智源研究院学术副院长，清华大学教授唐杰发布悟道2.0成果丨智源大会

{"type":"doc","content":[{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2021 年 6 月 1 日，由"},{"type":"link","attrs":{"href":"https:\/\/www.baai.ac.cn\/","title":"","type":null},"content":[{"type":"text","text":"北京智源人工智能研究院"}]},{"type":"text","text":"主办的"},{"type":"link","attrs":{"href":"https:\/\/2021.baai.ac.cn\/","title":"xxx","type":null},"content":[{"type":"text","text":"2021 北京智源大会"}]},{"type":"text","text":"在中关村国家自主创新示范区会议中心开幕，会期三天。智源大会是北京创建全球人工智能学术和创新最优生态的标志性学术活动，本文为 InfoQ 记者带来的现场报道。"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"以下内容根据唐杰的演讲速记进行整理，未经本人确认。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“悟道”大规模智能模型取得多项国际领先的AI技术突破和多个世界第一，同时也做出了多项行业贡献。建设并开放了全球最大的中文语料数据库WuDaoCorpora2.0，提出了自然语言评测新标准“智源指数”，包含6种主要语言、30余项主流任务与相关数据集，形成了大模型“权威考卷”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/5a\/09\/5a39474d96b6d3ec00eeeeb44ed30409.jpeg","alt":null,"title":"","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"生命模型，有赖于脑科学特别是神经科学的进步，首先要对大脑进行高精度解析，再在大算力的支持下进行高精度仿真模型。针对生命模型，智源研究院成立了生命模拟研究中心，负责开发高精度生命模拟仿真平台“天演”，构建和探索感觉、知觉、记忆等智能模型，争取成为新一代人工智能发展的不竭源泉。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"生命模型本身就是脑科学研究的中心任务，是探索人类自身、解密智能成因的科学探索需要。“天演”生命模型系统已经开展斑马鱼全脑、灵长类视觉、人类心脏、人类认知等模型构建，推动生命科学、医疗健康和人工智能等前沿科研探索和应用发展。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"最后一条路线的挑战最大，核心是构建地球乃至宇宙的物理模型，培育新一代自主人工智能，应对各类自然挑战。智源已经启动了“天道”物理模型建设步伐，从基本粒子模拟开始，已建立了“分子模拟研究中心”，突破了分子模拟“快”与“准”难以兼顾的瓶颈，取得重要突破，将在未来五年内打造全球最具影响力的材料数据库，推动人工智能驱动的新一代分子动力学模拟迈上新台阶。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"上述三条路径的核心任务分别是构建信息模型、物理模型和生命模型，共同需求是算力。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为建设超大规模智算平台，北京市专门安排了专项资金，建设体系架构先进、高速互联互通、可扩展高效并行的软硬件系统。在强大的智算体系支持下，三大模型性能快速提升，共同支撑新一代人工智能模型的建立，进而赋智各行各业，赋智万物。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"智源研究院经过两年多探索，已经确定将打造智能模型作为源头创新的核心任务，在机构设置、资源保障和机制体制上进行了系统部署，全力构建信息模型、生命模型和物理模型“智能三剑客”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"简单总结来说，首先是“悟道”信息模型：数据是智能的素材，通过“大数据+大算力+强算法”，训练超大规模信息模型，为各行各业智能化提供切实可用的共性基础智能模型，是当前及未来一段时间人工智能全面赋智经济社会发展的中心任务。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"然后是“天演”生命模型：生命是智能的第一载体，模拟仿真亿万年进化而成的生物神经系统和身体，构建生命智能模型，挖掘生物智能机制机理，是发展新一代人工智能的不竭源泉。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"最后是“天道”物理模型：自然是生命的摇篮，地球是人类的摇篮，构建地球环境物理模型，是实现类人智能和人机共处共融的必要条件。建立宇宙物理模型，发展强人工智能，推动智能走出地球、穿越时空。本次智源大会，各位将亲眼目睹这三大模型的最新进展。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第二方面的工作是建设学术生态。智源从成立之初，就开始遴选智源学者，建设智源社区，积极打造全球人工智能学术和技术创新生态。“智源学者计划”主要支持顶尖科学家开展人工智能重大方向的基础前沿研究，支持青年科学家开展开放性、探索性研究。包含智源首席科学家、智源研究员、智源青年科学家等六个层次。已经围绕六大研究方向，遴选上百位智源学者，形成了AI科研的“北京力量”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"下面介绍一下六大重点研究方向的目标及工作进展。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“人工智能的数理基础”方向旨在建立以数学和统计理论为第一原理的新一代人工智能方法论。已取得了一系列成果。深度势能团队首次采用智能超算与物理模型结合，将原来可能需要60年才能完成的分子动力学模拟任务缩短到了1天，成果获得国际高性能计算应用领域最高奖——戈登贝尔奖。应用层面，团队提出的MOML算法已应用在冬奥赛场温度和风速预测，预测准确率大幅提高。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“人工智能认知神经基础”方向旨在将认知科学、神经科学和计算科学交叉融合，为人工智能的架构和功能探索新的模型和发展思路。目前正在建立全球规模最大的高精度生物智能开源开放平台及模拟系统。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“机器学习”方向旨在解决现有机器学习算法存在的可解释性缺失、大数据依赖、模型场景化等问题，开展“用户友好的AI”研究。已经取得了阶段性进展。黄高团队提出的高效空间自适应推理，可实现高效的动态特征提取。