AI shows its wild side
A world of AI
Last month we attended the H2O AI World event, which showcased a variety of training sessions alongside talks on how AI is being used today. Throughout the two-day event we heard lots of interesting case studies, however, there was one in particular that stood out – “Wildbook”. Wildbook is a concept which blends together AI and Biology to help measure the population of endangered species by giving them their own ‘Animal Facebook’ account.
A Blend of AI and Biology
The first step in aiding endangered species is having a complete view of the current situation: population size, sex/age ratios, geo-locations and life span to name just a few key factors. Previous data collection methods were flawed, with large amounts of assumptions and estimation.
A new way of collecting this data (via photographs) was proposed by Wild Me, the charity behind Wildbook, which engages both local and global communities. These photos are processed by AI to assess unique identifiers which establish and isolate individual animals of each species. This method of photo collection differs greatly from stationary cameras as it permits a wider scope of the land in real time, while also avoiding the risk of being eaten by elephants!
Photographs can also provide additional vital information which can help with future analysis of a species if the experiment is repeated in the future for example location and health patterns of individual animals.
A Black and White Barcode
This notion was tested in Kenya on critically endangered Grevy’s zebras where an estimated 98% of the population live. In 1970 it was estimated that there were over 15,000 Grevy’s zebras living in Kenya, this figure reduced dramatically to 4,000 by the end of 1980 falling to just 2,000 at the start of the new millennium. These figures raised a variety of questions: what happened to these zebras? How accurate is the current method of measuring a population?
For one weekend in January 2016 Wild Me brought together 350 volunteers consisting of Conservation Managers, Country Government Officials, Pastoral Livestock Owners, and Academic Scientists. This experiment was conducted across 25,000 KM2 which was divided into 45 blocks, as volunteers split into teams to take photos of as many Zebras as they could find in their given location.
The experiment generated 16,866 individual images which were then processed by AI, analysing specific patterns in the animal genetic markings to identify and name 1,942 unique individual zebras. A statistical model of 95% confidence was then calculated to estimate that the actual population of Grevy’s zebras in Kenya was 2,250 (+/- 93). The model attempts to account for small areas which were excluded from previous experiments due to historically being out of the range of Grevy’s zebras or inaccessible/dangerous areas.
Alongside providing a strong population estimate, the images also showed that around 30% of the population were infants and juveniles. This indicates that the population appears to be sustainable with the adult death rate compensated by sufficient births.
Analysis on the geo-locations depicted that Laikipia (which historically wasn’t a natural part of the Grevy’s zebra range) now hosts over half of Kenya’s Grevy’s zebra population due to healthy rangelands. This data supports original concerns that as the human population grows, the natural habitat for Grevy’s zebras declines as competition for water/grazing resources increases.
In light of this study, the Kenyan government have amended the figures of the ‘Official Zebra Census’ to support these findings. In addition, from these findings, Wild Me have recommended a number of actions including restoring grasslands, improving water access and developing wild-life friendly infrastructure.
A Treasure Map for Poachers?
This study provides a vast amount of information to aid with the prevention of the extinction of Grevy’s zebras however, is it possible that this could backfire? This experiment highlights the key areas where the Grevy’s zebra are currently located making it easier for poachers to find which in turn could aid (as opposed to prevent) the extinction of these animals. The protection of this information is critical and needs to be a strong consideration for Wild Me. Strong data security is key as well as strict vetting before granting anyone access to more in-depth information on the locations of animals on the platform.
Wildbook is continuing to broaden its archive of animals who have their own account, extending the range of species available. Wild Me AI Programmes currently scrape the internet to find and process pictures/videos to produce more animal profiles. AI programmes are constantly improving with more and more animals being able to be uniquely identified, for example, a current project can recognise elephants based on the shape of their ears!
As technology increases the future potential for this project is exponential, with the exciting prospect of drones to aid this experiment could improve accuracy and range even in a shorter time frame.