Defense: US and Great Britain want to test AI together

Both the US and the UK have their own guidelines for AI models. It also means they need to be vetted for safety. Now the two countries want to develop and implement at least one joint experiment to monitor potential risks and test AI.

Advertising

In addition to joint reviews, there should be a close exchange of all information and opportunities to contact each other's experts. “AI continues to evolve rapidly and both governments recognize the need to act now to ensure a common approach to AI security that can keep pace with the technology's evolving risks,” it said. British Government Press Release. A treaty between the United States and Great Britain should not be confined to either country. Further partnerships are planned with countries all over the world.

AI regulations in USA, UK and EU

The joint venture's work also aligns with the necessary work discussed at the AI ​​Security Summit last November. Representatives from several countries, as well as some of the biggest AI providers, gathered there to talk about the opportunities and risks of the technology. The result was the “Bletchley Declaration” signed by 28 countries except the United States, Great Britain, European countries, Saudi Arabia and China. The primary purpose of the declaration is to identify safety hazards that have social impacts.

In the US, President Joe Biden has issued an executive order aimed at regulating AI. However, it was initially aimed at federal officials who needed to test AI models. To do this, they must first develop tests and standards and define potential risks. Great Britain has launched several AI projects. That includes it Regulatory authorities are requested to submit commentsHow to deal with AI – It's still April. For example, the Data Protection Authority should present what data protection looks like in AI systems and what needs to be taken into account.

European AI legislation was recently rejected by the EU Parliament. In some cases, it already provides some specific terms. Some particularly dangerous AI systems may not be used, for example Social Scores, i.e. systematic tracking of citizens using AI. Other systems may only be used in exceptional cases, such as real-time biometric surveillance, when national security is at risk. On the other hand, so-called general purpose AI (GPAI), is less restrictive. But: even here, standards and monitoring mechanisms are still being developed. What the risks of GPAI are are not yet decisively negotiated.

(emw)