AI governance project ideas

As you come up with project ideas, we thought it would be helpful to provide examples of five projects that we’d be excited to see participants complete. Please note that these are example ideas. While we expect a few people might use these as they are, we hope that most use them as a template for their own ideas.

Ideas

Imagining positive AI futures in a specific sector, and anticipating governance needed to get there

As we highlight in the resource guide for session two, existing literature on positive AI futures is often of poor quality, quasi-utopian, and not grounded in evidence. As a result, policymakers find themselves unable to imagine the world they’re building towards. Without a sense of direction, it’s hard to set coherent goals. This problem is magnified by the nature of dual-use technology, which will have positive and negative effects across society.

To solve these problems, your project could anticipate AI progress over a set period, like over the next ten years. You could rely on existing resources predicting capabilities increases over set domains, or you could rely on expert forecasting on more generalized capabilities increases. Using this as a baseline, you could imagine the benefits these capabilities would bring to a sector. 

With an evidence-based positive future in mind, you could design a narrowly tailored policy intervention designed to steer the technology toward that direction. Alternatively, you could analyze an existing policy proposal for the same attributes. From there, you could present your vision and policy interventions in a paper or a blog post.

Analyze AI governance proposals at the regional level

Policy professionals have paid close attention to national-level AI proposals in the US, UK, EU, and China. These proposals have been thoroughly analyzed. Less attention has been paid to AI policy movement at the regional level within nations. For example, Colorado recently became the first US state to pass a comprehensive AI regulation. Many experts expect other states to follow suit. We think these policies have been under-analyzed, resulting in an information deficit for policy professionals.

Your project could analyze a state or regional-level AI policy proposal. It could analyze the potential impact of one proposal across multiple domains, or it could examine a variety of proposals to see how each addresses a specific risk scenario. It could contrast a proposal with a national proposal to understand what gap it aimed to address. Finally, it could evaluate what lessons policymakers should learn from this example. Your findings could be written up in a paper or a blog post.

Designing novel policies for risk mitigation that scale with capabilities

As we discussed in session 7, adaptation is an under-explored concept for mitigating the risks of advanced AI. Your project could aim to apply this concept to a risk by developing a policy that begins the process for adaptation and scales as capabilities increase. For example, there’s considerable debate over whether existing LLMs could help actors create bioweapons, and many scientists think this risk might increase with capabilities.

You could design a novel risk mitigation framework or policy intervention that seeks to foster adaptation to these risks. You could examine proposals to limit the availability of DNA manufacturing equipment or that would require labs to know their customers. You could identify policies that might become more relevant as capabilities increase, evaluating their current and future necessity based on defined capabilities thresholds. Your findings could be written up in a paper or a blog post.

Build a policy tracker

Online policy trackers (click here for an example) give researchers, policymakers, and the general public an overview of relevant actions on an issue by governments. Existing AI trackers tend to cover all global AI policies or all actions taken by a national government. In my opinion, most existing AI policy trackers are difficult to navigate. They are also often too information-dense or lack information on the content of each policy action.

Your project could be to create a policy tracker, either tracking all actions of a relevant country or actions on a subset of AI policy across multiple countries. You could Ingest governmental action from a reputable source that updates regularly to populate the site. You could use an LLM to automatically summarize the policy based on the criteria you define. Your deliverable would be a website that others could use to gain insight into the policy areas you’ve chosen to track.

Examine a historical case study for relevance to international AI governance

Previous international agreements have sought to limit the proliferation of dangerous capabilities, foster cooperation, and set up new institutions. Now, with many actors calling for international governance for advanced AI, it’s worth examining past proposals to find relevant lessons.

Your project could analyze an existing international governance regime – like the International Atomic Energy Agency, the World Trade Organization, or the UN Conventions on the Law of the Sea – to identify whether elements of their structure, authority, or composition could be effective models for AI. Alternatively, you could examine historical case studies like the Montreal Protocol or the Strategic Arms Limitations Talks. Your findings could be written up in a paper or a blog post.

Other lists of project ideas

High-quality lists of open questions in AI governance can be found here and here.