Is humanitarian action leading the development of Microsoft's AI?
Blog|by Mary Branscombe|13 November 2018
AI is often in the news for scare stories about how many jobs it might replace, or for problems where machine learning systems replicate and amplify bias, but it can also be used to build powerful tools to improve the world. To support that, Microsoft is setting up a series of AI for Good initiatives that offer a mix of financial grants, access to Azure resources like Cognitive Services and machine learning VMs, support for preparing the data and labelling used in the project and expertise from Microsoft AI experts to help make projects a reality.
The latest, announced at Ignite 2018, is a $40 million, five year program to develop AI for humanitarian action: detecting and preventing human rights violations, improving disaster response and supporting the needs of children, refugees and displaced people.
Microsoft is already involved in the World Bank and United Nations’ Famine Accident Mechanism which is using machine learning to try and forecast when food crises are about to turn into famines.
AI can help with forecasting or detecting the early signs of disasters and with targeting aid; machine vision could spot the first signs of a natural disaster or detect which roads have been damaged or destroyed to route emergency response vehicles around them. One of the entries in this year’s Microsoft Imagine Cup student competition, Prometheus, used image recognition of NSA infrared footage to look for early signs of forest fires and send drones to high risk areas to take more footage that can be analysed.
Microsoft already uses predictive analytics and bot frameworks to try and understand human trafficking. It works with Operation Smile who uses machine vision to help surgeons choose techniques for individual operations, so they can treat more children. The company is also collaborating with medical researchers to use machine learning to create a genomic database for Sudden Infant Death Syndrome.
Researchers from Stanford and ETH Zurich used machine learning with years of economic data covering refugees in Switzerland and the United States to build a prediction model that can find the best places for refugees to settle to give them the best chance of finding a job.
Real-time speech translation could help people talk to lawyers and aid workers even if they don’t have a language in common. Microsoft is already creating a chatbot with the Norwegian Refugee Council, NetHope and University College Dublin, using LUIS, the language understanding cognitive service, speech recognition and machine translation to help displaced young people find free educational material that covers what they need to learn.
Another Imagine Cup project, SoulSailor, used similar techniques to build a chatbot to help refugees understand how to look for a job or get access to government services in their new country, to help them build a new life.
Microsoft has also announced the first project for getting a grant under the $25 million AI for Accessibility program. This is the second of its three AI for Good initiatives, following the AI for Earth program, and concentrates on using AI to improve communications and connections for people with disabilities, improve employment opportunities or lower the cost of technology that improves daily life. Zyrobotics is creating accessible STEM education materials like ReadAble Storiez, a reading fluency system that identifies students who need help learning to read but don’t have access to an occupational therapist, using custom speech models created with Microsoft Cognitive Services and Azure Machine Learning to customise their lessons.
The AI for Earth grants also show how the AI for humanitarian action program will work. AI for Earth started as a $2 million commitment and then increased to $50 million over five years and has already made 139 grants in 45 countries. They cover everything from tree species identification to using audio to identify which animals are in a particular area to understand whether conservation is succeeding to modelling how wild fires spread using reinforcement learning, to plant pest predictions to modelling ocean atmosphere dynamics, the impact of warmer water on plankton and the food chain, predicting ocean waves and creating a predictive data service that can give early warnings for floods and oil spills.
As the projects mature, Microsoft will make larger investments in the most promising projects – including ones that can become commercial services. It will also turn some of the AI advances made by investing in projects it supports into platform services developers can use. For instance, Microsoft supported the Chesapeake Conservancy who used Azure to create a more precise map of land use around the Chesapeake Bay to understand problems like sewage outflows and chemical runoff. Now, the organisation is working with Microsoft and geodata experts ESRI on a deep learning algorithm trained using Azure Batch AI to automatically analyse the land use in satellite imagery at a one metre scale. In time that will be available as an Azure Cognitive Service, but the collaboration with ESRI also resulted in the Azure Geo AI Data Science VM, which bundles up a range of AI, machine learning and data science tools with preconfigured Python and R interfaces to ESRI ArcGIS and Jupyter Notebooks for sample geospatial AI applications.
Working through the programs also makes it clear what’s needed for successful AI projects, like crowdsourcing and labelling data sets. That proved to be a problem for some AI for Earth projects so that program now includes grants for that, and they’re in AI for humanitarian action from the beginning. The data sets will be hosted on Azure and made publicly available for everyone to use.
As Lucas Joppa, Microsoft’s Chief Environmental Officer and head of the AI for Earth program, explained to us, “The great thing about machine learning is that a lot of the costs are up front. When you train, it’s about getting the data together and getting the model trained up. That’s where you need the human expertise. You have the costs of top machine learning talent, of data acquisition and of model training. There’s a huge amount of compute needed to train those models. After that, though, costs really drop off and what you’re left with is a model. A model is a bunch of math that you can run on a computer; all you need is to take new data and show it to that model and that gets very cheap, so we can commoditise the cost of this work.”
So not only is Microsoft funding specific projects that might individually contribute to a better world, it’s also helping to create data sets, tools, models and services that will be broadly useful to AI developers, whatever field they’re working in.
Contact the Grey Matter Technical Services team for an obligation-free discussion about Azure, and integrating AI and Cognitive Services technology into your application or project: +44 (0)1364 654100.
Contact Grey Matter
If you have any questions or want some extra information, complete the form below and one of the team will be in touch ASAP. If you have a specific use case, please let us know and we'll help you find the right solution faster.
By submitting this form you are agreeing to our Privacy Policy and Website Terms of Use.
