Scripture warns against false prophets. The Gospel of Matthew says that no one but God, “not even the angels of heaven,” knows the day or the hour of the end times. That’s worth remembering amid the flood of predictions about an artificial intelligence apocalypse.
This week saw two prominent researchers at frontier lab Anthropic casually throwing around probabilities for humanity’s extinction at the hands of runaway AI. Evan Hubinger, who leads the company’s alignment research, estimated its probability within the next decade to be greater than 10 percent. Jacob Coxon, who resigned from Anthropic this week, accused the leading labs of “gambling with our lives.”
These warnings have received outsize attention. Who can blame the public, already uncertain about the technology’s impact on their lives, for taking them seriously?
The forecasts are concerning, but they merit a certain skepticism. People working at the frontier may know more than most about what today’s models can do. But that does not give them special knowledge of what machines that do not yet exist will someday become — or how society will adapt to the challenge.
Unfortunately, politicians in both parties are starting to treat these fears as grounds for designing public policy. Democratic socialist Sen. Bernie Sanders (I-Vermont) and Rep. Greg Casar (D-Texas), chair of the Congressional Progressive Caucus, want to pause advanced AI development and permanently ban “artificial superintelligence” — a term that their proposed legislation defines with chilling sweep.
Under their plan, systems that merely match human cognitive performance across a broad range of tasks would qualify. A new Cabinet-level regulator would be empowered to police the technology. Violating the ban could mean up to 20 years in prison for individuals and dissolution of the offending company.
The dawn of the AI era cries out for humility and patience. Politicians should not seek to regulate a fast-moving technology they don’t understand. Bad legislation risks undermining the many potential upsides of AI, such as developing breakthrough medicines.
Silicon Valley’s doomers are pointing policymakers in the wrong direction — and in a way that happens to let the AI companies off the hook. Scientists at the frontier labs talk about their products as an otherworldly force — “an alien mind,” in the words of OpenAI’s chief scientist, Jakub Pachocki. This invites an all-of-humanity response to avert the worst-case scenarios — from suffocating domestic legislation to fanciful global treaties.
These scientists might instead fret about the fate of their employers, who may be held liable for damages caused by their inventions. In an essay published this past weekend, Pachocki admitted that building AI is more like experimentation than engineering. These researchers are not building complex missiles; they are playing with what might be the digital equivalent of nitroglycerin. When a lab blows up and takes out a city block, do we blame the chemicals?
Of course not. The legal system would ask who handled them, what precautions the labs took, what warnings were ignored and who decided the experiment should proceed.
Those are exactly the kinds of questions that Alabama Attorney General Steve Marshall (R) has begun asking. Last month, his office subpoenaed OpenAI for records surrounding the Hugging Face lab leak, asking about safeguards surrounding its testing and previous unauthorized intrusions. Unlike prophecy, this is a process that can produce evidence.
The prospect of discovery, injunctions and damages ought to make AI companies more cautious about preventing their next experiments from going awry. Liability offers better discipline than overbearing legislation, and it’s much more likely to yield results than negotiating unenforceable treaties with the Chinese Communist Party.
Silicon Valley’s prophets of doom want government to fear AI enough to prohibit an unknowable future. The better answer is to make AI companies fear the consequences of behaving recklessly in the present.
The post AI hysteria jumps the shark appeared first on Washington Post.
