Sanders and Casar file a bill to ban artificial superintelligence
Bernie Sanders and Greg Casar introduced the Ban Artificial Superintelligence Act on 23 September. The bill would prohibit systems defined as capable of destroying or disempowering humanity, and it sets a maximum penalty of 20 years in prison. Advanced-model work stays paused until a new Department of Artificial Intelligence writes the rules.
Artificial Intelligence··Night
The bill names what it would ban
Bernie Sanders and Greg Casar introduced the Ban Artificial Superintelligence Act on 23 September. The bill would prohibit development of systems it defines as capable of destroying or disempowering humanity. The Verge reports that the definition includes overthrowing the government. The same filing, in the account from Sanders's Senate office, would ban development of artificial superintelligence. The two accounts name one bill and one threshold: a system whose capacity is the destruction or the disempowerment of humanity, not a routine model update. Advanced models face a pause tied to a training-data threshold. Work resumes only with permission from a scientist-led Department of Artificial Intelligence. That is the shape of the prohibition both the news report and the sponsor's release describe, and it is the reason the bill is a ban rather than a labelling rule.[1], [2]
A new department would hold the pause
Advanced-model work stays paused until a new Department of Artificial Intelligence writes the rules. Sanders's office says that pause runs until the new federal department has safety rules in place, and that the bill would create a cabinet-level Department of Artificial Intelligence. The Verge adds the department's further duties: it would oversee removal of features judged dangerous and supervise the destruction of superintelligence. The pause is therefore not a study commission. It is a stop on advanced-model work, tied to a training-data threshold, that lifts only when that department gives permission. The sponsor's release and the news report agree on the department as the body that writes the rules before the work continues. They also agree that the bill's subject is artificial superintelligence, the class the text defines by the capacity to destroy or disempower humanity.[1], [2]
The ceiling is 20 years in prison
The bill sets a maximum penalty of 20 years in prison. The Verge reports that ceiling for a company leader or for a person acting outside a company, and says the bill's one-pager compares it to penalties for unlawfully developing nuclear weapons. Sanders's office states the same ceiling, up to 20 years in prison for violators. In the press release, as The Verge reports it, Sanders says easing off the gas is not enough when the race is toward a cliff. The penalty and the pause sit on the same text: development of the defined systems is prohibited, advanced-model work waits on the new department, and the maximum sentence is 20 years. The comparison to nuclear-weapons penalties is the bill's own one-pager, relayed by The Verge, not a separate criminal statute already in force.[1], [2]
