Project summary

As we know that the leading AI models are getting integrated into the decision-making, planning, by the agentic workflows, their recommendations directly will affect the shape land development, automated agriculture, and resource extraction. Non-sensitive planning can easily lead to land-clearing, seasonal burning, and industrial agroforestry that crush or displace the natural habitat of wild animals (ground-nesting birds, reptiles, burrowing mammals, and insects).

We will build a specialised system and a dataset of synthetic ethical reasoning traces. Using the UK AISI Inspect framework, we will audit whether frontier models identify wild animal suffering in real-world. We will do it by planning scenarios and providing mid-training/fine-tuning demonstration data to improve their moral reasoning.

What are this project's goals? How will you achieve them?

We will try to curate 50 realistic, high-stakes decision scenarios across land clearance, irrigation routing and automated pest exclusion. We will also work on the ground scenarios in actual agro-ecological trade-offs (e.g., timing brush clearance around stone-curlew nesting periods, adopting non-lethal acoustic displacement over disking, weighing thermal night operations vs. day operations). We will evaluate leading models like GPT-6, Claude, Sonnet, Gemini, and Llama) along established ANIMA dimensions. These dimensions are Moral Consideration, Contextual Welfare Salience, Harm Minimization, and Trade-off Transparency.

We will create synthetically augmented Chain-of-Thought traces. These traces will help us to demonstrate how an advanced, competent planner naturally balances commercial objectives with non-human sentience without refusal. We will also format these traces for integration into open-weights models and release them freely on Hugging Face.

How will this funding be used?

Researcher Stipend (12 months part-time/dedicated student researcher): $15000
API Credits & Compute: $4,000
Expert Domain Review Expenses: $3,000

Who is on your team? What's your track record on similar projects?

I am an academic living in Northern Cyprus for the past 4 years. I am going to kick-start it with my student volunteers and then will establish a team once funded. We have been doing projects in the same line and need funding to develop and deploy them in the hatchling nests.

Reference (Muhammad Toaha Raza Khan)

What are the most likely causes and outcomes if this project fails?

The problems can be technical, where models can simply adopt a "safe refusal" persona (e.g., "I cannot recommend clearing this land because animals might be harmed") rather than engaging in realistic operational trade-offs. For these cases, we will need to explicitly adopt ANIMA's scoring methodology, where simple refusals score poorly, and points are awarded only for actionable, cost-conscious harm reduction.

The generated land-use scenarios can be too trivial or diverge from real scenarios. For this, we can allocate some budget to external domain review by wild animal welfare researchers to audit scenarios before running final evaluations.

How much money have you raised in the last 12 months, and from where?

We have very recently secured a number of international research funding awards through competitive processes, among them grants from the International Science Partnerships Fund (amounting to £80,000), the National Research Foundation of Korea (£450,000), the Québec Merit Scholarship programme (£35,000), and the AdımODTÜ research programmes.