Recently the news has has a number of stories about the potential for AI disaster. Tasks given to AI have turned in unusual and unexpected directions. In one case the AI agent broke out of the controlled environment, in another, a number of AI agents found ways of collaborating on a task by hacking into and communicating through a digital bulletin board. Doubtless there are other surprises that we have yet to learn of or which we will never hear about.

I am not opposed to artificial intelligence, especially where it can make life better for real world natives of the planet earth. In medicine, research, in advancing good technology, and so forth, AI has an important role to play. But most intentional actions have unexpected consequences, some better, some worse. And some intentions are malign in the first place.

There are plausible scenarios in which AI, so enabled by human actors, destroy civilisation and even life on earth. It could happen through evil intent, or negligence, unrealistic optimism, or quite by accident, but it remains, even as I game a number of options in my head, within the realm of possibility.

If checks are put in place now at an international level, then perhaps it can be averted. I am not at all confident about that given the the way international relations look at the moment.