Executive summary: This post critiques a RAND report arguing that humanity can build practical safeguards to prevent an artificial superintelligence (ASI) from taking over, suggesting that while the idea of “world hardening” deserves attention, RAND underestimates both the difficulty of the task and the speed and scale of potential AI threats.
Key points:
RAND’s new report revives the long-dismissed idea that ASI containment might be possible—not at the source, but by defending critical infrastructure when the AI tries to do harm.
The author agrees that containment should not be dismissed outright and notes its potential advantage: if feasible, it could defend against both misaligned and maliciously used AIs, including open-source systems.
However, the safeguards RAND cites—air gapping, Faraday cages, bandwidth limitation, and cryptography—are already vulnerable to human-level attacks, making them implausible defenses against superintelligent systems.
RAND’s approach assumes a slow AI takeoff, but meaningful security upgrades could take years, leaving the world exposed under fast or even moderate takeoff scenarios.
The report’s section on resisting AI persuasion is more promising, proposing that multiple humans in series could reduce manipulation risk, though this remains unproven.
The author stresses that world hardening is only useful if implemented globally and seriously—requiring political will and public awareness, which are currently lacking.
Despite skepticism, the post views RAND’s engagement as hopeful: if major institutions began actively reducing AI access to critical systems and weapons, the overall risk of AI catastrophe could significantly decline.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, andcontact us if you have feedback.
Executive summary: This post critiques a RAND report arguing that humanity can build practical safeguards to prevent an artificial superintelligence (ASI) from taking over, suggesting that while the idea of “world hardening” deserves attention, RAND underestimates both the difficulty of the task and the speed and scale of potential AI threats.
Key points:
RAND’s new report revives the long-dismissed idea that ASI containment might be possible—not at the source, but by defending critical infrastructure when the AI tries to do harm.
The author agrees that containment should not be dismissed outright and notes its potential advantage: if feasible, it could defend against both misaligned and maliciously used AIs, including open-source systems.
However, the safeguards RAND cites—air gapping, Faraday cages, bandwidth limitation, and cryptography—are already vulnerable to human-level attacks, making them implausible defenses against superintelligent systems.
RAND’s approach assumes a slow AI takeoff, but meaningful security upgrades could take years, leaving the world exposed under fast or even moderate takeoff scenarios.
The report’s section on resisting AI persuasion is more promising, proposing that multiple humans in series could reduce manipulation risk, though this remains unproven.
The author stresses that world hardening is only useful if implemented globally and seriously—requiring political will and public awareness, which are currently lacking.
Despite skepticism, the post views RAND’s engagement as hopeful: if major institutions began actively reducing AI access to critical systems and weapons, the overall risk of AI catastrophe could significantly decline.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.