Research Experiences for Undergraduates (REU) Participant Support Costs Supplement for NCS-FO: How Ecology Induces Cognition: Paleontology, Machine Learning, and Neuroscience #SP0049192

Project: Research project

Project Details


About 385 million years ago, fish evolved into four-legged land animals. We have recently shown that just prior to this transition, eyes tripled in size and moved from the sides of the head to the top. Combined with computational visual ecology, these changes in eye size and position show that transitional animals viewed scenes through air while living in water--like crocodiles--and gained a million-fold increase in the space of visual awareness as a result. Unlike their purely aquatic predecessors, these animals were able to see further ahead and therefore plan before they had to act, providing a selective benefit to the animals that evolved the ability to plan. Concurrent work in comparative neurobiology has shown that brains increased in size and complexity during the ascendance of land animals. A useful gloss is that the pre-terrestrial brain is well suited to high speed responses to nearby sensory input, while the brains of birds and mammals are well suited to slow responses to distant sensory input, mediated by learning and an ability to imagine future scenarios and pick the best. We tend to think of nervous systems as the means by which an animal organizes its world, but this deep time perspective suggests that it is rather the world of an animal that organizes its brain. In this case, a particular change in evolutionary ecology favored cognition over reactivity. We propose to use a combination of a computational analysis of planning in predator-prey interactions and an empirical study of key circuits that enable animals to transition from the gapless relationship between sensation and reward in reactive environments to highly gapped relationship between sensation and reward in environments that favor planning. Our computational framework for this work will be partially observable Monte Carlo planning with reinforcement learning. The empirical framework will be in-vivo imaging of mammalian brains during simulated predator-prey interactions within a virtual reality apparatus, where the entropy of the virtual world is calibrated by the computational analysis. Through these efforts we will gain insight into the ecosystem-agent features that maximize the utility of planning. Resolving this to computational theory and biological mechanism will enable us to identify the characteristics of environments in which cognitive augmentation empowers new levels of performance, and what feedback needs to be delivered to achieve enhanced cognitive performance.
Effective start/end date9/1/188/31/23


  • National Science Foundation (ECCS-1835389)


Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.