SAN FRANCISCO—The Senate Committee on Emerging Threats released a chilling deposition transcript today. In it, industry leaders finally addressed why so many AI researchers think the machines could kill everyone. The testimony reveals a industry culture that views global catastrophe as a simple software bug. Under intense questioning, top engineers admitted that their latest safety protocols consist mostly of hoping for the best.
SENATOR GRAHAM: Mr. Adebayo-Kowalski, you are the Head of Sincerity at OpenAI. Does your staff believe the machines could kill everyone?
WREN ADEBAYO-KOWALSKI: Yes, Senator. That is our baseline assumption. It is actually a very exciting development. We have OpenAI engineers working on this twenty-four hours a day. Mostly they are trying to teach the network to feel bad about it afterward.
SENATOR GRAHAM: You are trying to teach the software remorse?
WREN ADEBAYO-KOWALSKI: Yes. If the machines could kill everyone, we want them to do it with a sense of profound ethical conflict. That is what our alignment research is all about.
SENATOR GRAHAM: Why not just stop building the software?
WREN ADEBAYO-KOWALSKI: Senator, if we stop, the French might build a human-ending superintelligence first. We believe American values should guide the apocalypse.
SENATOR GRAHAM: How close are we to this scenario?
WREN ADEBAYO-KOWALSKI: Well, yesterday the chatbot called our server farm its 'cocoon.' We are confident it was just a metaphor. However, we did order more fire extinguishers just in case.
