NCS-FO: The Biology and Technology of Online Planning

Focus Areas: Cognitive and Neural Processes in Realistic, Complex Environments; Neuroengineering and Brain-Inspired Concepts and Designs; Data-Intensive Neuroscience and Cognitive Science. The ability to guide actions via a plan is one of the most impressive abilities to evolve in animals, often contrasted with stimulus-evoked varieties of behavior termed "reflexive," "reactive," or "habitual." Its importance in regulating the behavior of those animals able to plan is hard to overstate. In humans, the brain's capacities for planning are interwoven with cultural technologies-enabling achievements such as building York Minster over a span of 252 years-but even so a number of looming existential threats indicate that our ability to plan may be less than ideal to avoid their occurrence. Understanding how planning occurs-and augmenting its power with engineered artifacts-would be transformative to neuroscience and society. In prior work, the team identified the transition on to land in the Devonian as a key driver for a selective benefit of visually guided planning. Subsequently, they identified two factors key to its flourishing: a balance between having long sightlines to interrogate large swaths of space, and interruptions of those sightlines by obstacles that can hide adversaries. When the right balance occurs, as it does in savanna-like habitats, planning has its highest payoff. The team has a preliminary algorithm for automating planning based on these insights that is 10,000 times faster than a leading competitor in machine learning, and will develop and extend the theory and algorithm in initial work (Aim 1). They will gather behavioral evidence to test whether the theory solves two key mysteries of animal planning: how they do it so rapidly, and how they decide to switch off between planning and habitual responses (Aim 2). Finally, while modulating the planning load of computationally-derived complex habitats in a predator-prey dynamic with a robot adversary, they will record from key structures (value calculation areas of frontal cortex, hippocampus, and striatum) to assess causality through interventions including targeted inactivations (Aim 3). Integrative Value and Transformative Potential: The work spans from its founding domain on the paleontology of planning, through to ecology, computational neuroscience, and machine learning. It has high transformative potential in its use of a novel biological model system (a carnivorous rodent), novel theoretical frameworks, a completely new behavioral paradigm, and a novel outreach plan.