Much scholarship considers how U.S. federal agencies govern artificial intelligence (AI) through rulemaking and their own internal use policies. But agencies have an overlooked AI governance role: setting discretionary grant policy when directing billions of dollars in federal financial assistance. These dollars enable state and local entities to study, create, and use AI. This funding not only goes to dedicated AI programs, but also to grantees using AI in the course of meeting their routine grant objectives. As discretionary grantmakers, agencies guide and restrict what grant winners do -- a hidden lever for AI governance. Agencies pull this lever by setting program objectives, judging criteria, and restrictions for AI use. Using a novel dataset of over 40,000 non-defense federal grant notices of funding opportunity (NOFOs) posted to the U.S. federal grants website between 2009 and 2024, we analyze how agencies regulate the use of AI by grantees. We select records mentioning AI and review their stated goals and requirements. We find agencies promoting AI in notice narratives, shaping adoption in ways other records of grant policy might fail to capture. Of the grant opportunities that mention AI, we find only a handful of AI-specific judging criteria or restrictions. This silence holds even when agencies fund AI uses in contexts affecting people's rights and which, under an analogous federal procurement regime, would result in extra oversight. These findings recast grant notices as a site of AI policymaking -- albeit one that is developing out of step with other regulatory efforts and incomplete in its consideration of transparency, accountability, and privacy protections. The paper concludes by drawing lessons from AI procurement scholarship, while identifying distinct challenges in grantmaking that invite further study.

It is notable, responsive, and appropriate that the majority of the Initiative projects had various interdisciplinary aspects to them--either in their teams or in the people they convened. Grantees felt that such interdisciplinarity was important to continue and in the long term the community will be healthier and more resilient for this. While the "diversity disaster" in the broader AI field is well known.7 Within the sub-field of AI ethics and governance, grantees noted that a field that relied on the same voices, geographies and, often, institutions would result in missing perspectives. With 15% of grantees being non-US based8 and 80% of funding support academic institutions, on this front, and in line with their international ambitions9 the Initiative could have done more. There remains an imperative to push for substantive diversity of thought and experience, both within geographies and across them.

This year, Kaggle started a new program called the BIPOC (Black, Indigenous, People of Color) Grant Program. It aims to empower underrepresented data scientists with support to advance their careers and aspirations. I am grateful that I was one of the few people who became a part of this wonderful program. All the students who became part of the program were assigned a mentor as well. I had done a few basic projects before I became a part of this program.

Since last January, when we launched our AI for Health program, we've been dedicated to using AI and data science to help improve the health of people and communities worldwide. As we reflect on how the world has changed this past year due to the pandemic, we want to take a moment to shed light on the great work our grantee partners are doing to tackle some of the most difficult health challenges. Our AI for Health program's commitment is to empower grantees. To date, we have awarded over 180 grants in four areas of focus, which include accelerating medical research, increasing global health insights, addressing health equity and building research capabilities. Access to Microsoft's technology such as Azure High Performance Computing, Azure Machine Learning, Power BI, Return to School Power Platform solution and the SmartNoise differential privacy platform have accelerated the progress made in grantee research.

Through our work in the Microsoft AI for Accessibility program, we have learned there are big gaps in mental health services around the globe. In some countries, there may only be one mental health professional per 100,000 people. When paired with the reality that 1 in 5 people have a mental health condition, we are asking how technology can and should be involved. In February, we shared our call for project proposals that aim to accelerate mental health research, data insights, and innovations using AI, and today we want to highlight the projects we're supporting. Of the 89% of people who screened positive for major depression through Mental Health America's online survey last year, 79% do not want to pursue psychotherapy or medications, yet 50% want access to digital tools.

What if we could treat the Earth like a computer, a system with an ever-flowing set of data that can be tracked, analyzed, and potentially even predicted. That's the gist of Microsoft's latest environmental initiative, which it's dubbed a "Planetary Computer." The company foresees a world where we can track just about anything happening in the world -- a forest fire in California, the river tides in Uganda -- and have all of that data readily accessible on a single AI-driven platform. If Microsoft succeeds it could reshape our relationship with the Earth entirely. Lucas Joppa, Microsoft's first chief environmental officer, boiled down the concept succinctly in an interview for the Engadget Podcast: "It's a platform that is intended to accelerate our ability to monitor, model and then ultimately manage Earth's natural systems to ask questions like, 'Where are the world's forests? Where are the world's wetlands? How fast are they changing?' And hopefully, what are the sorts of benefits that we are gaining from those ecosystems? What are the services that those ecosystems provision to people?"

Severe weather is impacting agriculture across the globe. The Midwestern United States has been continually flooding since March, inflicting $2.9 billion in property damage and threatening the livelihoods of farmers throughout the region. Internationally, food security is under threat from an onslaught of drought, while agriculture is already subject to the challenges of thin margins and complex global trade. Meanwhile, there is increasing pressure to do more with less to ensure food security for the growing global population. To meet these pressures, farmers are driven to get the most out of every harvest, even if that short-term focus may have long-term ill effects on the soil and ultimately their yield.

The rapid growth of text data has motivated the development of machine-learning based automatic text summarization strategies that concisely capture the essential ideas in a larger text. This study aimed to devise an extractive summarization method for A-133 Single Audits, which assess if recipients of federal grants are compliant with program requirements for use of federal funding. Currently, these voluminous audits must be manually analyzed by officials for oversight, risk management, and prioritization purposes. Automated summarization has the potential to streamline these processes. Analysis focused on the "Findings" section of ~20,000 Single Audits spanning 2016-2018. Following text preprocessing and GloVe embedding, sentence-level k-means clustering was performed to partition sentences by topic and to establish the importance of each sentence. For each audit, key summary sentences were extracted by proximity to cluster centroids. Summaries were judged by non-expert human evaluation and compared to human-generated summaries using the ROUGE metric. Though the goal was to fully automate summarization of A-133 audits, human input was required at various stages due to large variability in audit writing style, content, and context. Examples of human inputs include the number of clusters, the choice to keep or discard certain clusters based on their content relevance, and the definition of a top sentence. Overall, this approach made progress towards automated extractive summaries of A-133 audits, with future work to focus on full automation and improving summary consistency. This work highlights the inherent difficulty and subjective nature of automated summarization in a real-world application.

Last year, Microsoft took the wraps off its AI for Accessibility initiative, a $25 million, five-year program that focuses on leveraging the power of AI to help people with disabilities. Atlanta-based Zyrobotics became the first grantee for the program, gaining Microsoft's help in developing science, technology, engineering, and math (STEM) games and learning tools geared towards education of young children. And now, over a year later, Microsoft has announced 11 new AI for Accessibility grants, bringing the total number of grantees currently benefiting from AI for Accessibility up to 32. This news also aligns with the celebration of National Disability Employment Awareness Month, which carries on through the month of October. "The amount of potential that there is for software or hardware to better meet the needs of people with disabilities, and to raise the bar of what customers can come to expect of the role technology could play in their lives, is just an amazing opportunity."

Two Canadian startups have been named grantees of Microsoft's AI for Accessibility initiative, a $25 million, five year grant program launched last year to help NGOs, academics, researchers, and inventors accelerate their work for people with disabilities. "We have a huge opportunity and a responsibility to be making technology smarter and more useful for people with disabilities." AI for Accessibility wants to amplify human capability for people around the world with disabilities, by funding relevant projects that leverage and apply AI technology. The program is part of Microsoft's broader AI for Good initiative. This new round of 11 grantees brings the total projects included in the program to 32, which are spread across 13 countries.