Collaborative Research: NSF2026: EAGER: A Playground and Proposal for Growing an AGI.

在智能和信息系统部（IIS）的强大智能项目和综合活动办公室的NSF 2026基金项目的支持下，波士顿学院和布兰迪斯大学的研究人员正在通过综合符号或逻辑推理、通过与环境的交互学习以及最先进的神经网络来解决创造人工通用智能的挑战。受自然（例如，人类）智能结构的启发，由此产生的心智架构为他们擅长的问题部署了每一种策略（“Best of All Worlds”或BAW方法）。该项目的成功完成将促进人工智能和心理学/神经科学领域的一系列研究项目。从长远来看，AGI的发展有望为社会带来重大利益，使计算机能够根据与世界的经验发展抽象概念，并产生新颖的想法和发明。该项目还将有助于扩大学生培训和妇女及代表性不足的少数民族的参与。该项目旨在建立一个新架构的原型，并针对一个开放的任务进行测试，这个任务对于人工智能来说很难，但对于任何地方的幼儿来说都能掌握：揭示积木、容器和其他小物体的可见性。主要目的是建立一个虚拟世界，模拟婴儿可以探索、操纵和学习；建立一个包含BAW心理架构关键方面的模拟婴儿的工作原型；并评估代理在几个困难的开放式任务上的表现。这种架构有助于整合自然智能研究中的关键概念，这些概念在人工智能中很少使用：心智模型、探索性游戏和分块。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。

With support from the Robust Intelligence program in the Division of Intelligent and Information Systems (IIS) and the NSF 2026 Fund Program in the Office of Integrated Activities, investigators at Boston College and Brandeis University are addressing the challenge of creating Artificial General Intelligence by synthesizing symbolic or logical reasoning, learning through interaction with the environment, as well as state-of-the-art neural networks. Inspired by the structure of natural (e.g., human) intelligence, the resulting mental architecture deploys each of these strategies for the problems they excel at (the "Best of All Worlds”, or BAW, approach). Successful completion of this project will facilitate a range of research projects in AI and psychology/neuroscience. Long-term, the development of AGI is expected to have significant benefits to society, by enabling computers to develop abstract concepts grounded in experience with the world, and to generate novel ideas and inventions. This project will also help broaden student training and participation of women and underrepresented minorities.This project aims to prototype a new architecture and test it against an open-ended task that is difficult for artificial intelligence but mastered by human toddlers everywhere: uncovering the affordances of blocks, containers, and other small objects. The primary aims are to build a virtual world that a simulated infant can explore, manipulate, and learn from; build a working prototype of a simulated infant incorporating key aspects of the BAW mental architecture; and evaluate the performance of the agent on several difficult, open-ended tasks. This architecture facilitates incorporation of key concepts from the study of natural intelligence that are infrequently used in artificial intelligence: mental models, exploratory play, and chunking.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.