Research

Most of my work originates in a deceptively simple question: how do agents — animals, people, groups — gather information and make decisions in environments that are complex and constantly changing, and where their own behavior biases what they get to see?

The problems I find most interesting are the ones that seem way harder than they should be: situations where most people would benefit from a change, but the costs fall on a few, or come now while the benefits come later, and so the system stays stuck. I try to bring mathematical rigor to those questions without losing sight of the people and places at the center of them.

Conservation social science

I lead social science at The Peregrine Fund, a global conservation organization focused on birds of prey. As the organization’s first social scientist, I am developing and integrating social science research into our programs worldwide. I seek to understand why people interact with wildlife the way they do, and what it actually takes to change behavior meaningfully and sustainably. I design the research (household and attitude surveys, livelihoods assessment, monitoring and evaluation), supervise the in-country teams who carry it out, mentor an international group of early-career researchers, and work with program directors to build the results back into program design.

Current research includes:

• Human–wildlife conflict and the wildlife trade in East Africa and the Caribbean, including retaliatory poisoning of predators and scavengers, how much predator-proof livestock enclosures (bomas) reduce conflict, and the trade in wildlife parts.
• Lead-free ammunition adoption among U.S. hunters. Fragments from lead bullets left in carcasses and gut piles are a major cause of poisoning in eagles, condors, and other scavengers. Lead-free alternatives exist and perform well, yet switching among big game hunters has been slow. How can we accelerate this process through voluntary programs?
• Methods for reliable survey design. Attitude-survey data is often subject to substantial enumerator effects, i.e., which interviewer asked a question affects the answer. This creates major problems for accurately assessing the effects of social interventions for conservation. I’m reanalyzing existing survey data to measure these effects, and designing new surveys around two requirements: statistical rigor, and a realistic picture of what can be done in each country. This results in better measures of how our programs perform on the ground.

Earlier applied work

With Christopher Golden’s group at Harvard and a team on the ground in Madagascar, I studied households in the Makira Protected Area that depend on wild meat for food. We combined fruiting phenology, hunting records, and surveys of seasonal taste preferences to show how diet, hunting, and the forest change together over the year. That matters for any intervention that asks people to hunt less (Arehart et al. 2026, People and Nature, cover article). As a graduate student I served on Utah’s COVID-19 modeling task force (Meredith et al. 2021).

Theory

When does learning pay off? In my PhD I developed a minimal model of a forager learning the value of food types in an environment whose values and variances can change in any way. The model returns the optimal memory and exploration settings for a given pattern of change, so the fitness benefit or cost of learning can be quantified directly. One result: variance in resource value biases a forager’s estimates downward, which may help explain why animals often prefer less variable options (Arehart & Adler 2023; see also Arehart, Reimer & Adler 2023).

Collective learning and structural bias. Groups can be more accurate than individuals by pooling what their members know, but the pooling is vulnerable to bias. One postdoctoral project modeled populations of rational Bayesian agents who combine their own observations with what others share, under two hard-to-avoid biases: similar agents tend to talk to each other, and some kinds of information get shared more than others. These are enough to produce stable polarization, or a whole population misled. Worse, the conditions that make a population most responsive to the truth are the same ones that make it most vulnerable to polarization. The result is a structural tradeoff between accuracy and consensus (Arehart & Akçay 2025, preprint).

Forced discounting. I am developing the idea that structural conditions can raise the rate at which people discount the future, to the point where using up a resource is the rational choice rather than a failure of foresight. Manuscript in preparation.

Perceived variance. How do people make decisions about processes that are highly uncertain and play out over long timescales? Statistics can give us the optimal way to handle this, but most decisions are not made following statistical principles. I’m interested in how uncertainty enters the internal models people actually use.

AI and collective decisions

AI is changing how people get information, and the behavioral consequences are poorly understood. In conservation, new AI tools help track poachers but fail to address the motivations that drive people to poach. At the same time, AI opens a new frontier for behavioral experimentation and makes it possible to audit data quality at a scale that wasn’t possible before. Several problems now under study in AI alignment (reward hacking, preference learning, multi-agent dynamics) have close analogues in foraging theory and collective behavior.