Launched in January 2020, Microsoft’s AI for Health program is committed to improving the health of the world’s population. Since then, the AI for Health program has partnered with over 200 grantees on projects designed to accelerate medical research, build research capabilities, increase global health insights, and address health inequities.

Given that the COVID-19 pandemic surprised the world just months after program launch, the AI for Health program rapidly focused efforts on understanding, modeling, and visualizing COVID-19 infection, vaccination, and outcomes. As the pandemic has transformed to endemicity, the program has focused its efforts on three broad areas:

Current research

Public health. Applying visualization, data analytics, machine learning, and modeling to:

• Understand the relationships between social determinants of health and health outcomes, clinical care, health behaviors, and health status.
• Identify the social determinants of health that—if changed—would have the greatest return to the health of the population.
• Allow researchers and policymakers to develop and define indices of health risks to rapidly identify areas for intervention.
• Focus on relationships between local economic distress and social determinants of health and cardiovascular disease in data-rich cities (opens in new tab) (including New York City, Lisbon, Lausanne, Rio de Janeiro, and Singapore).

Imaging analytics. Efforts here have ranged from applying artificial intelligence and machine learning to:

• Early identification of leprosy from skin photographs in the Brazilian population, thereby preserving people’s fingers, toes, and limbs and, hopefully, accelerating the elimination of this ancient disease (with the Novartis Foundation). Read the paper: Reimagining Leprosy Elimination with AI Analysis of a Combination of Skin Lesion Images with Demographic and Clinical Data.
• Using cell-phone videos to identify children in Mexico that are at risk of retinopathy of prematurity the leading cause of preventable childhood blindness (with Clínica Oftalmológica Peñaranda, Red RoP de la Provincia de Buenos Aires, Centro integral de salud visual Daponte).
• Using captured images of tympanic membranes to identify Australian Aboriginal and Torres Strait children who have chronic otitis media, helping to prevent childhood deafness (with the University of Sydney). Read the paper: Evaluating the Generalizability of Deep Learning Image Classification Algorithms to Detect Middle Ear Disease Using Otoscopy 
• Radiological images, to identify, segment, and indicate abnormalities in the pancreas, the female breast, the liver, the lungs (with a multiplicity of partners).

Genomics and proteomics. Applying artificial intelligence, machine learning, and modeling to:

• Support the development of a biobank repository (opens in new tab) for the Mexican population. 
• Integrate genomics data to analyses of health outcomes from imaging.
• Identify potential drug binding sites from protein simulations with Folding@Home. Read the paper: Predicting the Locations of Cryptic Pockets from Single Protein Structures Using the PocketMiner Graph Neural Network

Looking forward, we anticipate continuing the above work and expanding efforts to include the application of large language models in our analytic repertoire. Further, we will continue to form deep, collaborative, global relationships with renown not-for-profit organizations (like Novartis Foundation) and academic institutions (like Tec Monterrey, Johns Hopkins University, New York University, and the Institute for Health Metrics and Evaluation at the University of Washington).

“The fact that a health problem can be predicted in advance will reshape the cost curve of healthcare.”

— Satya Nadella

It will also dramatically change the health and wellbeing of the world’s population. Artificial intelligence is the tool that allows for such advanced predictions; its application in healthcare will radically transform how healthcare is practiced and lead to a healthier, more productive, and more equitable world.

The post The Microsoft AI for Health program: Solving the world’s biggest health issues, one life at a time appeared first on Microsoft Research.

Every year natural disasters like flooding, hurricanes, tornadoes, and fire put millions across the globe at risk and cause losses worth trillions.¹ Climate change is expected to drive up the pace and severity of these events. The AI for Good Lab is currently doing a lot of work to combine geospatial data, like high-resolution satellite imagery, with innovative machine learning techniques to analyze the impacts of disasters, which can help governments respond once an event has happened. But we are going beyond that, developing tools to help predict natural disaster impacts ahead of time, making it possible to better target efforts to prevent loss of life and mitigate damage.

