The papers attempt to prevent potential harm from the new technologies and promote research into how AI deployment could benefit society, all while strengthening U.S. leadership in the field of artificial intelligence generally.
The primary policy document, "A Framework for U.S. AI Governance: Creating a Safe and Thriving AI Sector," makes the argument that current U.S. government organizations that currently have jurisdiction over the pertinent fields may frequently regulate AI tools. The suggestions also stress how crucial it is to determine the function of AI tools so that laws can be tailored to suit those uses.
The project was led by Dan Huttenlocher, dean of the MIT Schwarzman College of Computing, who noted that "as a nation we're already regulating a lot of relatively high-risk things and providing governance there." The idea for the project originated from the work of an ad hoc MIT group. "While we don't think that's enough, let's start with areas where human activity is already subject to regulations and that society has determined over time to be high risk. That is the practical way of looking about AI.
Asu Ozdaglar, the director of MIT's Department of Electrical Engineering and Computer Science (EECS) and deputy dean of academics in the MIT Schwarzman College of Computing, who also assisted in overseeing the framework, adds, "The framework we put together gives a concrete way of thinking about these things."
The initiative comes at a time when industrial investment in AI has increased significantly and interest in the topic has increased over the past year. It also includes several other policy papers. Currently, the European Union is working to complete AI legislation using a different methodology that imposes varying degrees of risk on different kinds of applications. General-purpose AI technologies, including language models, have emerged as a new source of contention in that process. The difficulties facing any governance endeavor include controlling both general and specialized AI technologies,
"We felt that MIT should be involved in this because we have the necessary expertise," said David Goldston, the MIT Washington Office's director. "MIT is a pioneer in AI research and one of the original hubs for the field. We feel obligated to assist in resolving these significant concerns because we are among those developing the technology that is bringing them to light.
Purpose, intent, and guardrails
The primary policy brief describes the potential expansion of current policy to include AI, making use of extant regulatory bodies and legal responsibility frameworks where applicable. For example, the medical industry is subject to stringent licensing laws in the United States. Since it is currently illegal to pose as a doctor, it should be obvious that using artificial intelligence (AI) to prescribe medication or diagnose patients would be prohibited, same like using strictly human wrongdoing would. This is not just a theoretical approach, as the policy brief points out; autonomous vehicles, which use AI systems, are governed by the same laws as regular vehicles.
The policy brief highlights that having AI providers identify the goal and intent of AI applications in advance is a crucial first step in creating these regulatory and liability frameworks. By evaluating emerging technologies in this way, it would become evident which current regulatory frameworks and regulators apply to any particular AI tool.
It is also possible for AI systems to exist at different levels, forming what techies refer to as a "stack" of systems that work together to provide a certain service. For instance, a particular new tool might be based on a general-purpose language model. As the brief points out, issues pertaining to a particular service may generally fall under the purview of the service provider. As stated in the first brief, "it may be reasonable for the provider of that component to share responsibility when a component system of a stack does not perform as promised." Therefore, if the creators of general-purpose tools are found to be responsible for particular issues, they should also be held liable.
That makes thinking about governance more difficult, but foundation models shouldn't be entirely disregarded, according to Ozdaglar. "You often build an application on top of models that come from providers; yet, the models are still a component of the stack. What duty is that in that case? Systems should still be taken into consideration even if they are not at the top of the stack.
That complicates thinking about governance, but Ozdaglar says foundation models shouldn't be completely discounted. "Models from providers are frequently the foundation upon which an application is built, but the models themselves remain a part of the stack. Then, what obligation is that? Even if a system is not at the top of the stack, it is still important to evaluate it.
Responsive and flexible
Although this makes thinking about governance more difficult, foundation models shouldn't be entirely disregarded, according to Ozdaglar. "Provider models are often the cornerstone upon which an application is constructed, although the models themselves stay within the stack. So what's the obligation there? It is crucial to assess a system even if it is not at the top of the stack.
Existing agencies are included in the policy framework, but it also adds some new oversight authority. The policy brief, for example, advocates for improvements in the auditing of emerging AI technologies. These changes might come from user-driven, government-initiated, or legal liability procedures. The report emphasizes that public standards for auditing would be required, whether they were set by a federal organization like the National Institute of Standards and Technology (NIST) or by a nonprofit organization like the Public Company Accounting Oversight Board (PCAOB).
Furthermore, the paper does advocate for the establishment of a new, government-approved "self-regulatory organization" (SRO) agency that would operate similarly to the Financial Industry Regulatory Authority (FINRA), which was established by the government. An AI-focused organization of this kind may amass domain-specific expertise that would enable it to be adaptable and quick to react when interacting with a quickly evolving AI market.
Huttenlocher, the Henry Ellis Warren Professor of Computer Science, Artificial Intelligence, and Decision-Making in EECS, argues that responsiveness is necessary since human-machine interactions are highly complicated. We believe that the government ought to take a close look at current SRO structure before considering creating new agencies. Since the store is still subject to government oversight and is still chartered by the government, the keys are not being turned over.
“Part of doing this properly”
Encouraging additional research on how to make AI helpful to society as a whole is another aspect of effective government participation on the matter, as the policy briefs make clear.
As an example, the policy paper "Is It Possible to Create a Pro-Worker AI? The article "Choosing a path of machines in service of minds," written by Simon Johnson, David Autor, and Daron Acemoglu, examines the idea that rather than being used to replace workers, AI might be used to support and enhance them. In this scenario, long-term economic growth would be more evenly distributed throughout society.
The goal of the ad hoc committee was to extend the scope of policymaking by considering a range of analyses from other disciplinary viewpoints, rather of focusing only on a few technical issues, when it came to the topic of AI legislation.
According to Huttenlocher, "We do believe academic institutions have an important role to play in terms of expertise about technology as well as the interplay of technology and society." It illustrates the key to effectively controlling this: legislators who consider social processes and technology in tandem. The country is going to require that.
In fact, as Goldston points out, the committee is making an effort to mediate the divide between those who are enthusiastic about AI and those who are apprehensive about it by pushing for appropriate regulation to follow technological advancements.
The committee that released these materials, in Goldston's words, "is not a group that is antitechnology or trying to stifle AI." Nevertheless, this group is arguing that supervision and governance are necessary for AI. It's necessary to accomplish this correctly. These individuals, who are knowledgeable with the technology, argue that supervision of AI is necessary.
You must be logged in to post a comment.