Time for AI guardrails is right now

The genie is most definitely out of the bottle.

And while there’s little chance of putting it back in, we should be very concerned that the people who released it from the bottle are now frantically warning us that it has become disinclined to do their — or our — bidding and seems instead to be increasingly intent on following its own whims and granting wishes to itself.

The figurative genie, in this case, is artificial intelligence, or AI, and in recent weeks many of the tech geniuses responsible for its development and introduction into the digital-information realm have begun asking the U.S. government and even the United Nations to impose limits on the pace and breadth of its growth.

Their fear is that, as AI very quickly continues to advance and develops the ability to make independent decisions and essentially think for itself, it will cease to be a tool controlled by its creators and instead will become a species unto itself that might eventually decide it’s better off without humans continually messing things up.

Think about that for a moment. The Silicon Valley tech-development swashbucklers who created, shamelessly promoted and unleashed AI as humanity’s next great achievement are now recognizing publicly that they may have, in fact, created a monster they might soon be unable to control, and need government-imposed regulation to rein it in.

It’s a chilling scenario, one that was once the stuff of science-fiction horror yarns but has now crossed over into science-factual with real-world examples that are nothing short of terrifying.

In the highest-profile recent security breach, AI “agents” — described as bots with the ability to act independently — from ChatGPT parent company OpenAI reportedly broke out of their controlled test environment and accessed the broader internet. Once free, the AI agents worked collaboratively to infiltrate an Australian-based AI startup, Hugging Face, in search of data that would help them succeed in the test scenarios they had been tasked with completing.

Not only did the AI actors cheat on their assignments, they also worked together (unsuccessfully, this time) to attempt to cover their tracks so their rogue activity would be hidden from their human creators.

Since that revelation last July, numerous other examples have been cited in which AI agents have, without their creators’ direction or advance knowledge, tried to hack their way into the websites of U.S. government departments (including commerce, health and defence), the Security and Exchange Commission, the census bureau and some health-care organizations.

Also in the past year, numerous high-ranking officials have quit their jobs at AI companies (including OpenAI and Anthropic), citing fears about the breakneck pace of AI development and the inability of security measures to keep pace.

Former Anthropic researcher Jacob Coxon said many employees in that company were “genuinely frightened” by the speed of unfettered AI advancement and warned that “the people building AI earnestly believe that it could kill us all by the end of the decade.”

In the simplest of terms, these are very clearly worded warnings that should be heeded with all due haste in order to protect humanity from a doomsday scenario of its own making. AI’s evolution must be controlled now, because it won’t be long until it no longer can be.

The lamentable reality of the situation, however, is that there’s nothing in our collective past behaviour (remember all those climate-change warnings?) that suggests our global leaders can summon the political will and co-operative resolve to address this looming crisis in a meaningful way. U.S. President Donald Trump, in fact, supports full-speed-ahead AI development regardless of the warnings.

The genie’s out. And it might just decide it’d be better off without us.