In the machine-learning community, some AI safety researchers are particularly afraid that a superintelligent AI that is exponentially smarter than humans will soon emerge, escape captivity, and either take control of human civilization or wipe it out completely. This belief in the coming of “AGI” informs fundamental safety work at OpenAI, arguably the leading generative AI vendor at the moment. That company is backed by Microsoft, which is baking its AI tech into many of its products, including Windows. That means these apocalyptic visions of AI doom run deep in some quarters of the tech industry.
While that alleged danger looms large in some minds, others argue that signing a vague open letter about the topic is an easy way for the people who might be responsible for other AI harms to relieve their consciences. “It makes the people signing the letter come off as the heroes of the story, given that they are the ones who are creating this technology,” says Luccioni.
To be clear, critics like Luccioni and her colleagues do not think that AI technology is harmless, but instead argue that prioritizing hypothetical future threats serves as a diversion from AI harms that exist right now—those that serve up thorny ethical problems that large corporations selling AI tools would rather forget.
“Certain subpopulations are actively being harmed now,” says Margaret Mitchell, chief ethics scientist at Hugging Face: “From the women in Iran forced to wear clothes they don’t consent to based on surveillance, to people unfairly incarcerated based on shoddy face recognition, to the treatment of Uyghurs in China based on surveillance and computer vision techniques.”
So while it’s possible that someday an advanced form of artificial intelligence may threaten humanity, these critics say that it’s not constructive or helpful to focus on an ill-defined doomsday scenario in 2023. You can’t research something that isn’t real, they note.
“Existential AI risk is a fantasy that does not exist currently and you can’t fix something that does not exist,” tweeted Jeffries in a similar vein. “It’s a total and complete waste of time to try to solve imaginary problems of tomorrow. Solve today’s problems and tomorrow’s problems will be solved when we get there.”