False images more real than life, biased algorithms to sort CVs or to carry out large-scale cyberattacks… How to regulate artificial intelligence and its excesses? While the European Parliament adopted, on Thursday in committee, the “AI Act”, the very first major legislation on the subject, the United States and China, the two main digital powers, are also making progress on the subject. But the approaches differ.
Unlike Europe, which wants to become the first continent to regulate AI, the United States remains cautious at this stage. No question for Washington to put a spoke in the wheels, while the country gave birth to OpenAI, the Californian start-up behind ChatGPT. At the beginning of May, the White House certainly reiterated the need for reliable, auditable and secure AI systems, during a meeting with the CEOs of OpenAI, Google, Microsoft and Anthropic. But regulation of AI is still in its infancy, especially at the federal level.
Regulation “by small touches”
At this stage, the White House has agreed on a AI Bill of Rights. Published in 2022, the text details five principles that should guide the construction of artificial intelligence algorithms. But the text is not a law and is not binding. Rather, it summarizes Joe Biden’s position on the risks of AI and is intended as a guide to good practice. This text followed an executive order issued in 2019 under Donald Trump aimed at “maintaining American leadership in AI”.
Despite these two texts, the United States has not yet “created a uniform and coherent federal approach to the risks of AI”, notes the Brookings Institute in a recent report. Federal agencies have not taken up the subject: at the end of 2022, only five of the 41 largest agencies had adopted an “AI plan”, according to Stanford University.
The FTC, the competition watchdog, has already indicated that it will “vigorously” apply current regulations “including in this new market” that is AI. “We must regulate AI,” wrote Lina Khan, the boss of the FTC, in a recent op-ed in the “New York Times”. Otherwise, the problems that have arisen with the Gafa are likely to be repeated, warns the senior official, who calls for making “the right choices”.
However, “the United States will not make a European-style ‘AI Act’, but rather regulation by small touches”, explains Winston Maxwell, director of studies at Telecoms Paris .
A pro-Party AI
Like Europe, China has already developed a regulatory project, with a recent legislative focus on generative AI. As in the United States, the global buzz caused by ChatGPT has reignited the competition between Chinese web giants and tech such as Baidu, Alibaba, Tencent or Huawei.
But Beijing is walking on a tenuous ridge line. The communist regime must leave enough room for its champions to stay in the race against the United States, while ensuring that the AI follows the Party line.
” The project [de texte] insists that AI-generated content must not sabotage state power, incite the overthrow of the socialist system, secession, or the destruction of national unity,” the China Daily ”, one of Beijing’s official media, in a recent article.
At the same time, some cities or provinces, such as Shanghai in 2022, have adopted pro-AI regulations aimed at promoting its adoption in public services and businesses.