权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

AI Institute: Planning: The Proteus Institute: Intelligence Through Change

AI 研究所：规划：Proteus 研究所：通过变革实现智能

基本信息

批准号：
2020247
负责人：
Joshua Bongard
金额：
$ 50万
依托单位：
University of Vermont & State Agricultural College
依托单位国家：
美国
项目类别：
Standard Grant
财政年份：
2020
资助国家：
美国
起止时间：
2020-10-01 至 2022-12-31
项目状态：
已结题

来源：
https://www.nsf.gov/awardsearch/showAward?AWD_ID=2020247&HistoricalAwards=false
关键词：
AI Institute Planning Proteus Intelligence

项目摘要

For nearly 50 years, AI researchers have improved their ability to simulate one aspect of how the brain works: experience-driven change in the connections between neurons to improve the organism’s -- or the computer program’s -- behavior. This mechanism now lies at the heart of modern AI technologies. However, 'synaptic plasticity' is but one of millions of adaptive mechanisms that evolved to help organisms grow and learn to grapple with their environments, and many of those mechanisms are not limited to brains. Intelligent problem-solving was already discovered by evolution around the time bacteria evolved to clump together, and is used widely by cells and tissues throughout the body. Imagine an Institute that helps biologists, neuroscientists, ecologists and psychologists work with computer scientists to turn hundreds of these adaptive mechanisms into code. Imagine further an AI that figures out how to combine these code pieces together to create AI programs, robots, and computer-designed organisms that inherit life’s breathtaking abilities to perform complex tasks, recover from unexpected situations, and work well with others. Organisms can do these things because they constantly adapt their brains, bodies, and coalesce into ever larger functional groups, like communities and societies. For this reason this project will use the next two years to lay the groundwork for the Proteus Institute, named after the Greek god of constant change. Along the way, the project will provide opportunities for students, policy makers, companies and the general public to influence how such technology is created.Living systems continue to outstrip the most adaptive state-of-the-art artificial intelligence (AI) and robotics. One reason for this is that, without exception, organisms and species constantly restructure themselves at all organizational levels, from the microsecond- to millennial time scales; most machines do not. Almost all AI and robots incorporate change at just one time scale – that of synaptic plasticity – and in one modality: neural networks. This project will thus plan the Proteus Institute, dedicated to studying embodied plasticity: how multi-level change supports intelligence in protean systems (cells, organs, organisms, and ecologies), and how best to channel those discoveries into protean machines (robots and computer designed organisms) and algorithms (machine learning methods). To achieve this, the project will construct a continuously-running evolutionary algorithm that designs and trains robots and ML algorithms, using the insights of basal cognition and the multi-scale control systems of developmental biology. This algorithm will be increasingly enriched by software patches that simulate new and potentially useful adaptive mechanisms. This will enable the algorithm to discover combinations of domain agnostic adaptive mechanisms in a growing set of embodied AI and ML substrates. The project will also host two workshops where it will elicit candidate biological mechanisms not yet incorporated into robots or ML methods from biologists, neuroscientists, ecologists and psychologists. During the second workshop, the project will convene computational researchers to begin incorporating those mechanisms, and invite policy, industry, and other stakeholders to discuss how the emerging biology-to-AI pipeline could facilitate technology transfer, education, workforce development, and policy.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.

近50年来，人工智能研究人员一直在提高他们模拟大脑工作方式的一个方面的能力：经验驱动的神经元之间连接的变化，以改善有机体或计算机程序的行为。这一机制现在处于现代人工智能技术的核心。然而，“突触可塑性”只是进化来帮助有机体成长并学会与环境搏斗的数百万种适应机制之一，其中许多机制并不局限于大脑。大约在细菌进化成簇的时候，进化就已经发现了智能解决问题的方法，并被全身的细胞和组织广泛使用。想象一下，一个帮助生物学家、神经学家、生态学家和心理学家与计算机科学家合作，将数百种这种自适应机制转化为代码的研究所。进一步想象一个人工智能，它知道如何将这些代码片段组合在一起，创建人工智能程序、机器人和计算机设计的有机体，这些程序、机器人和计算机设计的有机体继承了生命执行复杂任务、从意外情况中恢复以及与他人良好合作的惊人能力。生物体之所以能做这些事情，是因为它们不断地调整自己的大脑、身体，并结合成越来越大的功能群体，如社区和社会。出于这个原因，这个项目将利用未来两年的时间为Proteus研究所奠定基础，该研究所以希腊恒变之神的名字命名。在此过程中，该项目将为学生、政策制定者、公司和普通公众提供机会，影响此类技术的创造方式。生命系统继续超过最具适应性的人工智能(AI)和机器人技术。其中一个原因是，无一例外，生物体和物种在所有组织级别上不断地自我重组，从微秒到千年的时间尺度；大多数机器不会。几乎所有的人工智能和机器人都只在一个时间尺度上--突触可塑性--和以一种方式--神经网络--融合变化。因此，该项目将规划Proteus研究所，致力于研究体现可塑性：多层次变化如何支持多变系统(细胞、器官、生物体和生态系统)中的智能，以及如何最好地将这些发现引导到多变机器(机器人和计算机设计的生物体)和算法(机器学习方法)中。为了实现这一目标，该项目将利用基础认知和发育生物学的多尺度控制系统的见解，构建一个连续运行的进化算法，设计和训练机器人和ML算法。该算法将通过模拟新的和潜在有用的自适应机制的软件补丁而日益丰富。这将使算法能够在越来越多的人工智能和ML底物中发现领域不可知的自适应机制的组合。该项目还将主办两个研讨会，从生物学家、神经学家、生态学家和心理学家那里引出尚未纳入机器人或ML方法的候选生物机制。在第二次研讨会期间，该项目将召集计算研究人员开始纳入这些机制，并邀请政策、行业和其他利益相关者讨论新兴的生物到人工智能的管道如何促进技术转让、教育、劳动力发展和政策。这一奖项反映了NSF的法定使命，并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。