朱军团队提出第三代安全可控人工智能，研发出全球首个针对算法模型本身进行安全检测的技术平台RealSafe，提供从安全测评到防御加固的整体解决方案。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“智能体系架构与芯片”方向旨在从整机、编程、芯片、方法四个方面探索智能计算系统的发展。提出了“类脑计算完备性”及计算系统层次结构，相关工作发表于Nature。探索出分形冯·诺依曼体系、支持CNN+RNN的混合神经网络处理器的新型架构。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“智能信息检索与挖掘”方向旨在建立智能信息检索与挖掘领域的“北京学派”，已经研发并上线了“交互式智能政务助手”，支持了60多个北京政务应用。针对信息检索的可解释性和公平性，提出双稳健排序等模型，连续两年在ACM SIGIR会议上贡献了约10%的学术论文。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“自然语言处理”方向旨在提出“大数据+富知识+跨模态”共同驱动的新范式。在中文预训练模型、知识图谱、文本生成等领域，取得了一系列具有国际影响力的学术成果。研发出最大的中文预训练语言模型“悟道·文源”，研发的神经机器翻译工具包，成为国际三大神经机器开源软件之一。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"智源研究院和国际AI技术生态深入融合，建设智源社区，今年将汇聚10万以上海内外人工智能科研人员。日常，智源社区举办丰富的学术交流活动，包括学术报告会、主题沙龙等，以及一年一界的智源大会，打造聚合全球一流AI学者前沿思想、连接世界AI学术与产业资源的平台，构建全球人工智能学术和技术创新生态。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“青源会”是智源研究院发起的青年科学家学术组织，为海内外的AI青年科学家和技术人员建立的宽松、活跃的学术交流平台，促进学科交叉，支持青年科学家提出具有引领作用的原创思想，开创新的科学前沿。在6月2日上午举办的“青源学术年会”上，将举办“青源会”成立仪式，详细介绍“青源会”发展目标及未来规划。届时欢迎各位参加。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"当前，人工智能已成为经济发展的新引擎。智源研究院加快推进AI原创成果落地，加快人工智能深度应用，孵化AI创新企业，推动人工智能产业发展。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"健康医疗也是人工智能最热门的研究和应用领域之一，北京在发展智能医疗方面，有着得天独厚的优势。为此，我们成立了“健康医疗研究中心”，构建医疗大数据和高质量小数据等基础数据，加快智能医疗核心技术突破，加快人工智能在健康医疗领域的深度应用。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此外，我们与清华智能产业研究院开展深度合作，推动以“个人”为核心的、AI驱动的主动健康管理。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在推动AI原创成果转化方面，去年建立了多个创新中心，加速AI原创成果熟化及工程化。智源为创新中心提供场地与研发资金支持。目前，已建设智能信息处理、认知知识图谱、安全人工智能等8个创新中心。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一年来，智源创新中心建设取得重大进展，例如："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"视网膜芯片技术创新中心，已研制出超高速脉冲视觉芯片，用于高速场景、弱小目标、全时检测等机器视觉应用。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"安全人工智能创新中心，研发出RealSafe人工智能安全平台、DeepReal深度伪造内容检测平台等，并在金融、能源等行业进行应用。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"智能政务信息处理创新中心研发的政务问答系统，已在60多个北京政务场景中得到应用。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"疾病脑电智能技术创新中心，研发的临床脑电图智能评估系统，准确率已达到落地应用水平，产品覆盖全国11个省市，累计服务超过上万名患者。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"今年，我们按照国家和北京市促进科技成果转化的最新部署，积极整合技术、场景、人才、资金等资源，推出支持AI创业的“源创计划”，为AI创业团队开放大模型、数据集等生态资源，为来自学术界的AI科学家创业团队对接应用场景，为来自产业界的创业团队对接业界领先的AI技术，加快形成可落地应用的AI产品。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另外，还将通过创业基金等方式，加快孵化一批具有国际领先技术能力的AI创新企业，为北京AI产业的持续发展储备力量。“源创计划”将在6月3日上午召开的“AI创业论坛”正式发布，敬请期待。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"当前，AI伦理和治理成全球共识，国际社会正探索建立广泛认可的AI伦理原则，推进敏捷灵活的AI治理。智源研究院在2019年成立人工智能伦理与可持续发展研究中心，率先在国内开展AI伦理安全研究，并积极参与国际AI治理。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"AI伦理规范方面，2019年5月，发布了我国首个AI发展与治理准则——《人工智能北京共识》。去年9月，又发布了我国首个《面向儿童的人工智能北京共识》。另外，我们建设了我国首个“人工智能治理公共服务平台”，针对人工智能技术研发及应用中存在的潜在伦理问题提供检测服务，帮助AI研发机构研发更符合AI伦理规范的技术和产品。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"联合国在2015年正式通过了可持续发展目标，共17大项，涉及社会、经济和环境三方面的发展问题。为此我们发起成立了“面向可持续发展的人工智能国际协作网络与智库”，目前已汇聚了来自15个国家的专家学者。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"同时，我们也联合百度、小米、旷视、滴滴等企业发起了“面向可持续发展的公益计划”，目前已围绕城市治理、自动驾驶、劳动者就业影响、青少年心理影响等方面发布了首批4个研究题目，并已启动研究工作。后续，完成后的研究成果将会面向全球公开，服务于人类命运共同体的构建与实现。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"以上是智源研究院一年来的工作进展，我的介绍就到这里，谢谢大家。"}]}]}