We launched our new Office Hours Programme this autumn with the aim of bringing expert #tech4wildlife advice from data scientists directly to conservationists through 1:1 advice sessions. With the support of Microsoft's AI for Earth Programme, we created a unique opportunity for conservationists both thinking about using AI and those already doing AI work to discuss with a data expert both what’s possible and what to try next, which proved to be a valuable experience for those involved.
From considering the hard problems and accuracy rate of thermally-triggered photographs of elephants in the Nilgiris reserve in southern India, to questions on how to automate species classification of bioacoustic and camera trap data in a variety of marine, coastal and terrestrial ecosystems, the office hours sessions provided a unique hands-on direct approach to solving a need within our community for clarity, direction and reassurance with their individual conservation projects.
One participant commented how the session introduced him to “some really great resources and options to explore in order to implement AI for my unique purpose, it was exactly the thing that I needed. It has really helped me along with my project by providing me the options and confidence to tackle the task at hand. Big thank you to all those involved and for all the advice provided.”
The idea for piloting this programme came out of a conversation our team had Dan Morris from the AI for Earth team. Together, we’d been looking at some of the AI questions coming through with applications to our On the Edge Fellowship programme. We were brainstorming ideas for other ways to support some of the projects and conservaitonists who were looking for help with apply AI to their work, but who were not quite at a stage where a year long fellowship and dedicated support was the right fit. Dan had the brilliant idea that we run some ‘Office Hours’ session with he and his colleagues. An Office Hours event concept had been something we’d been thinking about for a while, so we jumped at the chance to test it out.
We approached this series as a pilot, and, as with every WILDLABS venture, we let the community shape how the events looked. Feedback from the community saw a strong preference for a 1:1 event format over a group virtual room to discuss specific hard problems related to the application of AI in individual conservation projects. Out of those who signed up for an office hour session, four were applicants from the On the Edge fellowship programme who we reached out to thinking the opportunity would be useful for them, and three reached out to us from the wider community after seeing our announcements in October.
Ahead of the 1:1 office hour sessions, we held a briefing call spanning multiple continents where we created a space for the nine conservationists joining the programme to meet one another, Dan Morris from Microsoft, and the WILDLABS team. As with all our WILDLABS events, we valued the variety of expertise, skill sets and research areas of our attendees and faciltated a open environment where everyone introduced themselves and their work, and had time to ask questions to clarify the support they needed. Dan, informed by what he’s learned from advising lots of other conservationists over the years, then led a breifing session where he explained the questions to have in mind and data to have on hand to get the most out of an office hours session. These included a focus on the volume and general distribution of sample data, and helping Dan as a mentor to understand the team the conservationists are working with and its level of both ML and non-ML experience and expertise.
AI for Conservation Office Hours proved useful even for those who in the end decided not to use AI in their project. Lizzie Croose, who works for the Vincent Wildlife Trust in Herefordshire, England, was considering using AI to sort through camera trap images of mustelids (due to the high rate of data bycatch) but after going through her work with Dan was actually encouraged away from spending additional time training an AI model and given advice on alternative image processing shortcuts that bypassed training an ML model.
As Dan put it, “Sometimes 20 minutes spent setting up your windows and your keyboard shortcuts is worth a year of building an AI model; this felt like one of those cases.”
Meanwhile, many of those who were advised to pursue AI in their projects did so for similar time saving and efficiency benefits. For example, Ollie Wearn, the technical advisor for Fauna & Fauna International in Vietnam, and the two young Vietnamese researchers Le Hoang Phuc and Anh Minh Nguyen he works with, are looking to build a platform to classify the different endangered and critically endangered gibbon species across the country. With 53 AudioMoths deployed across the gibbons’ entire range in Vietnam, they have a lot of data and they’re aiming to apply a CNN / deep learning model to process the data into male and female, and individual male gibbon voices if possible. Clarifying the specifics of what they could realistically get out of ML as well as their technical capacity to achieve this, the offfice hours session helped them reach the conclusion that machine learning is ultimately most effective when the focus is on saving time, not on achieving 100% perfection. If there’s something a conservationist can reasonably do with data such as classify and catalog it, using ML to shorten the amount of time it would take that person to get the right data to classify is an extremely worthwhile goal.
The global reach of the office hours pilot was both a clear strength of the programme but also alerted our team to think of ways to support conservationists in remote areas where having a reliable internet connection is far from guaranteed. For example, in the session with Yoba Alenga Extasié, who works on acoustic monitoring of bats in the Ngandja Wildlife Reserve in the far eastern DRC near the Rwandan border, the internet connection where he was based during the session was intermittent and then unfortunately cut out completely. The network constraints mean we moved the office hours discussion first to the call chat, and then over to email between Dan and Yoba so he could get his questions answered and advice on further steps. Whilst a video call format has clear benefits, we recognise that it is not a catch-all and we’re continuing to explore ways of providing alternative but equal quality support in cases and places with reduced internet capacity.
Keep an eye out for Office Hours 2022!
Our Office Hours pilot demonstrated that having access to expertise and mentoring in a 1:1 scenario is a valuable opportunity for conservationists looking to take the next step applying tools like AI in their work. Overall, the programme was a success that we’re aiming to expand and replicate in the near future. Office Hours will be back in early spring next year, so watch this space and stay tuned for updates on the application and sign-up process.
Interested? Let us know in our Office Hours Thread what other topic areas you'd like us to cover in our Office Hours Programme, or to register your interest in the second round of office hours!