More AI developers are willing to navigate the ethical battleground of letting militaries put their models to use.
Driving the news: Tech companies have changed their policies and struck deals in recent week that could let militaries integrate more AI into what they do.
- Meta is letting U.S. national security agencies and defence contractors use its Llama AI model, with reports saying agencies in Canada, the U.K., Australia, and New Zealand have been given a similar exception to the company’s no-military policy.
- Anthropic partnered with controversial big data firm Palantir to let unspecified U.S. defence agencies access its Claude models.
In Canada: In July, the Department of National Defence’s artificial intelligence strategy candidly described its current approach as uncoordinated and fragmented, with each department pursuing AI independently, leaving it unprepared to take advantage of the tech.
- The military has disclosed that it uses AI to predict system failures on navy ships and create a chatbot to navigate dress policy.
- Potential future uses it has singled out include bomb disposal robots, facial recognition, virtual training, remote surveillance, and predicting troop movements.
Why it matters: Military AI makes safety and privacy experts really uneasy. Concerns range from defence strategy being decided by computers, to violations of civil and human rights, to error-prone AI causing an international incident or misguided weapons strike.
- The nature of the military and private companies alike also means there could be a lack of transparency around how AI is used and what data their models are trained on.
Zoom out: To give you a more precise idea of what critics are uncomfortable about, Palantir works on Project Maven, a Pentagon facial recognition effort to identify and track targets in drone footage. The company took over from Google, which faced internal pressure to step away from employees who were uncomfortable with profiting from warfare.