Artificial intelligence is reshaping how governments operate, workers do their jobs, and how citizens interact with public institutions. But while technology is advancing rapidly, the framework for using it effectively and responsibly is still taking shape. In this collection of perspectives, policy, economics, education, and technology experts from the Harvard Kennedy School outline how governments can think about AI. As a tool to improve services, a source of new risks, and a force that requires careful monitoring, new skills, and continuous adaptation.
• Mark Fagan: On how governments should think about AI.
• Jason Furman: On the principles of AI regulation.
• Video by Teddy Svoronos, Mark Fagan, Bruce Schneier
• Workforce projects: Tracking AI adoption
• Bruce Schneier: On AI and elections
• Teddy Svoronos: On teaching generative AI.
Mark Fagan: How governments should think about AI
People need to think of AI as another tool in their toolkit to deliver configuration services with quality, efficiency, and fairness. AI is a very powerful tool. It’s an evolving tool. But it’s just a tool.
All government organizations, and by extension non-profit organizations, face the current dilemma of accomplishing more with less. Despite budget cuts, constituent and stakeholder demands are increasing, and government officials are under tremendous pressure to meet those expectations. AI provides a more efficient way to provide a significant number of configuration services.
Currently, if you have any questions about your child’s enrollment in a school, you should contact your school system representative by phone. Typically, the window for doing this is a business day. There may also be language restrictions. AI chatbots may be available 24/7. Being able to operate in multiple languages may provide more reliable and efficient information. That’s the benefit we get from AI.
That’s the good side.
I have some concerns. The first is data privacy. Governments have a lot of information, and one of their duties is to protect that information. When incorporated into AI algorithms, they must be non-identifiable at an individual level. Protecting information through anonymization, being careful about its use, and keeping it discreet are ways to address data privacy. The second real risk is hallucinations. The way large language models work is that they simply try to predict the next word based on a wealth of information. You can get stupid information. The solution is to verify, verify, verify.
There are two other issues of concern. One is prejudice. All models are trained on data. If that data is not representative of the underlying population, you may get biased information. This is an issue that has come to the fore when it comes to hiring talent. The final big risk is disinformation. Today, it is very easy to use AI to change information, images, videos, or share inaccurate information.
The government’s role here is to monitor and build trust with the community. We need people with AI expertise: technical expertise, machine learning, data science, and cybersecurity. Expertise in AI ethics and regulation is also required. What is the law in a regulated environment? What is ethical decision-making? Another important area of expertise is operating under uncertainty.
AI development is progressing in real time. It is very important to be agile, learn and adapt.
People within your organization need to know what AI is. You can also leverage the AI community. Many people, communities, government agencies, and nonprofit organizations are working in this area. Take advantage of their expertise.
Let’s talk about concrete guardrails. These consist of policy and ethical considerations. The European Union has passed the EU AI Law. It’s a risk-based pyramid. It starts with things that AI cannot do, such as human cloning, social tracking, and social scoring. Next, there are several things you can do, but you need to be transparent. And you’re basically free to do anything that doesn’t create risk.
Ethical considerations relate to what data is used. That chatbot I mentioned when we talked about admissions? Well, if someone has access to technology that allows them to access chatbots, that would be great. Many people don’t have that. Moreover, being accessible is not enough. It takes literacy to know how to interact with it. Guardrails and ethical considerations ensure quality, efficiency, and fairness.
mark fagan I am a lecturer on public policy. His research focuses on regulation and he is co-author of the new book Governing With AI: How the Public Sector Can Use Artificial Intelligence to Recommend Performance. He recently co-authored the AI Risk Assessment Framework.
—
Jason Furman: On the principles of AI regulation
As policymakers consider regulating AI, they should keep these five principles in mind:
- Consider both the benefits and risks. It is tempting to apply the precautionary principle and demand that AI prove safe and eliminate all risks before moving forward. But doing so risks missing out on the huge benefits that AI offers in areas such as scientific research, education and the broader labor market. Delaying any of these benefits can also be costly. The right approach, then, is the old-fashioned one of balancing the potential costs and risks of AI with the costs and risks of delaying AI adoption.
- Compare AI to a non-omnipotent human being. AI is biased. AI gets into a traffic accident. AI can get overconfident and make things up. But guess what? Humans do all of those things too, but often make them all worse. Additionally, AI can better learn from its mistakes. So instead of looking for perfection, look for improvement.
- Domain-specific regulation rather than AI superregulators. AI will be used in everything from cars to medical equipment to stock trading. Regulations are already in place for automobiles, medical devices, and stock trading. Rather than having a super-regulator for every use of AI, we should enhance the AI capabilities of existing regulators so that they can determine whether a car is safe or not, based on the output of the technology rather than its input. Is the medical device effective? And ask the questions already being asked today.
- Regulation should not be used as a means to protect incumbent companies. Big AI companies may also welcome regulation for well-intentioned reasons. They may also worry that they think they can abide by the rules but their smaller competitors cannot. We have benefited from strong competition in the AI space. Unnecessary regulations should not hinder competition and encourage the entrenchment of monopoly power.
- Not every problem has a solution. We need to work hard to ensure that AI does not help people create sexually abusive images or biological weapons. These issues need to be resolved immediately. But I don’t see how AI regulation can ensure that technology doesn’t increase inequality or make work more meaningful. Instead, we will need to find solutions to such problems outside of technology, for example through more progressive taxes and transfers.
