I am a undergraduate researcher at Indiana University Bloomington studying heuristics, decision-making, and artificial agents!
My research interests center on how agents (both artifical and biological) make adaptive decisions in complex environments. I am especially interested in heuristic decision-making, ecological- rationality, strategy learning, and the comparison between human cognition and artifical agents. Lastly, I am learning as much about modern AI and the engineering architecture that is used in these systems for future projects as well!
My longer-term goal is to research how agents acquire, modify, prune, and reorganize decision strategies across changing environments. I believe by distilling how humans do this, we can create systems that are truly adaptive and more generalized like humans are.
This my flagship research project. I've played a lot of league of legends and I've spent numerous hours coaching a range of players from beginners to experts (top 1%) and on reflection, my approach was always in the realm of "these are the decisions you should make in this situation". This led to my research question of: What information are experts attending to (or skipping), and how is this information weighted and then used to make a decision, and how does this compare to non-experts?
This is a project I'm also very excited for! This project is inspired by Google's Mleting Pot benchmark, which is a collection of multi-agent environments designed to study how artifical agents learn to interact in complex social settings. This is normally done with modern reinforcement learning agents that rely on large neural networks ranging from thousands to millions of parameters. My project takes an Occam's razor approach. I am using a minimalst continuous-time recurrent neural network (CTRNN) with just a handful of neurons. By evolving these simple agents in a capture-the-flag environment, I am investigating whether team based cooperative and competitive behaviors can emerge from a minimal neural machinery interacting with a structured enviornment. I believe by taking the inverse approach and using simpler systems it'll make it easier to identify the mechanisms responsible for the behavior. This design allows me to ask what the minimum requirements are for cooperation, competition, adaptation and other complex behaviors to emerge.
CV coming soon.
Email: nkurosky@iu.edu
GitHub: github.com/Nkurosky