In 2022, the Algorithmic Accountability Act was introduced in the U.S. The AAA addresses growing public concerns about the widespread use of automated decision-making systems (ADS). It proposes that organizations implementing such systems take several concrete measures to identify and mitigate social, ethical, and legal risks. As a legislative effort to regulate ADS across all sectors, the U.S. AAA marks a milestone in a global trend toward supplementing or replacing self-regulation in this area with legislation. The most influential example of this trend is the Artificial Intelligence Act proposed by the European Commission (2021).
However, the story does not end there. The U.S. now has its most comprehensive official artificial intelligence policy to date. Last week, President Biden signed a historic executive order requiring new federal standards for AI safety, security, and reliability. The order addresses various aspects of AI risk management and development, establishing new standards to safeguard Americans’ privacy, promote civil rights, protect consumers and workers, and encourage innovation and competition.
The sweeping order, nearly 20,000 words long, uses the term “artificial intelligence” to refer to predictive, perceptual, or generative automated software that can mimic certain human abilities.
The rapid rise of artificial intelligence—specifically, generative AI systems—has sparked intense concern and existential fears regarding concrete and demonstrable risks in the present. For example, AI models clearly exacerbate the problem of misinformation through visual deepfakes and the instantaneous generation of text. Machine learning algorithms encode biases that amplify and automate existing patterns of discrimination, potentially influencing human behavior in the long term. There are threats to privacy posed by the vast volumes of data collected through AI systems—including facial recognition software—and used to train new generative AI models. Artificial intelligence could also become a major threat to national security; for example, AI models could be used to accelerate the development of new chemical weapons.
The new executive order moves the U.S. toward more comprehensive AI governance, as—unlike previous actions—it goes beyond principles and guidelines, with certain sections requiring specific actions by technology companies and federal agencies.
The Executive Order calls for the following actions:
a) New standards for AI safety and security
As AI capabilities intensify, so do their potential risks and implications for the safety and security of individuals. Hence the requirement that developers of AI systems share the results of safety tests and other critical information with the U.S. government. Specifically, the Executive Order requires companies developing any foundational model that poses a serious risk to national security to notify the federal government when training the model and to share the results of all safety tests. These measures ensure that AI systems are safe, secure, and reliable before companies bring them to market.
Specifically, the requirement applies to models containing “tens of billions of parameters” that have been trained on extensive datasets. This transparency rule will likely apply to the next version of OpenAI’s GPT, the large language model that powers its ChatGPT chatbot.
b) Development of standards, tools, and tests to ensure that AI systems are safe, secure, and reliable.
The National Institute of Standards and Technology will establish rigorous standards to ensure safety prior to public release. The Department of Homeland Security will apply these standards to critical infrastructure sectors and establish the AI Safety and Security Council. In addition to critical infrastructure, chemical, biological, radiological, nuclear, and cybersecurity risks will be addressed.
c) Protection against AI-enabled fraud, by establishing standards and best practices to detect AI-generated content and authenticate official content. The Department of Commerce will develop guidelines for content authentication and watermarks to clearly label AI-generated content. Federal agencies will use these tools so that Americans can easily know that the communications they receive from their government are authentic.
d) Establishment of an advanced cybersecurity program to develop AI tools capable of detecting and correcting vulnerabilities in critical software, including ensuring that the U.S. military and intelligence community uses AI safely, ethically, and effectively in their missions, while taking steps to counter adversaries’ military use of AI.
The fact is that, without safeguards, AI can put Americans’ privacy at even greater risk. After all, AI not only facilitates the extraction, identification, and exploitation of personal data, but also increases the incentives to do so, because companies use data to train AI systems.
Thus, the Executive Order calls on Congress to pass data privacy legislation to protect all Americans, especially children, and outlines the following actions:
a) accelerate the development and use of privacy-preserving techniques—
including those that utilize cutting-edge AI and enable the training of AI systems while preserving the privacy of training data.
b) strengthen privacy-preserving research and technologies, such as cryptographic tools that preserve individuals’ privacy. The National Science Foundation will work with this network to promote the adoption of cutting-edge privacy-preserving technologies by federal agencies. The development process will involve “red-teaming,” in which benevolent hackers will work with model developers to proactively analyze vulnerabilities.
c) assess how federal agencies collect and use commercially available information containing personally identifiable information, strengthening privacy guidelines to ensure they account for AI risks.
