- AI researcher Roman Yampolsky estimates there is a 99.9% chance that AI will lead to the extinction of humanity.
- He said that so far there are no safe AI models, and future models are unlikely to be bug-free.
- Other AI researchers have been more conservative in their estimates of AI leading to extinction.
Podcaster Rex Friedman said in a recent episode that most AI engineers he's spoken to estimate there's a 1-20% chance that artificial general intelligence will eventually wipe out humanity.
Predictions vary depending on how you ask: for example, a recent survey of 2,700 AI researchers suggested there is only a 5% chance that AI will lead to human extinction.
But Fridman said it's important to talk to people who estimate the likelihood of AI wiping out humanity to be much higher. For example, in a podcast interview published Sunday, AI researcher Roman Yampolsky said there's a 99.9% chance that humanity will be wiped out within the next 100 years.
The AI researcher teaches computer science at the University of Louisville and has just published a book called “AI: Unexplainable, Unpredictable, Uncontrollable.”
He spent over two hours on Fridman's podcast discussing the risks of AI, and his predictions were pretty pessimistic.
Yampolsky said the chances of AI destroying humanity depend on whether humans can create bug-free and highly complex software in the next 100 years, which he said is unlikely because no AI model is completely safe from people trying to make it do things it wasn't designed to do.
“They've already made mistakes,” Yampolsky said. “There have been accidents, there have been jailbreaks. I don't think there is a single large-scale language model today where someone hasn't been able to successfully do something that the developers didn't intend.”
The first few versions of the AI model over the past two years have raised various warnings about potential abuses and misinformation, from deepfakes to create fake pornographic images of female celebrities to AI robocalls impersonating President Biden threatening to influence the election.
Google AI Overviews, based on Google's Gemini AI model, was the latest product release to hit the nail on the head. The new feature in Google Search was meant to provide quick, informative summaries for specific questions that would appear at the top of search results. However, it made headlines by coming up with nonsensical answers, such as suggesting that pizza should be made with glue or stating that no African country's name begins with the letter K.
Yampolsky said that controlling AI requires permanent safeguards: Even if the next few versions of GPT work, the AI will continue to improve, learn, self-correct, and interact with different players, posing existential risks and “one chance,” he said.
Sam Altman, CEO of OpenAI, the developer of ChatGPT, said: was suggested A “regulatory sandbox” where people can try out AI and then regulate it based on “what really goes wrong” and “what really works.”
In 2015, Altman warned (or maybe he was joking) that “AI will probably bring about the end of the world, but it will also create some great companies.”
These days, Altman says it's “all the sci-fi stuff” about AI that keeps him up at night, and that “it's easy to imagine situations where this could go wrong.”
Since ChatGPT took the world by storm in November 2022, various predictions have been made about AI, with many claiming that it could lead to the downfall of humanity.
But Yampolsky also warned that “we can't predict what smarter systems will do.” In the context of AGI, he likened humans to squirrels, saying AI will create things we don't even know exist yet.
But Yampolsky says there are three ranges of outcomes he foresees: the risk that everyone dies, the risk that everyone suffers and wants to die, and the risk that humanity loses purpose entirely.
The last refers to a world in which AI systems are more creative than humans and can do any job. In that reality, it's not clear what contribution humans would make, Yampolsky said, and he expressed concern that AI could start taking over human jobs.
Most in the field agree that AI poses some risks, but believe the likelihood of things ending up catastrophic is low: Elon Musk predicts there is a 10-20% chance that AI will destroy humanity.
Former Google CEO Eric Schmidt has said that the real dangers of AI — cyberattacks and biological attacks — will occur in three to five years. If AI gains free will, Schmidt's solution is simple: humans just have to turn off the AI.
