You normally don’t have to look very far to find someone denouncing artificial intelligence or buying into a robot-laden, dystopian future. For me, it’s that weirdo I went to high school with who dropped out of college after a semester. He is super well-informed. And, Facebook is the perfect place to share his groundbreaking, uh, theories.
OK, maybe the people who fear AI aren’t all college dropouts. In fact, Elon Musk has been quoted as claiming he and his fellow tech elitists might be creating their own means of destruction.
It’s hard to know how to feel about artificial intelligence. On one hand, just in the accounting profession it has massive potential to free CPAs to do more meaningful work. On another, if a robot can replace you (and do what you do in half the time with zero errors), why wouldn’t your employer invest in that, or at least want to learn more? And, on a theoretical third hand, AI is flat-out scary.
I mean, c’mon. Has no one seen Black Mirror?? I know what the potential is here.
Well, in typical human form, we’ve decided the best way to combat that third scenario is to have AI systems duke it out with each other. Like Fight Club.
And, unlike Fight Club, we definitely need to talk about this.
The concept, according to MIT Technology Review, comes from researchers at OpenAI, a nonprofit founded by Silicon Valley’s who’s who, including Elon Musk.
Research shows AI systems can sometimes develop unexpected and unwanted habits. Look no further than Microsoft’s AI chatbot that, in less than 24 hours, began tweeting that “Hitler was right” back in 2016. Yikes. And, it’s not always possible or plausible for humans to supervise the training of an AI program, which is where many of these issues are introduced.
This innovative approach would allow AI to debate with one another until it’s clear which one is correct. You can test out what this might look like here.
Read more at MIT Technology Review and let us know in the comments whether you think this solution, in your opinion, is worth pursuing.