Searle Tells Google Digital AI Has Zero Intelligence, Cannot Think

塞尔向谷歌断言：数字人工智能毫无智能，无法思考 由于计算机程序仅处理纯粹的形式语法而非内在语义，计算模拟无法复制产生意识思维的生物因果力。 约翰·塞尔（John Searle）认为，绝不能将功能性工程的成功与真正的心理理解相混淆，坚称单纯的形式语法无法产生语义理解。通过其著名的“中文房间”思想实验，他证明了计算机程序只是在操纵形式符号，而不具备任何内在意识或清醒体验。他区分了独立于观察者的生物现象与相对于观察者的计算解释，并得出结论：计算模拟可以对过程进行建模，但无法复制大脑的物理因果机制。尽管生物体本身就是已经能够思考的物理机器，但要创造真正的机器意识，需要复制神经生物学的因果力，而非仅仅扩展数字代码的规模。 约翰·塞尔认为，“强人工智能”混淆了语法层面的模拟与语义层面的复制，这意味着无论计算机程序多么复杂，单靠语法永远无法产生内在意识。 (11:57) 必须在内在且独立于观察者的现象（如生物大脑的思考）与相对于观察者的归因（如人类将电子状态转换解释为计算）之间划清关键界限。 (18:51) 对于机械智能反抗人类的担忧从根本上讲是完全错位的，因为数字计算机根本不具备任何内在心理状态或意识意图。 (27:14) 真正的人工意识需要复制大脑的生物因果力，而不是编写形式系统，正如模拟人体消化或暴风雨既不会产生实际的消化过程，也不会带来真实的湿润一样。 (32:01)

Philosopher John Searle argues that computational symbol manipulation can never produce genuine understanding because syntax is not semantics, meaning digital computers lack intrinsic consciousness. He asserts that while machines can simulate human-like behavior, consciousness is a specific biological causal process rather than an abstract formal program. Epistemic Ambiguities and Observer-Relative Phenomena (2:23) Searle diagnoses a pervasive conceptual confusion between epistemic claims regarding knowledge and ontological claims regarding existence. While scientific investigation must remain epistemically objective by evaluating claims independently of personal bias, this does not preclude investigating phenomena like consciousness, which possess an ontologically subjective mode of existence because they exist only as felt experiences. Furthermore, observer-independent entities such as mountains and molecules exist regardless of human cognition, whereas observer-relative phenomena like money, universities, and commercial computation exist entirely through the collective intentionality and attitudes of conscious observers. The Chinese Room and the Syntax-Semantics Gap (11:22) Searle uses the Chinese room thought experiment to refute the core premise of strong artificial intelligence. A person inside a room who executes formal instructions to match and output Chinese characters can flawlessly pass the Turing test to outside observers without understanding a single word of Chinese. Digital computer programs operate through purely formal syntax, whereas human thought requires semantics and intrinsic mental content. Searle rejects the "Systems Reply" and related counterarguments, pointing out that expanding the boundaries of the system or adding syntactic processing modules still fails to bridge the gap between mechanical symbol manipulation and conscious meaning. Intrinsic versus Observer-Relative Intelligence (18:51) Attributions of intelligence to computational machinery are strictly observer-relative, existing solely in the minds of the human programmers and users who interpret the machine's electronic state transitions. A pocket calculator performing an addition function or Deep Blue evaluating 250 million chess positions per second possesses zero intrinsic intelligence or conscious awareness; both merely simulate cognitive tasks. Consequently, Searle dismisses apocalyptic warnings of artificial intelligence rebelling against humanity, comparing fears of machine uprisings to worrying that closet shoes will revolt because people have walked on them for centuries. Biological Causality versus Computational Simulation (32:01) Duplicating a biological power is fundamentally different from running an abstract simulation. Searle compares building an artificial brain to creating an artificial heart: an artificial heart succeeds by replicating the physical causal power of pumping blood, whereas software programs merely execute formal models without replicating the brain's biochemical causal mechanisms. In debate with Ray Kurzweil, Searle rejects the claim that consciousness can emerge simply by executing complex algorithmic operations, emphasizing that altered states caused by substances like cocaine demonstrate that specific neurobiology and neurotransmitters, not abstract substrate-independent code, produce subjective experience. The Scientific Search for the Unified Field (67:39) Searle defines consciousness in ordinary terms as the qualitative sequence of subjective feeling and awareness that begins upon waking and ceases with sleep or anesthesia. Rather than pursuing the dominant "building block approach" that seeks isolated neural correlates for distinct sensations like perceiving the color red or hearing middle C, Searle advocates for a "unified field" model. Under this view, sensory perceptions do not generate conscious experience from scratch but instead modify an already existing conscious baseline, leaving neuroscience with the empirical task of discovering the physiological mechanisms that establish that unified field. Searle emphasizes that resolving the problem of consciousness will not come from philosophical deduction or formal computer programming, but from empirical neurobiological research determining how physical brains causally generate conscious experience.

