Humans do team well while AI agents do not. One example of this is the Dota 2 AI Open AI Five: Five AI agents would be in the game as five top Dota 2 players in the world, and they were broken into different teams with other human agents to fight against each other. However, the agents do not collaborate well. The overall goal of the project, STRONG (Strengthening Teamwork for Robust Operations in Novel Groups), is to probe the future of HMT (human-machine teaming) and to develop natural task general human-guided machine learning capabilities for future scenarios of teaming.
Here are three strands for this:
Strand 1: Design New Game - Speculative Game Design for Futuristic HMT. We’ll be looking at the idea of speculative game design - we think about not just how teaming looks now but also in the future.
Strand 2: Semantics of Human-Guided Machine Learning. One can do Interactions (rewards and punishment), interventions, and demonstrations to guide the agent which is based on machine learning models, like the Q-learning-based one.
Strand 3: Human-Guided Machine-Learning Agent Development. We’ll be looking at how to build agents that could adapt what the means ones want to teach on the fly and apply them across multiple games.