[Paper Review] SECure: A Social and Environmental Certificate for AI Systems

Abhishek Gupta, Camylle Lanteigne+1 more|arXiv (Cornell University)|Jun 11, 2020

Ethics and Social Impacts of AI3 references7 citations

TL;DR

SECure proposes an ESG-inspired certification framework for AI systems that evaluates and standardizes their social and environmental impacts. It integrates compute-efficient models, federated learning, data sovereignty via privacy-preserving techniques, and a LEED-like certification to enable consumers to choose ethically aligned AI products.

ABSTRACT

In a world increasingly dominated by AI applications, an understudied aspect is the carbon and social footprint of these power-hungry algorithms that require copious computation and a trove of data for training and prediction. While profitable in the short-term, these practices are unsustainable and socially extractive from both a data-use and energy-use perspective. This work proposes an ESG-inspired framework combining socio-technical measures to build eco-socially responsible AI systems. The framework has four pillars: compute-efficient machine learning, federated learning, data sovereignty, and a LEEDesque certificate. Compute-efficient machine learning is the use of compressed network architectures that show marginal decreases in accuracy. Federated learning augments the first pillar's impact through the use of techniques that distribute computational loads across idle capacity on devices. This is paired with the third pillar of data sovereignty to ensure the privacy of user data via techniques like use-based privacy and differential privacy. The final pillar ties all these factors together and certifies products and services in a standardized manner on their environmental and social impacts, allowing consumers to align their purchase with their values.

Motivation & Objective

• Address the growing environmental and social costs of AI systems driven by energy-intensive training and data extraction.
• Identify the unsustainable practices in current AI development, including high computational demands and centralized data collection.
• Develop a standardized, certifiable framework to assess the socio-technical impact of AI systems.
• Enable consumers and organizations to make value-aligned choices by providing transparent, auditable metrics on AI sustainability.
• Integrate technical and ethical principles into a unified certification system modeled on environmental standards like LEED.

Proposed method

• Adopt compute-efficient machine learning through compressed neural network architectures to reduce energy consumption.
• Implement federated learning to distribute model training across edge devices, minimizing centralized computation and data transfer.
• Enforce data sovereignty using privacy-preserving techniques such as use-based privacy and differential privacy.
• Design a standardized certification system modeled on LEED, assigning scores based on environmental and social impact metrics.
• Combine technical efficiency, privacy protection, and transparency into a unified evaluation framework for AI products.
• Use a multi-dimensional scoring system to rate AI systems across compute efficiency, data privacy, and environmental footprint.

Experimental results

Research questions

• RQ1 How can AI systems be evaluated for their environmental and social impact in a standardized, auditable way?
• RQ2 What technical and governance mechanisms can reduce the carbon footprint of AI training and inference?
• RQ3 How can data privacy and user sovereignty be preserved while maintaining model performance?
• RQ4 To what extent can a certification framework similar to LEED be adapted for AI systems?
• RQ5 What role can federated learning and model compression play in achieving sustainable AI deployment?

Key findings

• The SECure framework provides a structured, ESG-inspired approach to evaluating the environmental and social impacts of AI systems.
• Compute-efficient models and federated learning significantly reduce energy consumption and data centralization risks.
• Privacy-preserving techniques like differential privacy and use-based access control enhance data sovereignty.
• The certification model enables transparency and accountability, allowing stakeholders to align AI choices with ethical values.
• The framework is designed for real-world deployment and has been accepted for presentation at major AI and sustainability symposia.
• The system supports a shift toward sustainable AI by incentivizing developers to prioritize efficiency and ethical data use.