When artificial intelligence (AI) researcher Professor Milind Tambe learned at a World Bank summit in 2010 that poaching was going to wipe out rhinos, elephants and tigers, he realised his research could help rangers predict where poachers were going to set traps. “Too often the public sees AI as a great unknown and something to be feared, but actually it has extraordinary potential for societal and environmental wellbeing,” says Tambe.
Tambe, who is Director of the Center for Research on Computation and Society at Harvard University and Director of AI for Social Good at Google India, wants AI researchers interested in social impact to reject the notion that their work finishes when a research paper is published. “Often knowledge from research isn’t actually deployed in the field or used for large-scale field studies. As a result we don’t quite encounter the real problems being faced.
And if we don’t know what problems need to be solved, we’re not able to identify AI that needs to be created.”, he explains.
Following the World Bank summit, Tambe met with the Wildlife Conservation Society (WCS) in Uganda. There he was able to see for himself the challenges facing the rangers – the discrepancy between the staggering size of the area and the available resources. “Because they only patrol small fractions of these vast parks on any given day or month, they didn’t have a lot of data specifically on poaching incidents. As a result, their patrols were based on well-informed guesswork,” he says.
To help the rangers concentrate their efforts where they’re most needed, Tambe and his team developed an AI software package called PAWS (Protection Assistant for Wildlife Security). It uses behaviour models, game theory and machine learning to predict likely poaching hotspots, even in areas where there have been minimal patrols and, subsequently, little is known about what’s happening there.
Since they weren’t patrolling those areas, we could assume that they didn’t think those areas were important to patrol in,” says Tambe. “But our AI modelling pointed them there. They found a poached elephant. After that they found elephant snares. So before the poachers could kill more elephants, the rangers were able to remove these elephant snares. That was a very fulfilling moment, for my team and myself,” he says.
In most recent trials, rangers using the PAWS predictions saw anywhere between 5 to 20 fold increase in numbers of snares captured.
Because of its success in Uganda and more recently in Cambodia, PAWS is now set to be rolled out to more than 800 protected areas in over 60 countries. It is also being integrated with the database system, SMART patrol information system, Which was created by a partnership of 13 NGOs in wildlife conservation. “This means rangers around the world can download our software and use it to make predictions. Right now, people in different parts of the world are downloading the software, testing it in their park, and seeing how it performs. Because each country, each park has its own terrain, wildlife and challenges, each of these needs to be fed into the AI algorithm to make it effective,” he explains.
“And once we are past this testing stage, then the gates will be open so everybody can start using it to make predictions about poaching incidents, which I think is a massive change,” he says. Tambe also sees a wider application for PAWS in combating other environmental crimes including illegal fishing and logging.
Tambe’s first foray into AI for social good started in 2007, when he created a pioneering AI security program to assist LA airport in optimising security patrols. The success of this program, which has been replicated by other agencies focused on public safety, including the US Coast Guard to patrol major ports in the US, awakened him to the impact he could have. “It changed policy. It changed the way they were operating. And after that, I have not been the same. Until then I felt like I was playing by the rules of what other AI researchers wanted AI research to be. Once I saw that we could actually directly apply things in the field and it made a difference, I just said, ‘Okay, this is the way I want to conduct research from now on.’ I don’t want to ever go back to the way I was doing things,” he says.
Looking across the research landscape, Tambe sees an enormous opportunity for AI researchers to embed social good within research.
There are billions of people, particularly in resource-poor countries, who have not seen the benefits of AI. We’ve got these brilliant AI researchers all over the world, but they don’t have access to the non-profit world, who have problems they could solve. If we could connect the non-profits to the researchers, I feel like a big explosion can happen in terms of allowing people to work on those problems,” he says.
In his role as Director of AI for Social Good for Google India, he sees this challenge as a critical part of his work. “Last year we held a workshop where we invited a small number of university researchers and non-profits. The non-profits pitched their problems and the university researchers pitched their experiences and skill sets, then we ran a matchmaking session. In the end we were able to fund six of the 25 projects people pitched to us from those matchmaking sessions and those projects have been really successful,” he says.
Tambe is particularly proud of one of the collaborations, which has focused on helping Armman, a non-profit that runs health programs with expectant and new mums in low income communities. By using AI to analyse phone call patterns including which calls are being listened to and for how long – they are able to predict when mothers are about to drop out of their health program, with an accuracy of 80%. The non-profit is now able to dedicate the resources to develop targeted intervention plans to increase positive healthcare outcomes for mothers and their babies.
Tambe and his team at Google India are planning to run these workshops across several countries. “Now it’s going to be a bigger thing, not confined to India. Any non-profit in Southeast Asia, India or Africa can apply and researchers can apply from all over the world. We will do the matchmaking,” he says.
Successful proposals from the matchmaking sessions will receive initial funding of USD$20,000 to the non-profits and USD$10,000 to the researchers in research grants from Google. Where there is success, Google will continue to provide funding support.
For Tambe, this matchmaking is where the real difference can be made globally. “Governments around the world could do so much more to encourage AI researchers to work with non-profits. The amount of funding and so forth needed may not be that much. This kind of workshop can be run all over the world.”
Working like this would lead to a massive change in the way we will impact the planet with our work,” he says.
Follow Milind Tambe on Twitter | LinkedIn | Facebook
Article by Kylie Ahern
Photo supplied
Comments are closed.