Who, where and how will AI be regulated?

Concerns about artificial intelligence are probably the only thing that has progressed as rapidly as it has in the past year.

In the short term, chatbots such as OpenAI’s ChatGPT flood the world with harmful language and disinformation, automated decision-making systems that discriminate against certain groups, and a lack of transparency in many AI systems. Many are concerned that the problem will remain hidden. There is also the pressing concern of job losses as AI systems prove capable of matching or exceeding human performance. And in the long term, some prominent AI researchers worry that developing AI systems that are more intelligent than humans could pose an existential risk to humanity.

The rapid advancement of this technology has created a new urgency for efforts around the world to regulate AI systems. The European Union started first, and this week on June 14, one of its bodies, the European Parliament, took a step forward by passing a bill known as the AI ​​Law. However, China’s rule-makers are among the first to move to turn their proposals into actual rules, with countries such as Brazil, Canada and the United States following suit.

The EU and the UK have conducted contrasting studies. The former is a regulatory forward. The latter is laissez-faire.

Notably, some of the calls for regulation have come from the very companies that have developed the technology and benefited most from unrestricted commercial deployment. “OpenAI believes AI regulation is essential,” OpenAI CEO Sam Altman recently said in written testimony before the U.S. Congress. He also urged lawmakers to consider licensing requirements and safety tests for large AI models. Meanwhile, Sundar Pichai, CEO of Google and its parent company Alphabet, recently said there would be a need for a “global framework” to govern the use of AI.

But not everyone thinks new rules are needed. The non-profit Center for Data Innovation supports the no-intervention approach taken by the UK and India. These countries intend to use existing regulations to address potential problems with AI. Hodan Omar, a senior policy analyst at the nonprofit, said: IEEE spectrum The European Union will soon feel the chilling effect of the new regulations. “By making it harder for European digital entrepreneurs to launch and grow new AI businesses, the EU is also making it harder for jobs, technological progress and wealth creation,” she said.

What does the EU AI law do?

The course of events in Europe will certainly serve as an example for governments around the world to learn from. In April 2021, the EU’s European Commission proposed an AI law that would adopt a risk-based hierarchy. AI applications that pose “unacceptable risks” will be banned. High-risk applications in areas such as finance, justice, and healthcare will be subject to increased scrutiny. Disclosure is required for limited-risk applications, such as the use of chatbots.

As mentioned above, on Wednesday 14th June, the European Parliament passed a draft of this law. While this is an important step, it’s only one step in the process. Parliament and another EU body, the European Union Council, have also proposed amendments to the law since its entry into force in 2021. Negotiations on the proposed amendments began in July and are expected to reach agreement on the final terms by the end of 2023. If the bill follows its normal schedule, the law will come into force in two years.

Connor Dunlop, European Public Policy Director at the nonprofit Ada Lovelace Institute, said one of the most controversial proposed amendments would be biometrics, including facial recognition systems currently used by law enforcement. It said the European Parliament is proposing to ban certification surveillance.

In the United States, a national law was proposed in 2022, but did not yield any results. The White House then issued a blueprint for his AI Bill of Rights. it is non-binding.

Another hot topic is the Congressional Amendment that seeks to cover recent advances in the “Foundation Model,” large-scale, flexible AI systems that can be adapted to a wide range of applications. “AI law is designed as a product law,” explains Dunlop. “Risk is defined by its intended purpose.” But that framework focuses on the companies and organizations that implement the technology, leaving the developer of the underlying model untouched. “What the European Parliament is trying to do is add another layer to the regulation,” Dunlop said. “For things like data transparency, only platform developers can make the system compliant, and it is very difficult for downstream implementers to do this.”

How do other countries regulate AI?

The active policy arena is not limited to Europe. Jeffrey Ding, assistant professor of political science at George Washington University and author of the ChinAI newsletter, says there is a false belief in the West that China doesn’t care about AI governance. In fact, China’s regulations are already in force, starting with the Recommended Algorithms rule, which went into effect in March 2022, calling for transparency for service providers and a way for citizens to opt out, he said.

Then, in January 2023, the Chinese government issued initial regulations governing generative AI, with further draft regulations proposed in April 2023. Ding says his rules about recommender algorithms and generative AI stem from common concerns such as: “The Chinese government is very concerned about algorithms for the general public that have the potential to shape societal perceptions,” he says.

China’s early rules on generative AI required websites to label AI-generated content, banned the production of fake news, and required companies to register their algorithms and disclose information about their training data and performance. Some of the language is broad enough to give governments considerable leeway in enforcement, such as the requirement that AI providers “dispel rumors” generated by AI-generated content.

“By focusing on civil rights, we can demonstrate protections that are independent of the technology being used, whether it is an Excel spreadsheet or a neural network.”
—Suresh Venkatasubramanian, Brown University

The draft rules go even further, requiring AI companies to verify the veracity of all data used to train models. “It’s an impossible attempt,” Ding said. “If that rule were actually introduced and enforced, it would impose very large costs on these AI providers.” It notes that the scope of the regulation will be narrowed to general public products, leaving only B2B applications. “Companies may respond by paywalling their models, so that only companies can access them and are not subject to government regulator scrutiny,” he says.

Lawmakers are off to a slow start in the United States, where many companies and research institutes are developing cutting-edge AI models. A national law was proposed last year, but did not yield any results. And in October 2022, the White House will issue a blueprint for a non-binding AI Bill of Rights, positioning AI governance as a civil rights issue and ensuring citizens are protected from algorithmic discrimination, invasion of privacy, and other harm. said there is.

Suresh Venkatasubramanian, a computer science professor at Brown University, co-authored the blueprint while working for the White House Office of Science and Technology Policy. He said the blueprint hints at a civil rights approach in hopes of creating flexible rules that can accommodate rapidly changing technology. “By focusing on civil rights, we can demonstrate protections that are independent of the technology being used, whether it is an Excel spreadsheet or a neural network,” he says.

Venkatasubramanian said that while individual federal agencies are enacting regulations on AI in their respective fields, there is “broad consensus that something should be done” at the legislative level. And indeed, U.S. Senate Majority Leader Chuck Schumer announced in April that he was distributing a draft “high-level framework” for AI regulation.

AI companies doing business around the world will find it difficult to comply with all local rules unless countries reach a global agreement. An intergovernmental political forum known as the Group of Seven (G7) has already started discussing AI governance, but no one expects the organization to move quickly. Meanwhile, European officials have suggested companies around the world could sign a voluntary AI code of conduct, saying they will submit a draft in the coming weeks. There is no time to waste, said European Commission Vice-President Margrethe Vestager at her recent meeting.