SBIR Phase I: A Human-Aware Platform for Socially Collaborative Personal Artificial Intelligence (AI) Assistants

这个小企业创新研究(SBIR)第一阶段项目的更广泛影响是使人工智能(AI)助理变得主动，使他们能够为用户提供更好的服务。目前，商业人工智能助手对用户的请求做出反应。该项目开发的技术将为人工智能助手提供情景感知，以了解用户的生活并预测他们的需求。该技术还将使社交智能能够以适当的方式主动支持用户。这个SBIR第一阶段项目将把这些技术应用到一种消费产品中，帮助用户进行时间管理和实现目标，同时在他们的日常生活中建立和加强健康、令人满意的习惯。积极主动的个人AI助手有可能提高每个人的生产率、便利性和生活质量，并以更大的独立性和健康促进适当的老龄化。根本性的科学进步还将使新一代人工智能助手在各个行业的潜在应用成为可能，推动经济增长并创造就业机会。该项目解决了启用主动式AI助手的两个核心技术挑战：用户的上下文感知和代理发起的交互。情境感知包括AI代理对当前用户状态和活动的实时理解，以及对过去用户习惯的长期理解。该项目建议开发混合计算模型，将来自视觉、声音和地理位置数据的多模式用户观察的机器学习与对历史用户观察执行长期推断和预测的概率图形模型相结合。虚拟体现的AI代理将利用这些上下文感知表示来与用户进行实时、面对面的协作。该项目提出研究和开发一种动态调度方法，使代理能够主动地与用户进行通信。这些模型将集成到一个更广泛的系统中，以帮助用户进行时间管理。该系统将实施端到端架构，在处理用户数据的同时保护用户隐私。技术解决方案将基于与效用和用户接受度相关的量化指标进行验证，方法是在最终用户家中部署原型，为期数周，并对他们对主动人工智能助手的主观体验进行调查。该奖项反映了NSF的法定使命，并已通过使用基金会的智力优势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 of this Small Business Innovation Research (SBIR) Phase I project is enabling Artificial Intelligence (AI) assistants to become proactive, empowering them to provide better service to users. Currently, commercial AI assistants respond to user requests reactively. The technologies developed in this project would provide AI assistants with the situational awareness to understand users’ lives and predict their needs. The technology will also enable social intelligence to take the initiative to support users in appropriate ways. This SBIR Phase I project will apply these technologies to a consumer product for assisting users with time management and meeting goals while establishing and strengthening healthy, desirable habits in their daily lives. Proactive personal AI assistants have the potential to improve productivity, convenience, and quality of life for every person, as well as to promote aging in place with greater independence and wellness. Fundamental scientific advancements will also enable a new generation of potential applications for AI assistants across sectors, fueling economic growth and creating jobs. This project addresses two central technical challenges for enabling proactive AI assistants: contextual awareness of users and agent-initiated interaction. Contextual awareness includes the AI agent’s real-time understanding of current user state and activity, as well as a long-term understanding of past user habits. The project proposes to develop hybrid computational models combining machine learning of multimodal user observations from visual, acoustic, and geolocation data with probabilistic graphical models that perform long-term inference and prediction over historical user observations. A virtually-embodied AI agent will leverage these contextual awareness representations to conduct real-time, face-to-face collaborations with users. The project proposes to research and develop a dynamic scheduling approach to proactively enable the agent to communicate with users. These models will be integrated within a broader system that assists users with time management. This system will implement an end-to-end architecture for protecting user privacy while handling their data. The technical solution will be validated based on quantitative metrics related to utility and user acceptance by deploying the prototype in end users’ homes over a multi-week period and conducting surveys about their subjective experience of the proactive AI assistants.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.