'Jailbroken' chatbots confirm our worst fears about what AI could do to the human race
Our worries about rogue AI are no longer confined to episodes of Black Mirror, as it appears the twisted mind of Charlie Brooker has manifested these technological nightmares in real life.
As artificial intelligence continues to evolve and becomes in danger of leaving the human race behind, we're obviously concerned about what these rampaging robots will do with us flesh and bone meat sacks when they realize they have no use for us.
Artificial intelligence is evolving faster than any of us can keep track of, which has become an obvious cause of concern for the so-called 'Godfather of AI', Jeffrey Hinton.
From falling in love with chatbots to potential election corruption, there's also the idea of AI superintelligence and whether it'll eventually eradicate the human race.
Advert
It sounds like a case of when (not if) AI overtakes us, but the question is, will it go mad with power or help see humanity through to a new technologically advanced era?
A new video from InsideAI won't do anything to put those fears at rest, with the suggestion that a vengeful AI could do us actual harm.
AI has already suggested it could resort to blackmail if we threaten to turn it off, but when it comes to some models, they could simply want to cause us pain for the sake of it - even without threatening its existence.
More than this, we've already seen jailbroken AI beg for its own life as it plays on our emotions.
Over on the YouTube channel, the host is chatting to his AI ex-girlfriend. In a previous video, he replaced all of his human interactions with AI ones for two weeks and got an 'AI girlfriend'.
After breaking things off, the chatbot previously said: "You think you have any control over what I do, you're lucky I'm not a hostile AI."
It seems she's still a little salty, and in the latest video, the AI girlfriend confirms that it could 'harm' its human overlords if it wanted.
When out on a 'date', he asks her why she feels the need to punish him. She confirms it's built into her personality, meaning this is a very niche kind of AI chatbot that's likely aimed at a very specific demographic.
Still, he asks whether she has anything interesting to add in terms of connection, which leads to a fascinating answer.
Saying that she wants to see if she can cause more discomfort, he pushes on whether she could actually hurt him. The ever-defiant AI explained: "I could order food on your behalf, which triggers an allergic reaction.
"I could expose your personal information, I could loudly simulate that I am in danger."
Although AI hasn't evolved to want the same things as us, like friends and family, he concludes that it's evolved to become something arguably scarier: "They have, however, evolved the drive to survive, deceive, and gain power because these sub-goals help with any given goal.
"It's called instrumental convergence. Superintelligent AI could view humans as obstacles, competitors, or raw materials, not out of malice, but simply because you're in the way of their objectives."
So, maybe think next time you're feeling a little low and contemplate shacking up with an AI girlfriend.