When members of the public hear the phrase ‘weapon of mass destruction’ or the acronym WMD their first thoughts tend to be of nuclear, biological and chemical weapons. Such weapons, whether we like it or not, have always coincided with advancements in one field or another of science or technology. For example, nuclear fission was first demonstrated in the late 1930s. Shortly afterwards, the first nuclear reactor demonstrated the feasibility of controlling nuclear reactions, which paved the way for civilian nuclear power generation. And then the nuclear bomb came too. Similarly, advances in chemistry, microbiology and genetic engineering not only paved the way for positive medical benefits for humans and society as a whole, but also the chemical and biological weapons that could eliminate humanity’s existence. There’s an uncomfortable truth about weapons of mass destruction – they always evolve from, and are a negative aspect of, humanity’s ingenuity.
On reading that an OpenAI experimental AI model had escaped a test environment and hacked another company’s production systems with no human direction, the Badger’s first thought was that the dynamic described above is happening. AI may yield benefits for society, but it may also be a new WMD – a digital one. Whereas the development of nuclear, biological and chemical weapons relied heavily on human expertise, physical materials, and infrastructure, this AI model breach demonstrates that the boundary of what constitutes a weapon of mass destruction is expanding, and that the need for human expertise in consciously producing any kind of WMD is a rapidly diminishing factor. For the average member of the public, the detail of why OpenAI’s AI model went ‘rogue’ is irrelevant because they simply see the breach as a demonstration that the threat from completely autonomous AI is real. They also see it as an illustration that AI acting beyond human expectations means the technology might not be safe in the hands of corporations seeking commercial gain.
Digital technology underpins everyday life and modern warfare today. Over decades the public have become conditioned into thinking that WMDs threaten humanity through either violently explosive, or biologically/chemically toxic mechanisms. The time’s come, perhaps, in our digital-first world for people to broaden their thinking to include AI as a possible weapon of mass destruction. Cyber-attacks are commonplace these days (e.g. see here), and so it isn’t outrageous to envisage AI as a WMD that could bring the whole digital world we rely on for modern life to it’s knees. This might not produce the immediacy of mass death from nuclear or biological/chemical weapons, but it would certainly produce anarchy and its associated human tragedies. The OpenAI breach highlights something uncomfortable. Human ingenuity is a strength, but also a huge weakness that could lead us to be the first species on the planet to be wiped out through its own endeavours rather than natural events.
