How Is AI Being Used and Governed? Stanford HAI 2019 Fall Conference, AI in Government

TL;DR

AI is being adopted across government, but public support depends on how its use is justified and governed. A Stanford-led project involving 15 law students and 10 computer science PhDs examined roughly 150 federal departments, agencies, and sub-agencies and identified about 170 AI use cases. The panel also addresses skepticism, status quo preferences, equity, fairness, accountability, privacy, and employment concerns. Read on for the key findings and governance questions.

Transcript

• All right, welcome. I'm Dan Ho. I'm a professor at the law school here also with affiliations with the political science department and CPR and with HAI. And we heard this morning from a really interesting discussion between Eric Schmidt and Mori Chisaki about regulating big tech. And I think this panel builds on that discussion really nicely. Th... Read More

Key Insights

• 😒 Citizen support for algorithmic government is not universal and varies depending on the justifications provided for their use.
• ❓ Citizens tend to prefer the status quo over algorithmic innovations, even when the innovations offer potential improvements.
• ⚾ Trust in algorithms is not solely based on their performance, and citizens evaluate algorithms differently from humans.

Questions & Answers

Q: How is AI being used in government?

The panel describes AI adoption across a broad range of government activities. A Stanford-led project examining federal agencies found roughly 170 use cases across about 150 departments, agencies, and sub-agencies, with a focus on civil regulatory purposes.

Q: What are the main obstacles to citizen support for algorithmic government?

Support varies according to the justification offered for using algorithms, and no single justification receives universal support. Other obstacles include a preference for the status quo, skepticism about algorithmic decisions, and concerns about job loss, inequality, and reduced social mobility.

Q: Do citizens prefer algorithmic innovations over the status quo?

Citizens tend to prefer the status quo, even when an algorithmic innovation could offer improvements. This preference is one reason support for algorithmic government remains limited.

Q: How do citizens evaluate algorithms compared with humans?

Citizens may judge algorithms negatively even when they agree with the decisions those algorithms make. Their trust is not solely dependent on performance, and they apply different standards when assessing algorithms and humans.

Q: What did the Stanford federal-agency AI project examine?

The project examined how federal agencies use AI for civil regulatory purposes, a frontier distinct from state and local uses in criminal justice. Its team surveyed roughly 150 federal departments, agencies, and sub-agencies and surfaced about 170 AI use cases.

Q: Who conducted the review of AI use across federal agencies?

David Engstrom reported preliminary results from a project co-led with Dan Ho, Tino Cuellar, and NYU’s Cathy Sharkey. They assembled 25 students: 15 law students and 10 computer science PhD students.

Q: What government AI issues does the Stanford HAI panel address?

The session has two main components: describing how AI is being adopted in the public sector and exploring possible governance pathways. Its subjects include civil regulation, privacy, transparency, surveillance, equity, fairness, accountability, and public opinion about AI tools.

Q: What role does New York City’s Automated Decision Systems Task Force have?

The task force is charged with developing a process for understanding algorithms used by New York City. Its work particularly emphasizes equity, fairness, and accountability, and panelist Brittny J. Saunders serves as its co-chair.

Summary & Key Takeaways

• Citizen support for algorithmic government varies depending on the justifications given for their use, with no single set of justifications receiving universal support.

• Citizens tend to prefer the status quo over algorithmic innovations, even when the innovations offer potential improvements.

• Trust in algorithms is not dependent on their performance, and citizens evaluate algorithms differently from humans.

• Opposition to algorithmic government is correlated with concerns about job loss, increased inequality, and hampered social mobility due to automation and AI.