Predicting cyclone damage and heat wave impacts in India

One example is our work with SEEDS (opens in new tab), a nonprofit disaster response and preparedness in India, which received a grant in 2019 from Microsoft for financial and technical support to mitigate damage from cyclones. The path of a devastating cyclone can be predicted 48 hours or so of its arrival, but manually collecting information about houses in its path is very time intensive. The Lab helped SEEDS develop an AI model, called “Sunny Lives”, using high-resolution satellite imagery of the areas likely to fall in the cyclone’s path and applied advanced data analytics and machine learning to identify the houses at highest risk. The model was piloted for risk assessment of cyclone induced flooding during cyclone Nivar and cyclone Burevi in 2020 and showcased promising results. The information provided by the model and risk scoring pipeline enabled SEEDS and its on-ground partners to target outreach to those communities with measures such as distributing an advisory in multiple languages to thousands the day before a cyclone arrived.  

In another project in 2022, the Lab worked with SEEDS to refine the AI model to generate risk information for heat waves in India. India has seen twice as many heat waves between 2000-2019 as between 1980-1999.² The AI model focused on heat waves affecting around 125,000 people living in slums in New Delhi and Nagpur. The Lab leveraged data from local risk maps developed by a partner that scored buildings with such parameters as built-up density, vegetation, proximity to a water body and rooftop material. The risk map is overlaid onto a regular map which can then be displayed on a smartphone, helping teams in the field figure out where to issue warnings, or where the local authorities need to direct resources. Microsoft President Brad Smith noted at the Web Summit (opens in new tab) that the work done in collaboration with SEEDS is “an extraordinary combination of high-tech and low-tech coming together… It’s part of what the world needs to do to adapt to climate change.”

Expediting damage assessments for the Red Cross 

The Lab is currently working together with the American Red Cross on disaster response by applying our current Microsoft damage assessment tool to the Red Cross’s data on hurricanes and tornadoes in the U.S. from 2022. The research aims to investigate preliminary impact and damage assumptions to determine ways to improve and expedite insights that can help inform ground operations.   

Addressing food insecurity with machine learning

Millions of people worldwide face chronic food insecurity. Nearly 828 million people were hungry in 2021, or 9.8% of the world population, according to the Food and Agriculture Organization of the United Nations¹. In order for aid workers and governments to prepare for and deliver timely assistance, early warning systems are needed to help them quickly forecast the incidence of hunger as conditions change due to natural disasters, war and other factors.   

In 2021, the Lab conducted a study jointly with the Catholic Relief Services (CRS) to apply machine learning techniques to this problem. Data from household surveys performed on the ground in southern Malawi by CRS teams was combined with machine learning algorithms, creating a model to predict future food insecurity at the household level in near real-time. The model proved to have an 83% accuracy rate in predicting food security outcomes and could generate accurate forecasts up to 4 months in the future.   

Using AI to address inequities in public health 

A huge need exists to address inequities in health and healthcare for underserved populations, and to reduce disparities between populations across the globe. Microsoft is committed to combatting health disparities based on gender, race, ethnicity, or income level. In 2020, Microsoft invested $60 million in AI for Health, a five-year project to apply AI to improving global health. To read about the Lab’s role in this work, visit AI for Health | Microsoft AI.  

The Lab’s current work looks deeply at inequities in public health risks and disease prevalence in rural areas. Disparities in healthcare access, health status, screening, and broadband access lead to less healthy outcomes for rural dwellers compared to urban. A great deal of data is available about health risk behaviors, disease prevalence, health status, health services utilization, healthcare quality, local economic conditions and health outcomes, but no easy way to compare local data between non-metro and metro areas.

The Lab is developing a visualization tool, known as the Rural Health Index Map, to assist policymakers in understanding health disparities at the county and zip code levels. Embedded analytics will allow for simulations, showing how changing a parameter, such as a risk factor or economic distress, might impact health outcomes. This tool might help policymakers in a number of ways, by providing insights that help them decide on investments to improve local health status, for example, or to see how changes to rural broadband access can potentially impact public health.

