Cleantech
AI Helping to Ensure Clean Water, Improve Yields, Protect Forests

Artificial intelligence is as scary as it is exciting. On the one hand, it may eliminate vast job categories, while on the other, it will enable humans to do more interesting work and achieve incredible advancements that would have otherwise taken much longer.

Computer scientists at Intel are deploying their skills in this area, to protect the earth and we humans on it through a dedicated unit called AI for Social Good. They are developing applications for everything from classifying icebergs to monitoring whale health digitally through their snot. Seriously, check out the snotbot.

We caught up with Anna Bethke, Head of AI for Social Good at Intel, as well as several of Intel’s software innovators to learn more about their work and what kinds of breakthroughs to expect.

Head of AI for Social Good” has to be the sexiest job title we’ve ever heard. Can you briefly explain Intel’s approach?

Anna Bethke: Intel is committed to advancing uses of AI that most positively impact our world; and there are both groups and individuals across many of our business units that are applying the incredible work they do every day in hardware, software and algorithm development to achieving this goal. As the head of this cross-company initiative, I act as the glue between the Intel employees who are able to help, and those socially impactful organizations that need extra technical guidance to move their missions forward.

AI can have infinite applications. How do you select the projects you engage in and the partnerships you develop? Is the cost effectiveness of using AI, over more traditional forms of technology, part of what’s considered? Are these projects philanthropic or commercial?

AB: Intel uses AI capabilities to impact the world in two ways. One is to support social good organizations with AI technologies and expertise to accelerate their positive work in the world. For example, we worked with nonprofit RESOLVE to create TrailGuard AI — an AI technology to detect poachers entering Africa's wildlife reserves; and alert park rangers in real time, so poachers can be stopped before killing endangered animals. RESOLVE is partnering with National Geographic Society and the Leonardo DiCaprio Foundation to deploy the TrailGuard AI cameras this year. The second is to support research efforts to ensure AI is more transparent, less biased and more accessible to all. An example you're familiar with is Peter Ma's Clean Water AI system to detect harmful bacteria in the water in real time. The organizations we partner with often see AI as being more cost effective than other, more traditional forms of technology; and when AI isn’t the best solution for whatever reason, we always advise organizations to seek the best solution for them. We partner with both nonprofit and commercial organizations, as both types of organizations have the very real capability to positively impact our world.

Let’s look at a few examples:

Rosemarie Day, Intel Software Developer:

You have developed a model to retrieve data from LandSat satellites and other data sources to monitor deforestation using TensorFlow. Your model can be used to classify plants, and analyze deforestation and growth of plants across different regions and over time. Who will use it, how will they use it, and what are some examples of decisions they will be able to make as a result?

Rosemary Day: Over time, I plan to adopt a wider data set used to classify plants and analyze deforestation across the globe. I’m hopeful that governments and large corporates will look to partner with one another to help make decisions that can change the world — like combatting climate change and better predicting weather patterns for susceptible groups of people. It is something that is impacting us all and that we all need to think about. 

Peter Ma, Intel Software Developer:

You have used AI to test water samples and show the level of contamination by placing drops of water under a digital microscope and connecting it to a screen. Can you explain the AI application and how it enables cheaper, faster or easier water testing than tools that exist today?

Peter Ma: Our current prototype places water drops under microscope and connects it to an Intel Movidius Neural Computing Stick. By doing this, we can inference the camera image and classify different bacteria through the deep learning neural network in near real time. Our next iteration of the prototype will automate the water sampling process, so that our clients can connect it directly to the water sources. Current water testing mainly relies on chemical strips — a method that is effective but slow, as we have to wait for chemical reactions. But if a new type of bacteria emerges, it would cost millions to develop and deploy a new type of chemical strip. Because Clean Water AI is optical, and it relies on image data to train neural networks, it makes detection of new types of bacteria much easier. It runs on the $79 Movidius Neural Compute Stick, a cost-effective solution that turns a low-cost laptop into a high-powered computer engine, capable of advanced AI capabilities, that can often cost thousands of dollars. Long term, this project is slated to productize to scale into a wide variety of industries and third-world countries.

Risab Biswas, Intel Software Innovator

Your goal is to help farmers identify diseases affecting their tea and other plants, so that they can avoid crop losses. How are you using AI to address this problem? Will farmers be able to afford the technology, and what percentage of their yields might it save them?

Risab Biswas: We can use AI applications to find and identify diseases in plants by using the advanced capabilities of image recognition and classification. Long term, my vision for this project is to scale the use case to develop an app, and eventually drones to find and identify diseased plants. Currently, we have a 95 percent accuracy rating for identifying plant disease, thus based on this data I'm confident the percentage of yields will improve with this technology. More specifically, according to our experimentation on both labs and the farms, we have statistically calculated that the yield can be increased to 30-40 percent. Moving forward, we have a planned pilot where we will be testing the system to determine a more accurate calculation.