Leveraging AI for social good

The hardware, software, and algorithms that automatically tag our images or recommend the next book to read can also improve medical diagnosis and protect our natural resources. Jack Dashwood and Anna Bethke discuss a variety of technical projects at Intel that have enabled social good organizations and provide guidance on creating or engaging in these types of projects.

There’s a growing desire to bring autonomous algorithms and systems into all aspects our daily lives to make our jobs, chores, and downtime easier and more enjoyable. At the same time, there are many individuals in our society who are impoverished, in danger, gravely ill, or otherwise generally requiring assistance. AI can help this situation by detecting bacteria in water quickly and accurately, identifying children who are at risk of being a victim of sexual trafficking or exploitation, helping a doctor identify cancer and diseases more quickly, and developing drugs in a more cost-efficient manner. AI is not limited to helping people in need. We can use AI to study and protect wildlife, help restore historical landmarks, and monitor our planet. AI is being used to increase crop production with a reduced amount of resources, to help feed a planet of more than 7 billion. It can also be used by first responders to study our cities and our surrounding environments to plan for and respond to disasters, saving countless lives in the process. Jack Dashwood and Anna Bethke discuss a variety of technical projects at Intel that have enabled social good organizations and provide guidance on creating or engaging in these types of projects. One example of AI for social good is the TrailGuard camera, a smart camera device that utilizes deep learning based workloads to detect poachers in wildlife reserves such as the African Grumeti. Improving dramatically upon classical motion sensors thanks to modern object detectors, the TrailGuard camera is able to avoid common false positives such as animals moving through the camera frame, but correctly flag humans in off-limits areas of the reserve. One major challenge of these “camera traps” is the risk during installation and maintenance. Conservation staff are at risk any time they are in the field setting up or maintaining these devices, in addition to the risk that the location of these camouflaged devices being exposed during this work. To minimize risk due to frequent maintenance, Resolve decided to implement its deep neural networks on the ultralow power Intel Movidius Myriad 2 vision processing unit with the hopes of taking battery life from an approximate two months to a targeted 12 months for the new design. In order to achieve this ultra-efficient design, Resolve and Intel engineering had to design a low-power hibernation mode in addition to an efficient sleep/wake capability in the device. Jack and Anna discuss the power saving design decision made to enable the device to operate on a battery while deployed in the wild for months at a time. Miniaturizing the device was also crucial to successful camouflaging. Reduction of PCB design to just 4.5 inches long by 0.5 inches wide means the cameras can be convincingly embedded into tree bark and rocks without detection. Jack and Anna contrast early designs against the deployed design to show how new silicon solutions with dedicated inference capabilities are enabling much smaller form factors, with fewer components, and ultimately lower cost—crucial factors for nonprofits relying on the funding of other organizations. Lastly, the TrailGuard program is growing in number of deployments around the world, and future versions are being developed to go beyond poacher detection but actually flag when rare species of animal make an appearance. Join in to learn about the data considerations of the black rhino for example, where image datasets of the creature are tightly held by governmental bodies trying to preserve this endangered species.