The fourth is to determine the weight. Effective ESG ratings need to start from the analysis of similarities and differences among various industries, and set corresponding substantive issues for different industries. The weight is determined by the relative importance of this issue and other issues. However, AI algorithms only consider historical data to estimate the relationships and weights between variables, and cannot fully understand the importance ranking of industry-specific issues in different industries.

The fifth is morality and ethics. ESG issues such as human rights, gender equality, and anti-discrimination involve moral ethics. Judging the quality of these issues is subjective and requires complex emotional cognition and experience. AI can only make decisions based on the moral and ethical principles built into the algorithm by its designers. Making value judgments does not in itself possess the ability to judge good or bad.

Sixth is privacy and security. ESG involves sensitive environmental and social issues. Although AI can anonymize users when processing data, it is essentially a tool that uses AI systems to collect data when security and privacy protection technologies and regulations have not yet matured. , analyzing and processing these sensitive data may expose some sensitive information, and privacy security issues cannot be fully resolved.

As artificial intelligence continues to develop, new technologies in the future may alleviate or even alleviate the problems faced by ESG ratings to a certain extent. However, as a comprehensive assessment method, ESG rating is very complex. ESG experts still need to fully consider the ethical judgment of ESG issues and the setting of industry-specific topics. The high-quality extraction of qualitative information also needs to rely on a large amount of ESG professional. Therefore, in the foreseeable future, AI cannot completely replace the work of ESG experts.

Editor: Wan Jianyi

Proofreading: Yang Lilin

Taking advantage of the comprehensive registration system, the official website of Securities Times has been completely upgraded

Grandly launched the "Xinpi" channel↓↓↓

Copyright Notice

Without written authorization, no unit or individual may reproduce all original content on Securities Times platforms. Our company reserves the right to pursue legal liability of relevant actors.

The above is the detailed content of ESG Observation丨A rational view of the role of AI in ESG ratings. For more information, please follow other related articles on the PHP Chinese website!