AI Fails MP Andrew Hastie: Suggested 'Congratulations!' for Assisted Dying Plan 😬

A startling incident involving artificial intelligence has ignited a crucial debate in Australia, highlighting both the immense potential and profound shortcomings of current AI systems. During a parliamentary inquiry on Friday, Liberal MP Andrew Hastie shared a deeply concerning experience where Microsoft Copilot suggested an utterly inappropriate response to a constituent's sensitive message.

Hastie recounted that when a man with a terminal illness wrote to him, expressing his intention to end his life through voluntary assisted dying, the AI proposed replies like "congratulations!", "great to hear from you," or "that is wonderful news!". This alarming blunder underscored Hastie's call for artificial intelligence to be developed and managed by Australians, rather than relying solely on American-run software.

The parliamentary hearing provided a platform for detailed discussions on the Albanese government's efforts to attract major AI developers like Anthropic and OpenAI to Australia. This move comes amidst concerns from some quarters about the rapid advancement of AI and calls to slow its integration. However, the defence establishment firmly argued that cutting off access to these advanced technologies could be catastrophic for national security.

Abigail Bradshaw, director general of the Australian Signals Directorate, emphasized the daily necessity of sophisticated AI for her agents to counter "highly capable malicious cyber actors." She warned that if digital adversaries possess more advanced AI models than Australia, they could identify and exploit vulnerabilities long before Australia can detect them. Bradshaw also revealed a critical vulnerability: Australia's top cyber-intelligence agency is "entirely reliant" on US-housed processing, a dependence that would "dramatically" hinder its capabilities if severed during a conflict.

Recognizing this critical dependence, the defence department's chief AI officer, Chris Crozier, outlined government initiatives to localize its technological infrastructure.

This includes contracting Google to establish a network of interconnected data centres across Australia, designed to be independent of US control.

Crozier stressed the importance of having all AI compute, applications, data, and decision-making capabilities based within Australia to prevent external interference. Bringing major tech firms into Australia would significantly aid this localization effort.

The Office of AI, situated within the prime minister's department, stated that having these companies operate under Australian law would empower local law enforcement to combat threats more effectively. Strict conditions on access could also ensure Australia benefits from the latest AI tools, gaining early and assured access, detailed technical insights, and better integration with leading experts.

The government is actively developing standards for AI companies operating in Australia. These include requirements for reporting issues and breaches, and potentially reserving computational power for local research. There are also proposals for large data centres to engage with communities, be situated away from residential areas, and support new renewable energy initiatives.

However, one significant hurdle remains: copyright law.

AI developers are hesitant to train models in Australia without substantial reform in this area.

#AIAustralia #TechEthics #NationalSecurity #AndrewHastie