SBIR Phase I: Machine Learning Actors to Improve Connectedness across Remote Teams

这个小企业创新研究（SBIR）第一阶段项目的更广泛的影响/商业潜力是开发机器学习驱动的参与者（ML参与者），促进朋友、陌生人、同学和同事在虚拟世界中用户生成的空间中的社交接触。向虚拟工作、学习和社交的转变伴随着重大的社会混乱。在过去的几年里，美国各地的人们报告说，他们的孤独和孤立程度越来越高。研究表明，游戏是团队建设的强大工具，非玩家角色对建立同理心有重大影响，该项目使用ML演员作为免费、多人、合作游戏的基石，旨在将远程工作者聚集在一起。这个小企业创新研究（SBIR）第一阶段项目旨在解决使机器学习参与者在用户生成的世界中可行的挑战。为了在虚拟世界中发挥作用，机器学习演员需要驾驭不熟悉的环境、玩家对话和难以预测的行为。角色需要在大量的数据上进行训练，并有一些人类的监督。该项目旨在证明，没有技术背景的用户可以从大量数据中训练机器学习参与者，然后将这些参与者部署在虚拟环境中，在虚拟环境中，他们可以对环境和玩家的选择做出反应。这个项目有三个主要步骤：1)从成千上万的电影剧本和游戏日志中学习一个大型的多模式分层任务网络，2)将该模型连接到3D环境中的角色，3)与远程团队一起测试游戏以衡量功效和乐趣。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact /commercial potential of this Small Business Innovation Research (SBIR) Phase I project is to develop machine learning-powered actors (ML actors) that facilitate social encounters between friends, strangers, classmates, and coworkers in user-generated spaces across the Metaverse. The shift towards virtual work, learning, and socialization has been accompanied by significant societal disruption. Over the past few years, people across the United States reported increasing levels of loneliness and isolation. Building off research that shows games are a powerful tool for team building, and non-player characters have a significant impact on building empathy, this project uses ML actors as the building blocks of free-to-play, multiplayer, cooperative games designed to bring remote workers together socially. This Small Business Innovation Research (SBIR) Phase I project aims to address the challenge of making ML actors viable for user-generated worlds. In order to be effective in the Metaverse, ML actors will need to navigate unfamiliar settings, player dialogue, and behaviors that are hard to predict. Characters will need to be trained on vast quantities of data with some human supervision. This project seeks to prove that ML actors can be trained from large amounts of data by users of no technical background and those actors can then be deployed in a virtual environment in which they are responsive to their environment and player choices. This project has three main steps: 1) learning a large multimodal hierarchical task network from thousands of movie scripts and game logs, 2) connecting that model to a character in a 3D environment, and 3) testing a game with remote teams to gauge efficacy and enjoyability.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.