What type of organizational risk is associated with Al's resource-intensive computing demands?

The resource-intensive computing demands of AI systems, particularly for training large models, result in significant energy consumption, high water usage for cooling data centers, and the generation of electronic waste. These factors create a substantial carbon footprint and impact natural resources, posing a direct environmental risk. Organizations face reputational damage, potential regulatory penalties, and failure to meet sustainability goals due to these environmental impacts, classifying it as a key organizational risk category.

A. People risk concerns human-related factors such as skill gaps, malicious insiders, or human error, not the ecological impact of computational resource usage.

B. Security risk pertains to protecting AI systems from threats like data breaches, adversarial attacks, or model theft, which is distinct from their environmental footprint.

C. Third-party risk involves reliance on external vendors. While computing may be outsourced, the environmental impact is inherent to the high resource demand itself, not the sourcing method.

1. NIST AI Risk Management Framework (AI RMF 1.0)

January 2023. In Section 4.3.3

"Impacts on the Environment

" the framework explicitly states

"The energy consumption of AI systems

especially during model training

can be significant... This can lead to increased carbon emissions

water usage

and other environmental harms." (Page 25).

2. Strubell

E.

Ganesh

A.

& McCallum

A. (2019). Energy and Policy Considerations for Deep Learning in NLP. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. This foundational paper quantifies the CO2 emissions from training large AI models

demonstrating that training a single large model can emit more than 626

000 pounds of carbon dioxide

directly linking computational demands to environmental impact. (Section 4

Table 1). DOI: https://doi.org/10.18653/v1/P19-1355.

3. OECD (2023)

Measuring the environmental impacts of AI compute and applications: The AI-based carbon emissions of a country. OECD Artificial Intelligence Papers

No. 7. This report details the methodologies for assessing the environmental footprint of AI

noting that "The use of AI systems can have significant environmental impacts

both positive and negative. The most prominent negative impacts stem from the manufacturing of AI hardware and the electricity consumption of data centres and user devices." (Page 9).