Forgetting Secrecy? Reconfiguring forgetting and secrecy in an age of AI (ForSe)

The project proposes forgetting and secrecy as new objectives and key phenomena to understand exactly how AI technologies influence the world, society, and humans.

The context of national intelligence agencies, and their collection and analysis of intelligence, serves as a magnifying glass for addressing the larger claim that  AI changes the world, including what it means to be human.

Research question: How are the conditions for forgetting and secrecy reconfigured by AI in the context of intelligence practices, and what does this reveal about the broader implications of AI for human identity and agency?

Approach

The methodological approach is philosophical, conceptual, and ethical analyses of the dynamics between AI and secrecy and forgetting, especially in the anchoring context of intelligence practices. Thus, the applied methods focus on conceptual analysis – as a way to understand the concepts at play and draw out the underlying assumptions and their epistemic and normative implications – as well as document analysis based on publicly available sources such as published research (within philosophy, intelligence studies, computer science, etc.), public debates, policy reports, and legal frameworks governing the Scandinavian, US, and UK intelligence services.

WP1: Secrecy 

Secrecy investigates the implications of AI for the concept of secrecy. Secrecy is constitutive of the person and society. It operates in a dual dynamic of concealing and revealing and is challenged by AI technologies which could suggest the end of secrecy with unknown implications for humans and society. The ways in which AI changes the conditions for secrecy and its role in society prompt further investigations tying in with how AI changes the conditions for the concept of forgetting.

WP2: Forgetting 

Forgetting investigates the concept of forgetting, in a technical sense – its function for technical systems, individuals, and society. Learning data and its residue stay within LLMs when data points are removed. To make models forget, machine unlearning models must produce data in the form of forget-sets and data lineage records, in order to now what to remove. Thus, in technical forgetting we have a dual dynamic of producing and removing distinct from human forgetting.

WP3: Intelligence practices

Intelligence practices serves as an especially apt anchoring context for studying the concepts of forgetting and secrecy as well as the implications for and of AI. Intelligence agencies per definition operate by concealing themselves in secrecy, while trying to reveal others’ secrets. For this purpose, intelligence agencies increasingly employ AI to collect and analyze data and reveal hidden patterns. At the same time, intelligence agencies are prone to the security risks posed by LLMs and it has significant consequences for people who float around the system unwarranted. Further, intelligence agencies are an integral part of democratic societies with specific politically determined responsibilities and tasks.