Do half of AI experts think that there is a 10% chance the technology will kill off humanity? And does it matter?
The claim had appeared widely in publications that might consider reviewing a survey beyond identifying a clicky purported result. Santa Fe Institute Professor Melanie Mitchell looked more deeply into the matter than the journalists had. The 2022 survey contacted 4271 people, of whom 738 responded. Of those, 162 answered the question, “What probability do you put on human inability to control future advanced AI systems causing human extinction or similarly permanent and severe disempowerment of the human species?” Of them, 81 answered 10% or higher.
Mitchell went on to say that while not an expert in surveys, there were some obvious problems, like not mentioning a timeline for extinction and the—as I am putting it—the likelihood that the sample is not statistically representative of AI researchers because of self-selected bias. I’d also add the question of whether any of the people has any particular expertise in social, psychological, political, economic, or other areas that would be needed to consider the survey question. What expertise would they have that might give them insight into an answer, other than they are involved in AI? It would be like asking whether people designing automobiles in the 1950s could have predicted climate change from carbon emissions.
Now for the thornier second question: Does it matter even if the study is deeply flawed? Yes, it does.
Humans have thousands of years of history that shows innovation and invention that turned into terror and slaughter, sometimes regretting it afterwards. The story that Alfred Nobel created the eponymous Peace Prize after seeing the results of inventing dynamite or multiple physicists on the Manhattan Project who felt as though they should not have helped enable such a weapon.
How many chemists looked back with dismay about products that became environmental scourges? Mike Tompsett, who helped develop color digital photography, reportedly regrets the selfie. Philo Farnsworth, who first transmitted a television image, hated what it eventually became.
Much of this recitation is a list of people who couldn’t see moderately into the future, or who convinced themselves that their work was at least relatively benign. Reflected in history, a poll of AI experts, whether representative or not, in several score thought there was a significant chance of wiping out humanity might be a result should be terrifying. Have these people no moral core or even concept of risk? Do they not realize that elimination of all humankind means them as well as everyone they hold dear? Would you, in a similar position, move along, hoping that things would work out?
That is the true danger that appears repeatedly in social groups, governments, businesses—that those in various degrees of power possess the ability to gamble with the fate of others and then the willingness to carry out the actions.
How often is war truly necessary? Rarely. Dangerous products reach distribution to consumers because of the cost for proper testing and consideration when there’s profit to be had in the short turn. Let a company run by someone known for being tempestuous and impulsive test a rocket without greater attention to safety. Move fast and break things, the rallying cry of parts of the tech elite.
It is not new and rather quite old. Julius Caesar overturning the Roman Republic. The horrible history of colonization across the globe, with Africa a continent repeatedly put under the figurative and literal lash. China’s subsumption of Tibet.
Find it in U.S. urban planning. Robert Moses in New York City who may have done some good things but also destroyed neighborhoods sacrificed to his vision of highways. The old West End in Boston, leveled to make way for Government Center. Richard Daley’s influence on Chicago through his building projects.
See it in finance, the desire for boundless profit, and the way historically it has led to war and conquest. In the monopolies of the nineteenth century and industrialism that turned people into fodder in a way removed from feudalism by only a more refined operational method. The Great Depression, a product of recklessness and firm belief by officials in their own certainty. Or the Global Financial Crisis/Great Recession, with companies and individuals making insane choices, figuring they would make money, get away with it, and things would be fine enough. Instead, we had utter collapse and more than a decade of recovery.
Products bad for health and sold by bamboozling marketing campaigns by people with no soul or conscience. The willingness of politicians to prey on division, deepen it to a gulf, all as a play for power.
Those at the top are always willing to gamble with everyone else’s lives. To risk catastrophic damage to people, communities, and the delicate planetary balance of nature so they can achieve what they want, allowing everyone else to pay the price.
These are the dangers: the sociopaths and psychopaths, and the weak people working for them, who don’t care. And it is only society at large that can stop them.