On October 31, the White House released the full text of a long awaited Executive Order (EO) on Artificial Intelligence. The EO comes as the United Kingdom kicks off its AI Safety Summit this week with a lot of major players, and the European Union scrambles to get its AI Act done by the end of the year. The EO marks the most significant US action on AI to date.
The EO has more teeth than any previous efforts from the Executive Branch, but it also seems to intentionally avoid dramatic constraints on potentially innovative applications and appears to have limited impacts on the current generation of popular generative AI models like OpenAI, Claude, Bard, and others. The 20,000+ word EO covers a sprawling range of big ticket items such as safety and security, privacy, competition, equity, and workforce.
Perhaps the most concrete development is the invocation of the Defense Production Act to require new large language models trained on over 10^26 mathematical operations or 100 trillion trillion operations (OpenAI’s GPT-4 is currently based on 20 trillion trillion) to submit the results of their red-team safety exercises to the federal government. There are also additional, more stringent guidelines related to biological synthesis projects to help protect against bad actors using AI for bioterrorism.
Reactions in the US to the EO have been neutral to positive. Some have praised its broadness and thoughtfulness. Others have noted that implementation will be the true test of impact. Others, including technology trade association NetChoice have called it a “back-door regulatory scheme” that could stifle innovation.
A Global View…
Although President Biden’s EO is the most significant action taken in the US to date, it is one of many major efforts happening globally, including the EU AI Act, the UK AI Summit, the G7 Hiroshima AI statement, and the UN Advisory Body on AI.
Further, it remains just an executive order. That means, for the most part, it stays in force only as long as the sitting president chooses to keep it in force. Without Congressional action to create a federal AI law, this executive action on AI could be revoked or changed with a change in administration. And with Senator Schumer continuing to lead a series of AI fora, meaningful, full-scale Congressional legislation is not expected in the immediate term.
That being said, some are calling the EO an important point of reassurance to the EU, which is in the final stages of negotiating its sweeping AI Act. The holdup for the AI Act reportedly is stemming from member country concerns that it could curtail homegrown innovation while giving U.S. companies a leg up. The EO is an indication that the US is taking regulation of AI seriously.