Supervisors

Dr Benno Simmons, Centre for Ecology & Conservation, University of Exeter (Penryn Campus)

Dr Tatsuya Amano, Centre for Biodiversity and Conservation Science, University of Queensland

Join a world-leading, cross-continental research team

The University of Exeter and the University of Queensland are seeking exceptional students to join a world-leading, cross-continental research team tackling major challenges facing the world’s population in global sustainability and wellbeing as part of the QUEX Institute. The joint PhD programme provides a fantastic opportunity for the most talented doctoral students to work closely with world-class research groups and benefit from the combined expertise and facilities offered at the two institutions, with a lead supervisor within each university. This prestigious programme provides full tuition fees, stipend, travel funds and research training support grants to the successful applicants.  The studentship provides funding for up to 42 months (3.5 years) for both home and international students.

Eight generous, fully-funded studentships are available for the best applicants, four offered by the University of Exeter and four by the University of Queensland. This select group will spend at least one year at each University and will graduate with a joint degree from the University of Exeter and the University of Queensland.

Find out more about the PhD studentships click here

Successful applicants will have a strong academic background and track record to undertake research projects based in one of the three themes of:  Healthy Living, Global Environmental Futures and Digital Worlds and Disruptive Technologies.

The closing date for applications is mid-day Friday June 28th 2024 (BST), with interview to be w/c 29th July 2024 (tbc). The start date is expected to be Monday January 6th 2025.

Please note that of the eight Exeter led projects advertised, we expect that up to four studentships will be awarded.

Supervisors

Exeter Academic Lead: Dr Benno Simmons

Queensland Academic Lead: Dr Tatsuya Amano

THEME - Digital Worlds & Disruptive Technologies

Project Description

To address global biodiversity loss, we need to understand how biodiversity is changing. However, current levels of biodiversity change far exceed our monitoring capabilities. Traditional field surveys alone cannot fill the gap, and so scalable, passive monitoring technologies, like camera traps, bioacoustics, and eDNA, have emerged as solutions. Camera traps are the most widely used of these technologies. However, for camera trap data to be useful, any species in each image or video must be identified. Manual identification is slow and, thus, AI tools to automate the process have proliferated. These species identification AIs are increasingly being used by conservationists, as well as banks and businesses looking to comply with biodiversity net gain, profit from biodiversity credits, or transition to being nature positive. AI can produce revolutionary benefits, but it can also create harms if not developed responsibly. Species identification AI is in its infancy: the first seminal study was published five years ago; current models generalise poorly and can be highly inaccurate; and we have no understanding of how accurate these systems need to be, or the consequences if they are not. That AI is being deployed today, in the presence of these knowledge gaps, is concerning.

An unreliable AI could miss critical changes in species populations, resulting in missed opportunities for conservation and loss of crucial ecosystem functions and services. On the eve of the AI revolution in biodiversity monitoring, this PhD will deliver the first research into responsible AI for species identification. This work is urgently needed to harness the full potential of this transformative technology. This project is pioneering and interdisciplinary, at the interface of biology, computer science, statistics and responsible AI. It aims to ensure that models are 1) safe and avoid unintended harms; 2) free of biases that favour some species or taxa over others; 3) accountable to humans, offering explanations for their decisions. Responsible AI for biodiversity monitoring is an unexplored area, and therefore the project’s outputs will define the foundational frameworks and methods of this new field. Moreover, these approaches will be transferable to other AI-based monitoring technologies, such as eDNA and bioacoustics, ensuring broad relevance. The PhD will therefore have significant impacts across AI and ecology research; banks and businesses; policymakers and NGOs/charities. The student will benefit from access to a wide range of stakeholders — including WWF, Devon Wildlife Trust, and Natural England — giving direct pathways to impact outside academia.

Background: Businesses, NGOs and governments are increasingly using AI to monitor biodiversity at scale. Generally, AI is combined with a passive sensor technology, such as camera traps, satellites, bioacoustics, or eDNA. The most widely used of these is camera traps, where AI is used to identify species in images and videos in order to produce insights about ecosystem health. AI holds great promise, but also can cause harms if not developed responsibly. Camera trap AI is nascent and often highly inaccurate, yet is being deployed widely today. This is concerning given that we currently have no understanding of how accurate these systems need to be, or the consequences if they are not. Bad AI systems could miss species declines, resulting in bad conservation outcomes and misallocation of funds. This PhD will develop the first research into responsible AI for biodiversity monitoring, allowing the full potential of this revolutionary technology to be harnessed.

Aims, methods and deliverables:

O1: Recommendations for AI accuracy.We need to understand how accurate species identification AIs must be for reliable population trends. Using camera trap data, population trends will be calculated using unmarked models at varying levels of artificial labelling errors. The minimum labelling accuracy needed to avoid erroneous conclusions (e.g. trend in the wrong direction) will be determined across space and taxa.

O2: Developing a safety standard. The first safety standard for species identification AIs will be developed, incorporating the kinds of images that AIs most often get wrong, but which are important to classify correctly from a conservation perspective.The performance of leading species identification AIs will be evaluated against the benchmark and the benchmark will be publicly released to be adopted as a standard.

O3: A new evaluation metric Standard. AI evaluation metrics are not suitable for conservation as they treat all mistakes equally e.g. misclassifying an endangered species is equal to a non-threatened species. If the field has, as its north star, an evaluation metric that ignores ecological reality, then AIs will be chasing the wrong goal, with potentially perverse outcomes. This project will therefore develop a new evaluation metric, to be adopted as a standard, that weights misclassifications by phylogenetic distance and trait dissimilarities, like conservation status.

O4: Develop transparent, explainable Ais. AI models used in conservation and funding decisions must be explainable and accountable. Explainable AI techniques like Testing with Concept Activation Vectors (TCAV) will make AI decisions interpretable by relating them to human-understandable concepts. A Bayesian modelling framework will also be developed that ensures uncertainty from species identification is propagated through to the final insights like population trend. Expertise. Benno Simmons (primary) specialises in using AI and technology for biodiversity conservation. Tatsuya Amano is a conservation scientist with expertise in using statistical modelling approaches to overcome information gaps. The student will be primarily Exeter-based, using the JADE II HPC for computation, and will spend 12 months at UQ to work with Dr Amano to develop the novel statistical methods for O1 and O4.

Entry requirements

Applicants should be highly motivated and have, or expect to obtain, either a first or upper-second class BA or BSc (or equivalent) in a relevant discipline.

If English is not your first language you will need to meet the English language requirements and provide proof of proficiency. Click here for more information and a list of acceptable alternative tests.

How to apply

Apply now

You will be asked to submit some personal details and upload a full CV, supporting statement, academic transcripts and details of two academic referees. Your supporting statement should outline your academic interests, prior research experience and reasons for wishing to undertake this project, with particular reference to the collaborative nature of the partnership with the University of Queensland, and how this will enhance your training and research.

Interview notifications date TBC

Please quote reference 5154 on your application and in any correspondence about this studentship.

Owing to essential maintenance, our system will be unavailable between 17:00 BST on Thursday 27th June and 09:00 BST Monday 1st July 2024.

The application deadline has therefore be extended until 12:00 BST on Wednesday 3rd July.

Please accept our apologies for any inconvenience caused.

Summary