Footnotes

¹ Disaster Relief and Rehabilitation (SEEDS) (opens in new tab)

² AI is helping vulnerable communities in India better understand heat wave dangers (Microsoft) (opens in new tab)

³ UN Report: Global hunger (FAO.org) (opens in new tab)

The post Fundamental rights and building resiliency appeared first on Microsoft Research.

During the COVID pandemic, as workplaces were closing and a global economic crisis ensued, global unemployment was heading towards a quarter of a billion people. It became apparent that expanded access to computer science education and digital skills would be critical for recovery, especially for people hardest hit by job losses: those with lower incomes, women, and underrepresented minorities.

To help address these needs, Microsoft launched a worldwide initiative to help people acquire digital skills, using data to identify in-demand jobs and the skills needed to fill them, providing free access to online learning paths as well as low-cost certifications and free job-seeking tools. Over 66.2 million learners in 249 countries and territories have engaged with digital skills content through this initiative. The AI for Good Lab’s involvement includes making data and analytics available to governments around the world so they can better assess local economic needs.

Addressing cybersecurity skills gaps in the workforce

2021 brought unrelenting headlines about cybersecurity attacks. Nation-state actors tampered with the software supply chain and hacked into sensitive government files; criminal ransomware groups attacked schools, penetrated hospitals and shut down a critical national pipeline. With this came the daunting realization that a serious shortage in skilled cybersecurity workers has hampered a strong response. The shortage is so severe that 3.5 million cybersecurity jobs are expected to be opened globally by 2025.

Microsoft has responded by launching a national campaign with U.S. community colleges to help train and recruit 250,000 people into the cybersecurity workforce by 2025. Since October 2021, collaborations have begun with 181 US community colleges across 44 states. The AI for Good Lab has contributed significantly by researching and providing data to help identify where the gaps lie, as well as data visualizations to help with understanding the results.  

The AI for Good Lab put together visualizations and a state-by-state dashboard (opens in new tab) to shed light on cybersecurity skills gaps in the U.S., to help traditionally excluded populations enter the cybersecurity workforce.  

The efforts were then extended to an additional 23 countries: Colombia, Denmark, France, Germany, India, Ireland, Israel, Italy, Japan, Korea, Mexico, New Zealand, Norway, Poland, Romania, South Africa, Sweden, Switzerland, and the United Kingdom. These countries have significant gaps in their cybersecurity workforces, both in terms of the number of and demand for cybersecurity professionals and the lack of diversity, increasing the vulnerability to cyber threats. We created a dashboard (opens in new tab) of these countries to better understand their cybersecurity gaps.

Addressing digital inequities

Education, healthcare, jobs, and other critical services increasingly require people to have access to a fast, affordable broadband internet connection. But more than one out of three people in the U.S. are not using the internet at what the Federal Communications Commission (FCC) defines as a standard broadband rate. Racial and ethnic minorities, people with disabilities and income-insecure families disproportionately lack a proper broadband connection. And many rural and tribal areas do not have the infrastructure to deliver broadband access.

This year, the AI for Good Lab partnered with Microsoft Airband to focus on developing a tool to empower policymakers to identify regions with digital equity gaps that can help direct funding and programmatic investments to the communities most impacted by the digital divide. By aggregating public data from the U.S. Census Bureau and the FCC with research from consumer advocate BroadbandNow, along with Microsoft’s own broadband usage data, we illuminate insights at the census tract level across 20 digital equity indicators, including broadband availability and usage, education and poverty rates. We seek to empower state agencies with easily understandable data as they consider how and where to address economic opportunity gaps.

(opens in new tab)

Identifying digital literacy gaps at the local level

Digital literacy is the ability to use technical devices to navigate the digital world, using reading, writing, technical skills, and critical thinking.¹ Individuals equipped with digital literacy skills can be better prepared to improve their livelihoods in a digital economy. Our team is currently working with top universities on a tool to measure digital literacy. This tool could be used by researchers and governments to gain further understanding of variables impacted by digital literacy and where to make investments to increase positive outcomes arising from digital literacy.

