AI-enabled products come with their share of challenges. Here's how Vulcan partnered with Sama to use artificial intelligence for wildlife conservation.
AI-enabled products come with their share of challenges. Here's how Vulcan partnered with Sama to use artificial intelligence for wildlife conservation.Vulcan, the Seattle-based organization built by Microsoft co-founder Paul Allen, has a long history of supporting research and initiatives that make a global impact. Now the Vulcan Impact team is continuing its commitment to better protect wild plant and animal species and their habitat by using artificial intelligence for wildlife conservation.
AI-enabled products that can record and monitor African wildlife come with their share of challenges. In addition to requiring massive amounts of training data, the diversity of the data must account for species, landscape, cultural relevance and human influence.
“We ran into a problem when we were trying to detect cows in the imagery. We had a ton of pictures of cows from Washington, where we are, but cows look different in Africa,” Gracie Ermi, Research Software Engineer at Vulcan, pointed out. “Diversity in the dataset has been super challenging.”
Unmanned aerial vehicles (UAVs) have proven to be a viable way to capture large amounts of data, however, these aerial surveys result in countless hours of video footage that can make finding value in the data collected challenging.
If processed by humans alone, the work can prove to be mundane when there’s nothing of interest on the screen for hours on end. This is where machine learning proves useful, and the accuracy of the model depends on the accuracy of the data used to train the algorithm.To ensure the highest quality training data, Vulcan partnered with Sama, hiring a dedicated team of data annotators to put bounding boxes around key areas of interest in videos and images, and then pass the data back to Vulcan’s machine learning team to build various ML models.Ben Suidman, a Senior Program Manager at Vulcan, shared, “There was good clear communication about expectations and ease of use of the platform. Often, what looks like a hot stone sitting out in the sun may later turn out to be a termite mound once it’s cooled down.""We had to provide information about the dataset to Sama, letting them know a set of 2,000 - 10,000 images were collected with certain objects in mind. That’s how we directed them, and ultimately that’s how the system was successfully trained.”
It was imperative that Vulcan partner with an expert on training data annotation given that any mistakes could lead to inaccuracies in the ML model. The Sama team went through a training period aimed at delivering quality annotation at scale and developing subject matter expertise for Vulcan’s specific use case.“Once you have a trained team, it pays off because they know what to look for and how to spot certain objects in the imagery. Some objects just blend into the background in images that have been taken from a certain angle from the sky, so you need a trained eye to spot those particular objects,” Suidman said.To date, Sama has labeled over 600,000 images for Vulcan, having achieved a quality SLA of 95% in support of their efforts to use AI for wildlife conservation. With Sama’s help, Vulcan is able to expedite the processing of data collected from UAVs, without compromising on quality.
The Vulcan Impact team’s work to survey wildlife in sub-Saharan Africa is just a small fraction of the work being done for wildlife conservation.On any given day, there are multiple projects in flight, all working toward the ultimate goal to protect endangered species and ensure stable and thriving generations of wild animals.Vulcan’s effort to enhance remote identification of animals has the potential to make a huge impact on wildlife conservation, allowing monitoring to be done over a larger area and faster than can be done on foot, or even in vehicles.Additionally, by automatically detecting visual anomalies with artificial intelligence, Vulcan hopes to enable rapid-response human wildlife conflict and loss of habitat, and potentially use this technology to monitor ecosystems or update censuses of animal species.Vulcan’s efforts could also provide even greater situational awareness to rangers in the parks of Africa and other parts of the world, alerting them in advance of pernicious activities, so they’re better prepared to respond.
“We want to enable and empower agencies to have more confidence in the data that they’re using. If we can lower the uncertainty in the data, we can provide these agencies with higher confidence in the numbers so they can act on them,” Pooja Mathur, Senior Product Manager at Vulcan explained.
Vulcan sees the technology they are developing eventually being used for additional use cases, including predictive analytics to forecast how changes in our environment and ecosystem will affect the animals and wildlife around us.
The Vulcan impact team credits its success in developing technology products that make a global impact to its collaborative mindset.Pawan Nrisimha, Director of Product Management at Vulcan shared, “It’s not about competitive analysis, it’s about partnership analysis. In philanthropy, in tech for good, you are always looking for partnerships.""It’s about figuring out the puzzle and thinking about, ‘How can I fill in a hole, while somebody else fills in another hole.’ The most important thing is to build key partnerships to work together and not step on each other.”In working with Sama, Vulcan has significantly improved its turnaround times to process training data, allowing its algorithms to thrive.
“Having good data is crucial to any machine learning problem. Working with Sama has been great because not only are we able to get the data in the correct format to train our machine learning algorithms. There’s the added bonus of the impact Sama is making in people’s lives. It creates a win-win all around,” Mathur expressed.
Sama provides ML Professionals and AI team Leads with an indispensable solution for Computer Vision data labeling.
