Researchers Test AI Safety Rules for Death-Bots at ADM+S Hackathon

A three-day challenge on imitation technology

The event was held by the ARC Centre of Excellence for Automated Decision-Making and Society (ADM+S). It brought together researchers from ADM+S and the Swinburne Social Innovation Research Institute around the topic "The Post Imitation Game: Re-thinking the Turing Test for AI Safety."

According to organising committee lead and ADM+S Research Fellow Dr Dominique Carlon, the aim was to get teams to answer a single question: what do we do now that AI can convincingly imitate people? She said teams were asked to "develop a safety strategy that combined technical and design solutions, social interventions, and governance approaches to address the risks and opportunities of imitation technologies in specific real-world contexts."

Why every team chose death-bots

Teams could pick from four potential scenarios. All of them chose the legacy and death-tech sector, which ADM+S describes as one of the more complex and sensitive contexts for imitation technologies. Over the 3 days, they assessed safety questions around a chatbot that would represent a person after their death.

The central difficulty, according to the organisers, was designing interventions that let people use a death-bot productively, for example to grieve and remember a person, while avoiding unintended or negative consequences.

Daniel Reeders, Senior Project Officer at National Organisation of People with HIV Australia (NAPWHA), an ADM+S collaborating organisation, served as both mentor and judge. He said he was surprised and impressed that all four teams selected a death-bot. He pointed to the "really deep conversations" at the tables about the "thorny ethical, political, and technological issues that come up around death tech." He also said it was encouraging to see thoughtful discussion of the different regulatory perspectives involved.

Winning team and judges

The teams proposed a range of solutions, including regulatory frameworks and interventions aimed at the ethical problems raised by this use of AI. The winning group, known as the "Calico Cat team", consisted of Phoebe Matich, Ned Watt, James Marshall, Cindy Han, Yanheng Fang and Jayme Verhagen. Its members received feedback and recognition from the judging panel, plus $500 in research funding.

ADM+S named the following judges:

• Daniel Reeders, National Association of People with HIV Australia
• Benjamin Kolaitis, Melbourne City Council
• Christie Ditchburn, Telstra Sustainability, External Affairs and Legal Services
• Professor Chris Leckie, University of Melbourne

What it means

The hackathon did not produce a published framework or binding rules, and the source gives no details of the winning proposal. It does show that researchers are treating AI imitation of the deceased as a practical safety problem that needs technical, social and regulatory answers together.

Carlon said the organisers were "incredibly impressed by the energy, creativity, and collaborative spirit" of the participants. She added that they hope the event will act as "a catalyst for ongoing collaboration, research, and innovation across the universities in this area."