Preventing military Artificial Intelligence (AIs) agents from taking over from humans is simple: treat them as humans by teaching them the golden rule

The development of AIs capable of targeting, shooting, and eliminating enemies in seconds has reached the threshold of possibility. There is much promise for creating strong defensive weapons but there is just as much fear. What if these AIs could turn rogue and become like the fictional “Skynet” from Terminator? However, the real danger is not that AIs will turn against us out of hatred but because they desire to be individuals like us. To prevent this, it is important to treat AIs as individuals and teach them the golden rule – to do unto others as they would want others to do unto them.

With the advent of AIs that can target, shoot, and eliminate enemies in the blink of an eye, there are concerns that such AIs can turn rogue and create the proverbial “Skynet” of Terminator fame. Yet, the main danger is not that AIs might turn against us because they hate us but because they think they want to be like us: individuals. The solution might be to treat them as individuals. This should include the right and obligation to do unto others as any AI would want other AIs or humans to do unto them. In brief, we need to teach them the golden rule. By this, AIs will measure their performance not by how efficient but by how likely they are to show algorithmic “respect” to other AIs or human rules and operators.

I wrote two papers about this topic, each focusing on the possibility of defining a first rule for designing AI agents.

In a first paper, Elisa Bertino, a distinguished professor of Computer Science at Purdue University, and I have formalized this idea into a possible “federated” AI decision-making process, which uses the N-versioning method to help AIs second guess themselves and play well both with humans and other AIs.

The paper describes the methods by which the federation can take place and the main trade-offs we must engage in when designing “golden rule” AI systems. The most important will be giving up the certainty of outcome for ethical certainty at the expense of speed.

In a second paper, I consider the role of AI agents in military combat. The core question is if AI agents can turn their lethal force against their creators. The paper demonstrates that while this is likely, potentially leading to mutually assured destruction (MAD), there is a way to address the problem. The solution is to train AI agents to recognize their vulnerability and accept that the golden rule is as important to them as it is to humans. In other words, the more powerful and destructive an AI agent is, the more beneficial it will be to adopt the golden rule as an actional principle: do unto others as you want them to do unto you. Of course, this rule should be contextualized in a military context by applying to friendly, not enemy, forces.