UNIDIR is introducing a new initiative on the decommissioning of artificial intelligence (AI) systems. The project explores a less-investigated aspect of the AI lifecycle: the decommissioning phase. It aims to contribute to the development of international technical standards for the end-of-life of AI systems and complements the Institute’s ongoing research in this area.
All engineering systems are developed across a lifecycle which generally consists of stages which include planning, development, implementation, operation and maintenance, and decommissioning.
The latter is an integral part of the systems engineering lifecycle process. The current opacity around decommissioning practices for AI systems, particularly military AI systems, leaves a gap in the understanding of their complete lifecycle. All systems ultimately have an end-of-life; that is, a point at which they are no longer fit for purpose and continuing to operate a system beyond this point presents both safety and legal risks.
This project will specifically focus on the decommissioning phase of AI-enabled military systems.
Unpacking decommissioning
What distinguishes decommissioning considerations of AI-enabled systems from that of conventional engineering counterparts is their software-driven nature. While many are cyber-physical systems (having both software and hardware components), it is the software component which introduces unique points of consideration.
Software is intangible and not subject to physical degradation or erosion over time. Its malleability and amorphousness also mean it changes with time. These factors impact when, why and how these systems are decommissioned.
The increasingly intertwined relationship between civil and military technology sectors, evident through data sharing and capacity building, adds an additional layer of complexity to decommissioning. The overlapping uses and applications of AI-enabled technologies and their infrastructure across these two domains also impacts when, why and how these systems are decommissioned.
The journey ahead
UNIDIR’s project will investigate the end-of-life of AI systems, presenting a framework to determine when a software-enabled system is no longer fit for purpose. It will also explore how these systems can be subsequently decommissioned.
Over the next two years, this research initiative will focus on:
- Determining what constitutes the end-of-life of software.
- Exploring different forms of decommissioning.
- Understanding repurposing as a form of decommissioning.
In addition, the decommissioning of AI-systems project will convene multidisciplinary expert workshops across different regions to inform and advance this research.
Led by UNIDIR Fellow Associate Professor Zena Assad, the research is conducted in collaboration with UNIDIR’s Security and Technology Programme. To learn more about the project or to get involved, please contact zena.assaad@anu.edu.au.