With regard to the promotion of civil rights, the irresponsible use of AI can exacerbate bias and other abuses in the justice system, healthcare, and housing. The Executive Order directs agencies to combat algorithmic discrimination, while also instructing authorities to provide clear guidance to the market to prevent AI algorithms from being used to exacerbate discrimination
To ensure fairness throughout the criminal justice system, efforts are being made to develop best practices regarding the use of AI in sentencing, probation and parole, pretrial release and detention, risk assessments, surveillance, crime prediction and predictive policing, and forensic analysis.
On the other hand, AI can bring real benefits to consumers—making products better, cheaper, and more widely available. However, AI also increases the risk of injuring, misleading, or otherwise harming consumers.
To protect consumers while ensuring that AI can improve the lives of Americans, the Executive Order promotes the responsible use of AI in healthcare and in the development of affordable medications.
Furthermore, the Executive Order shapes AI’s potential to transform education by creating resources to support educators in implementing AI-enabled educational tools.
On the other hand, AI is radically changing America’s job market and workplaces, offering the promise of improved productivity as well as the dangers of increased workplace surveillance, bias, and job displacement.
To mitigate these risks, support workers’ ability to bargain collectively, and invest in workforce training and development that is accessible to all, the Executive Order directs the development of principles and best practices to mitigate the harms and maximize the benefits of AI for workers, preventing employers from undercompensating the workforce, unfairly evaluating job applications, or interfering with workers’ ability to organize.
With regard to promoting innovation and competition, the Executive Order ensures that the U.S. leads the way in innovation and competition by catalyzing AI research in the United States through a tool that will provide researchers and students with access to key AI resources and data in vital areas such as health and climate change. The goal is to foster a fair, open, and competitive AI ecosystem, providing small developers and entrepreneurs with access to technical assistance and resources, helping small businesses commercialize AI advancements, and expanding opportunities for qualified immigrants to study and work in the United States by modernizing and streamlining visa criteria, interviews, and evaluations.
Recognizing that the challenges and opportunities of AI are global, the Executive Order seeks to expand bilateral, multilateral, and multisectoral commitments to collaborate on AI, manage its risks, and ensure that the technology is safe, reliable, and interoperable.
The U.S. Administration is working with allies and partners to establish a strong international governance framework for the development and use of AI, collaborating with Australia, Brazil, Canada, Chile, the European Union, France, Germany, India, Israel, Italy, Japan, Kenya, Mexico, the Netherlands, New Zealand, Nigeria, the Philippines, Singapore, South Korea, the United Arab Emirates, and the United Kingdom.
In short, there are many first steps being taken in many directions. Although the policy itself is not strictly speaking a regulation, it represents a major step toward imminent regulation because it gathers a vast amount of data through all the working groups dedicated to AI and the agencies’ research and development efforts. Collecting this information is essential for the next steps; after all, to regulate, one must first understand what is happening.
However, as broad as the new policy may be, there are still notable gaps. The executive order says nothing about the specific protection of biometric data privacy, including facial scans and voice clones. There are also gaps when it comes to addressing the government’s use of AI in defense and intelligence applications. Even where the order appears to be effective, there may be a considerable gap between what policymakers expect and what is technically feasible. “Watermarking” is an example of this. The new policy directs the Department of Commerce to identify best practices for labeling AI-generated content within the next eight months—but there is no robust, established technical method for doing so.
Therefore, the executive order alone is insufficient to resolve all the problems posed by the advancement of AI, as it is inherently limited in its authority and can easily be reversed. For this reason, the order itself calls on Congress to pass legislation on data privacy. However, there is no denying that this is a significant step forward that fills a policy gap. The European Union has been working publicly to develop the EU AI Act—which is close to becoming law—while the U.S. has been unable to make similar progress. With the executive order, there are steps to follow and changes on the horizon.
The difficult questions that need to be addressed concern which decision-making criteria and evidence should be considered legitimate—or at least socially acceptable—for different public and private decision-making processes. Answering these questions requires a positive vision of what future societies should look like. Policy makers must go beyond attempts to ensure “minimal” algorithmic accountability, focusing instead on designing public governance mechanisms that enable organizations to reach justifiable compromises within the bounds of legal permissibility.
__________