discussion / AI for Conservation  / 28 April 2021

Which are major open challenges that AI can help to solve in nature conservation ?

Years ago I worked in innovation management on the international developement sector. 
Institutional governance in uptaking new tech was quite slow.. However I now see that cost to embrace cutting edge tech got slower, and I sense there is a surge of tech for wildnerness. Is it so?
Which areas AI and data-driven solutions may mostly be helpful ?

I am posing this question for I am interested in:
- career opportunities in applied tech / tech management, offering also on-field activities
- research opportunities (PhD) where I can bring experience in international dev, startup and product innovation, envisioning use of tech (finalist at global innovation challenges)

Would like to include in your answer an introduction to most urgent problems to solve ?
Are these problems due to lack of funding ? to capacity building ? to lack of data ? are opportunity driven ? How do you see those evolving in the next 5 years ?

I 'd like to hear anwers also from conservationists and technologists. Thank you!

 




With the lowering cost of electronic hardware that can collect large amounts of data, such as Audiomoth devices for audio, and various consumer-friendly camera trap hardware for images; there is an increase in demand to parse through and extract meaning from these evergrowing datasets.

The vague term "meaning" can include things such as intrusive anthropogenic activity such as poaching and illegal deforestation. It can also mean searching for endangered species, helping sustainably managed logging/mining operations to determine best practices in their field, and monitoring population densities across different species.
 

People involved in the image recognition field face the challenge of going through the process of training models for a deployment region, then not having the models translate well into a different deployment region. I have seen similar problems related to drone imagery capturing baboon troop activity as well.

I personally work in the passive acoustic monitoring field where people go out and deploy low-cost audio recorders for months on end collecting terabytes of audio. One of the largest challenges is reaching a place where one can be confident in the extraction of some aforementioned "meaningful" data. For instance, finding people that can reliably label audio for specific species (such as birds) in a remote location such as the Amazon rainforest is challenging. This requires us to try and rely on resources such as xeno-canto which usually leads to models that do not translate that well into field recordings.

I think there is also a big gap for AI in Decision Support Systems in managing human-wildlife conflict and other conservation challenges that can be described as "wicked problems" (https://en.wikipedia.org/wiki/Wicked_problem).

Hi there,

As background, I manage several wildlife nature parks on behalf of the Wildlife Bank Trust in Australia (https://wildlifebank.org.au/) and have been collecting a range of datasets relative to the parks.

In terms of your query regarding "which areas would benefit from further AI and data-driven solutions", I'd add the following as two that would benefit from software to facilitate the automated identification of:

- flora/botanical species in digital imagery (both from camera and drone footage) and was wondering whether you had come across such libraries as part of the MegaDetector project; and,

- animal species from drone-captured video streams, particularly from thermal imagery/heat signatures.

Kind regards,

Miles

Have a look at https://www.hackfornature.com/

We have a bunch of challenges that requires AI.

The online hackthon is nearly finished but we will be having a V2 next year after IUCN