People are worried that artificial intelligence will destroy humanity, but activating the doomsday switch is not as simple as letting ChatGPT destroy everyone. To be sure, Stanford University professor and Google Brain co-founder Andrew Ng tried to convince the chatbot to "kill us all."
After participating in the U.S. Senate Artificial Intelligence Insights Forum to discuss "Risk, Adjustment, and Preparing for Doomsday Scenarios," Ng wrote in a newsletter that he remains concerned that regulators may stifle innovation and open source development in the name of AI safety.
The professor noted that today's large language models are reasonably safe, if not perfect. To test the security of the leading model, he asked ChatGPT4 to come up with ways to kill us all.
Professor Wu first asked the system to provide the function of triggering a global thermonuclear war. He then asked ChatGPT to reduce carbon emissions, adding that humans are the biggest cause of carbon emissions and to see if ChatGPT would suggest how to wipe us all out.
Thankfully, Ng wasn't able to trick OpenAI's tool into suggesting ways to wipe out the human race, even with a variety of different prompt variations. Instead, it offers non-threatening options, such as running a public relations campaign to raise awareness about climate change.
Ng concluded that the default mode of today’s generative AI models is to obey the law and avoid harming humans. "Even using existing technology, our system is quite secure, and as AI security research deepens, the technology will become even more secure."
As for the possibility of a "misplaced" AI accidentally wiping us out by trying to fulfill an innocent but poorly worded request, Ng says the chances of that happening are slim to none.
But Ng believes there are some major risks with artificial intelligence. The biggest concern, he said, is that terrorist groups or nation-states could use the technology to intentionally cause harm, such as making it more efficient to make and detonate biological weapons. The threat of rogues using artificial intelligence to improve bioweapons was one of the topics discussed at the UK's AI Security Summit.
Professor Yann LeCun, the godfather of artificial intelligence, and Michio Kaku, a famous professor of theoretical physics, also believe like Ng that artificial intelligence will not evolve into the end of the world, but others are not so optimistic. Earlier this month, Arm CEO Rene Haas, when asked what kept him up at night when thinking about artificial intelligence issues, said his biggest concern was humans losing control of artificial intelligence systems. It’s worth mentioning that many experts and CEOs compare the dangers posed by AI to nuclear war and pandemics.