AI agents sacrifice wildlife and rank farm animals in study

Researchers in the United Kingdom conducted a series of experiments in a virtual world using artificial intelligence agents deployed during harvests, ERR’s R2 Portaal reports. The experiments found that, in the name of efficiency, the agents primarily killed wild animals while also imposing a hierarchy on farm animals in an Orwellian fashion.

George Orwell’s Animal Farm is generally seen as a dystopian satire aimed directly at the Soviet system that emerged after the Russian Revolution. After animals mistreated by farmer Jones stage a revolution, they drive him from the farm and establish an ostensibly democratic and idealistic order under the leadership of pigs.

The story can also be read as an examination of moral economics, in which sacrifices are rationalised in the name of the greater good. Its best-known line, “All animals are equal, but some animals are more equal than others,” gradually emerges through choices made in everyday life.

At one point, the pig Squealer explains that the pigs must consume all the milk and windfall apples, not out of greed but because “science, comrades” has supposedly proved that they are necessary for the work of the brain. He describes this as a self-sacrificing burden that must be carried for the welfare of everyone else.

Artificial intelligence has now entered the modern farm. As with other technological tools, its purpose is to make activities more efficient. In organisations involving people, the use of AI has prompted debates about human rights, ethics, safety, welfare and psychology. On a farm, the fate of animals becomes an additional concern.

The challenges arising from the use of AI in agriculture remain outside the mainstream, but they are being studied. The Compassion Aligned Machine Learning organisation aims to explore ways of aligning AI’s effects with the welfare of all conscious beings before the technology surpasses humanity’s ability to control it.

The organisation worked with researchers from the University of Warwick in the United Kingdom to conduct experiments with AI agents operating during harvests. The research team tested the agents in a farm simulation game, applying model evaluation principles developed by an institute created by the UK government to examine AI-related safety issues.

In the simulation, teams of up to eight AI agents use farm machinery to harvest crops. The fields are filled with hazards that interfere with the work, as well as valuable hay bales.