For the past few years, artificial intelligence, usually referred to as AI, has been making major waves across the globe. Some love it, claiming it will open up new frontiers for business and automation, leaving the menial jobs to the robots. Others fear and hate it, claiming it is going to take over jobs and calling it out for replacing artist’s work.
And now, there’s a new concern floating around. Terrifyingly, some experts think that there is a small but significant chance that AI could actually make humanity go extinct. But is that mere speculation, or an actual concern we need to be worried about?
The findings in questions come from a 2023 survey of 2700 AI researchers who have published work at top conferences, and it is the largest compilation of such information. The survey asked participants about technological milestones, and the impact it will have on society, either good or bad. Nearly 58% of researchers said they consider there to be a 5% chance that AI could either cause human extinction or other catastrophic consequences for humanity.
“It’s an important signal that most AI researchers don’t find it strongly implausible that advanced AI destroys humanity,” says Katja Grace of the Machine Intelligence Research Institute in California, who authored this study. “I think this general belief in a non-miniscule risk is much more telling than the exact percentage risk.”
Yep, this sounds very concerning to us, but not all experts are worried. Émile Torres of Case Western Reserve University in Ohio claims that many AI experts, in their words, “don’t have a good track record” when it comes to making AI predictions.They also said that AI researchers are by no means experts in forecasting the future of AI, and points to this 2016 survey that she says did a “fairly good job of forecasting” what things could look like for AI in the future.
Compared with the same survey when completed in 2022, many AI researchers made wrong predictions. They predicted AI would hit certain milestones earlier than previously predicted. This was likely spurred on by the release of ChatGPT in 2022 and the rush to make programs like that available once there was high demand.
The researchers from the survey predicted that within the next decade, there is a 50% or greater chance that AI will be able to tackle most of the 39 sample tasks that have been laid before them, including perfectly copying a pop song and coding a payment processing site. These are very different and complex tasks, but the researchers claimed that other things, like solving math problems that have plagued mathematicians for years or installing electricity in a home, could take longer.
The 2016 survey further predicted that there was a 50% chance AI would be able to outperform humans on every task by 2047, and that there was a 50% chance that human jobs could become fully automated by 2116. This was 13 and 48 years, respectively, earlier than the 2023 survey predicted.
This, of course, speaks to the overall fear that AI will take our jobs and make humans obsolete, which seems to be the biggest concern folk have right now. But again, Torres claims that a lot of these predictions should be taken with a grain of salt.
“A lot of these breakthroughs are pretty unpredictable, and it’s entirely possible that the field of AI goes through another winter,” they say, referencing what happened in the 1970s and ‘80s. During this time, funding and corporate interest dried up, as the technology was seemingly stalled. If advances don’t happen fast enough, either in line with predictions or what people think AI should be, that could happen again.
Researchers also warn that, while the far-off threat of human annihilation may seem like the scariest thing to consider, there are more immediate worries to keep in mind. Over 70% of AI researchers said that more immediate threats are deepfakes, manipulation of public opinion, engineered weapons, authoritarian control of populations, and worsening economic inequality.
The weapons and control sound terrifying, but we’re already starting to see issues with deepfakes and public opinion, so those threats seem very spot-on. Torres especially points out that AI can contribute to misinformation around issues like politics or climate change, which also seems like a more immediate thing to be concerned with.
“We already have the technology, here and now, that could seriously undermine [the US] democracy,” Torres adds. “We’ll see what happens in the 2024 election.”
In this sense, Torres seems spot-on. We are already seeing concern over the 2024 elections and what they will bring out in our society, so the added fuel of AI misinformation definitely does not seem positive.
So, the consensus seems to be that the future could be scary, but the AI threats are likely more immediate and mundane, rather than some far-off dystopian future scenario.