Expanding inclusion: Opening labs in the global south

The AI for Good Lab’s research shows that there are approximately 5 data scientists in the Global North for every 1 in the Global South – meaning a significant gap exists in the Global South’s ability to turn data into insights for decision-making and action. This dearth is even more severe in Africa, where for every 1 data scientist, there are 14 data scientists in the Global North. To help address this gap, Microsoft recently announced that it will open a $27 million facility in Nairobi, Kenya (opens in new tab). The AI for Good Lab is building upon this investment by opening its first two locations outside the U.S. in Nairobi, Kenya, and Cairo, Egypt, to develop AI talent, build solutions and new partnerships. The Lab will begin its work by addressing the climate crisis in Africa. These efforts are discussed further in the sustainability section.

Additionally, the Lab has recently partnered with the government of Uruguay on a new Microsoft AI and IOT Lab (opens in new tab) that will support startups in the region to deliver new AI innovations in climate change and other pressing areas.

Footnotes

¹ Digital Literacy | Microsoft

The post Inclusivity in the growing digital economy appeared first on Microsoft Research.

Global renewables watch: A living atlas to map earth’s solar and wind energy installations 

One way Microsoft’s AI for Good Lab is helping address this challenge is through a newly-announced partnership with Planet Labs PBC (opens in new tab) and The Nature Conservancy (opens in new tab). The project, called the Global Renewables Watch (opens in new tab), is intended to map the entirety of the world’s utility-scale solar and wind installations, and to measure the capacity of these installations using AI and high-resolution satellite imagery. It will allow users to evaluate the progress of every country’s transition to clean energy and track trends over time. It will also provide unique spatial data regarding land use trends to help promote “nature positive” development. This work will help researchers and policymakers make data-driven decisions in renewable energy development.

• Project Geospatial Machine Learning 

Supporting biodiversity and wildlife preservation  

• Project Accelerating Biodiversity Surveys with AI 

A quarter of the planet’s wild animal species are threatened with extinction; they must be preserved before it’s too late. The Living Planet Index (opens in new tab) (LPI)—which tracks populations of mammals, birds, fish, reptiles, and amphibians—reveals an average 69% decrease in monitored wildlife populations since 1970, and a decline of 83% for monitored freshwater populations. Wetlands have been reduced by 1/3 and coral reef habitats have declined by 50 percent.^2,3 The AI for Good Lab has worked on many projects that promote biodiversity by harnessing AI to analyze large amounts of data collected by conservationists. Remote sensing has become the most powerful tool for conservationists to estimate wildlife populations and understand how animals use their habitats. It is dependent, however, on tremendous amounts of manual data annotation. It can take years for a small NGO to annotate millions of images or audio recordings for a single project.

Our Lab applies machine learning tools to a variety of image sources—including motion-triggered camera traps, aerial cameras, and microphones—to accelerate the work of ecologists. For instance, in 2019 the AI for Good Lab partnered with the nonprofit conservation group Wildlife Protection Solutions, which had gathered a vast and overwhelming amount of image data using remote cameras to monitor the status of different animal species. The Lab helped them implement AI and machine learning tools to analyze and simplify the data.

• Project Bioacoustics 

We also work on this problem by taking large volumes of bioacoustics recordings and applying deep learning models to them. This can give bio-scientists profound insights into environmental features, such as the presence or absence of species in the ocean or rainforest. By automatically processing and analyzing bioacoustics data, we can cut detection and annotation time by as much as 93%.

There is an urgent need for acquiring information to help protect our wildlife and preserve our wetlands which have a major impact on carbon sequestration and biodiversity. Some of our current geospatial work includes detecting beluga whales from satellite imagery, wetland mapping and carbon measurement.

---

Footnotes

¹ Internal Displacement Monitoring Centre (IDMC.org) (opens in new tab)

² WWF: Living Planet Report 2022 (panda.org) (opens in new tab)

³ Global Wetland Outlook (ramsar.org) (opens in new tab)

The post Creating a sustainable future appeared first on Microsoft Research.