Experts from various fields – ranging from data security to social studies – agree that AI regulation is needed in order to hold AI systems accountable. So far, the USA did not participate in the Global Partnership on AI – a project of the G7 nations that aims at establishing AI regulations. In an “America first” approach the USA want to ensure that they remain a global hub of innovation and that technological advances are consistent with their values. Thus, instead of an international agreement, the US government aims for national AI standards. For this reason, a set of 10 principles that agencies should consider when devising laws and rules for the use of artificial intelligence in the private sector was released by the US government last week. These principles can be divided into three areas: public trust and transparency, fairness and risk management, and progress and safety.
Public trust and transparency
Public trust in AI needs to be ensured by promoting reliable and robust, thus trustworthy AI applications. Moreover, transparency regarding AI applications and the disclosure of information should be used to strengthen this trust. For this reason, the public should also be included in the process when defining AI rulesets.
Fairness and risk management
AI applications should aim for fairness and non-discrimination. Societal benefits and distributional effects should thus be considered when defining rulesets regarding AI. Moreover, a risk-based approach should be used to determine which AI-based risks are acceptable and whether the expected benefits outweigh the expected harm.
Progress and safety
Regulations need to adapt to the fast-paced changes and updates of AI applications and data security (confidentiality, integrity and availability) needs to be ensured for data that is processed, stored or transmitted using AI systems. Agencies are supposed to coordinate AI-related policies to ensure consistency and to share experience.
While these are valid points, the principles face criticism as they are vague and act as recommendations rather than binding rules. Martijn Rasser, a senior fellow at the Center for New American Security said: “Anything that an agency produces could be shot down, given the vagueness.” Moreover, the White House’s Office of Science and Technology Policy further stressed that they want to limit regulatory “overreach” and urged other nations to follow this example they set. Regarding the strict data protection regulations, the EU has put in place it is safe to assume that the EU nations will aim for stricter regulations that could include a demand for explainablility of AI systems or human in the loop safety mechanisms.
Sources:
[1] aitrends | White House Releases 10 AI Principles for Agencies to Follow
[2] THE VERGE| White House encourages hands-off approach to AI regulation