What is the White House executive order on AI?
The White House issued an executive order on Artificial Intelligence on October 30, 2023 and rescinded it on Jan 20, 2025. In addition to introducing measures aimed at fostering US competitiveness and avoiding current harms from AI, this piece of legislation was notable for also mentioning existential risks
A risk of human extinction or the destruction of humanity’s long-term potential.
The executive order does:
-
Acknowledge existential risks.
-
Call for increased research into AI safety
.AI safetyView full definitionA research field about how to prevent risks from advanced artificial intelligence.
-
Direct relevant US government agencies to develop standards for evaluating frontier models, as well as to produce yearly reports on the existential risks posed by AI.
-
Require developers of frontier models to share their safety results and test procedures.
-
Require companies to report training runs over 10^26 FLOP
.FLOPA unit used to quantify computing power. One FLOP represents one calculation with floating-point numbers (numbers with a decimal point). “FLOPs” can be the plural of “FLOP”, but is confusingly also sometimes used interchangeably with “FLOPS”, meaning “FLOP per second”.
-
Emphasize the need for international collaboration.
The executive order does not:
-
Impose restrictions on the types of training runs that can be done.
-
Mandate particular safety procedures.
-
Propose a standard of safety, nor demand that a model meet any safety standard.
-
State that any governing bodies, or regulatory agencies, for AI safety should be created.
Zvi Mowshowitz summarizes it as "a request for government reports on what might be done in the future, plus some very mild reporting requirements imposed exclusively on a few giant corporations". Smaller groups may also start to be affected in the future when compute
Shorthand for “computing power”. It may refer to, for instance, physical infrastructure such as CPUs or GPUs that perform processing, or the amount of processing power needed to train a model.