ChatGPT and Advanced AI Face New Regulatory Push in Europe
European legislators to add new provisions for powerful AI to pending bill
PARIS—European Union lawmakers want to give regulators new powers to govern the development of technologies like those behind ChatGPT, the biggest push so far in the West to curb one of the hottest areas in artificial intelligence.
The breakneck pace of AI development in recent months requires a new set of rules tailored to powerful, general-purpose AI tools, a group of influential EU lawmakers say in an open letter they plan to publish Monday.
The group of lawmakers, who have been charged with hammering out a new draft of what the bloc calls its AI Act, say they are committed to adding provisions to the bill aimed at “steering the development of very powerful artificial intelligence in a direction that is human centric, safe and trustworthy,” according to a copy of the letter reviewed by The Wall Street Journal.
“With the rapid evolution of powerful AI, we see the need for significant political attention,” the lawmakers write. They add that the EU’s pending bill, which policy makers aim to pass into law later this year, “could serve as a blueprint for other regulatory initiatives in different regulatory traditions and environments around the world.”
The EU Parliament statement, coming from some of the most powerful voices in crafting the AI Act, adds new momentum to calls from some researchers and technologists for regulators to slow or pause development of very powerful AI tools, in part to allow regulators to catch up.
Late last month, a group of AI researchers and tech executives including Elon Musk signed an open letter from the Future of Life Institute that called for a six-month moratorium on the training of the next generation of AI tools to give time for the regulators and industry to set safety standards. A separate group of AI ethicists and researchers wrote another open letter last week urging the EU to include provisions that cover possible risks from “general purpose AI.”
Dragos Tudorache, a Romanian member of European Parliament who co-leads the body’s work on the AI Act along with Italian member Brando Benifei, said the group crafted Monday’s letter in part in response to the Future of Life Institute’s letter. In addition to calling for a pause in training advanced AI models, that letter described one possible risk from AI as “loss of control of our civilization” and urged regulators to step up work with AI developers.
“We share some of the concerns expressed in this [Institute] letter, even while we disagree with some of its more alarmist statements,” the EU lawmakers said in Monday’s letter. They add that they think that regulation can help humanity reap the benefit of AI and avoid “more challenging future scenarios.”
“Together, we can steer history in the right direction,” they add.
The EU’s AI bill is nearing its final stretch of debate as more regulators globally are jumping into the fray. China’s top internet regulator earlier this month proposed rules to control artificial-intelligence tools similar to ChatGPT.
In the U.S., the Biden administration has begun examining whether checks need to be placed on the tools. Last month, Italy temporarily banned ChatGPT on privacy grounds and the U.K. released a paper suggesting regulators oversee development of AI with a focus on tools’ safety, transparency and fairness.
It isn’t clear yet what new provisions the EU Parliament will propose adding to the AI Act. Monday’s letter says the rules are needed for a type of AI tools called foundation models, which are trained on massive sets of data, and underpin some of the most recent advances in AI. Foundation models include some large language models like those behind tools like ChatGPT from Microsoft Corp. -backed startup OpenAI, which can respond cogently to textual questions.
The signatories of Monday’s letter are set to meet this week to hash out their proposals and agree to a common position, which the full Parliament would vote on in May. In the past, some members of European Parliament have called for much more robust transparency about how such powerful AI tools are built, fairness requirements for the data used to build them and ongoing audits of their safety and predictability.
After the European Parliament agrees on a draft, which it is expected to do next month, it will negotiate with members of the EU’s Council, which represents member states. The Council settled on its own AI Act draft late last year, but its draft left open to the European Commission, the bloc’s executive arm, to determine any specific requirements for general-purpose AI. Both chambers must vote to approve a compromise text for a bill to become a law—something policy makers say they hope to do later this year.
The Swedish government, which holds the Council’s rotating presidency, has called the legislation a priority and said it was prepared to convene discussions as soon as the Parliament has agreed on its position.
The European Commission recognizes the need to consider specific rules for general-purpose AI and will support lawmakers in their efforts, an official said.
Adding any new rules to the AI Act that govern the development of general-purpose AI tools like foundation models would represent a major shift in approach. Until now, the legislation has taken what policy makers call a risk-based approach and saved the toughest rules for what the EU deems to be risky applications of AI, such as a ban on most police use of facial-recognition technology. But new rules for foundation models would apply to such technology no matter to what end use they might be put.
Tech companies and their lobbyists have in the past argued that the law should stick to a risk-based focus on specific AI applications, and not put too many restrictions on AI development because it would impede innovation. But some tech researchers have also joined academics and technologists in expressing support for rules that would slow a race by companies to roll out advanced new AI tools.
Current drafts of the bill would impose fines of up to 6% of a company’s global revenue in case of noncompliance.
Monday’s open letter also included other, broader calls to action from the EU lawmakers. They urged the EU’s executive arm and President Biden to convene a high-level global summit on AI, aimed at agreeing to preliminary governing principles for how to deploy the technology. They also suggested that the Trade and Technology Council, a U.S.-EU body that gathers again next month, hammer out an agenda for the summit at its next meeting.
The legislators also called on companies and AI laboratories to “significantly increase transparency towards and dialogue with regulators” and “ensure they maintain control over the evolution of the artificial intelligence they are building.”