加州大学伯克利分校哲学教授 John Searle 在 Google 演讲中主张，计算机程序仅依赖纯粹的形式句法操作，绝不可能产生真正的语义理解或意识；意识是一种生物大脑因果机制产生的本体论主观现象，而非可仅凭算法模拟实现的计算过程。 认识论与本体论的混淆 (2:23) Searle 指出，围绕人工智能与心智哲学的讨论常陷入概念混乱，关键在于混淆了“客观/主观”的认识论含义与本体论含义。认识论上的客观指不依赖个人偏好、可通过经验检验的事实判断（例如伦勃朗生于1606年），主观则指审美或个人评价；而在本体论上，山脉与构造板块具有不依赖主观体验的客观存在，疼痛或意识体验则属于本体论主观，必须由意识主体感受才存在。神经科学家常因意识具有主观性而断言其无法被客观研究，Searle 认为这是一种歧义谬误：科学研究完全可以以认识论客观的态度，去探究本体论主观的意识现象。 观察者依赖性与计算的本质 (5:23) Searle 进一步区分了“观察者独立”与“观察者相对”两类现象。山脉与分子在物理世界中独立存在，而货币、财产、婚姻与 Google 等社会制度只因人类意识的集体赋予才成立，一旦主观态度瓦解，这些建构便不复存在。现代计算与机器“智能”同样完全属于观察者相对的属性。计算器或商业计算机在物理层面上仅发生电路状态转换，其“进行算术”或“具备智能”的意义完全来自设计者与使用者的外部解释。人类用重力公式 $s = \frac{1}{2}gt^2$ 描述落体，并不意味着落体本身在进行内在计算；图灵机并非大自然既有的物理实体，而是人类对物理过程的解释模型。 中文屋论证与句法语义的鸿沟 (11:22) 针对强人工智能声称运行适当程序的计算机能真正理解语言的观点，Searle 提出了著名的“中文屋”思想实验。一个完全不懂中文的人被关在房间内，仅依照一本英文规则手册机械地匹配和置换传入的中文符号，输出在外人看来完全流利准确的中文回答。即便该系统通过了图灵测试，屋内操作者依然对中文毫无理解。Searle 据此总结出核心命题：“句法不同于语义”以及“模拟不同于复制”。计算机程序仅通过形式规则处理未经解释的符号标识，而人类心智则具有内在的意向性与语义内容，单纯扩展系统边界（如“系统回复”）或在代码中添加标注为“语义”的模块，依然只是增加形式规则，无法从形式句法飞跃到真正的语义理解。 生物因果力与模拟复制的边界 (18:51) 面对机器觉醒或反叛的人工智能末日论，Searle 认为这建立在混淆内在智能与观察者相对智能的虚幻推论上。计算机完全缺乏意识状态，担忧机器叛变如同担忧鞋柜里的鞋子会因被践踏而报复人类。他反驳 Ray Kurzweil 将神经递质置换为符号的类比，强调人脑是通过特定的电化学过程（如突触间隙的多巴胺、去甲肾上腺素和GABA机制）产生高阶意识状态的生物因果机器。正如人工心脏必须真正复制泵血的物理因果力而非仅仅模拟血流，若要制造真正思考的人工机器，必须复制大脑产生意识的具体因果力，而非依靠计算机算法进行形式模拟。计算机可以完美模拟披萨的消化过程或暴风雨，但模拟消化无法真正消耗食物，模拟暴风雨也不会把人淋湿。 神经科学转向与统一场模型 (51:49) 在探讨如何界定与研究意识时，Searle 给出了常识性操作定义：意识是指从无梦睡眠中醒来开始、直到再次入睡或麻醉为止的持续感知与清醒状态，其核心特征在于具备主观的质感体验。他赞同神经科学界打破旧有学术禁忌、直接研究意识神经相关物的经验转向，但对当前主流的“积木块途径”（试图先攻破看到红色或听到中央C的单项神经机制以推导整体机制）表示怀疑。Searle 倾向于“统一场”模型，主张知觉并非凭空创造意识，而是在大脑预先建立的统一主观意识背景场域中进行局部的质感修饰；因此，破解意识谜题的根本在于揭示大脑大尺度系统如何建立这一基线场域。 Searle 强调，人类本身就是具有思考能力的生物物理机器，原则上并不排除未来利用非生物材料复制出具备意识的人造物，但通过编写符号程序实现强人工智能的技术路径从哲学原理上是不可行的。破解心智奥秘的最终解答将不会来自哲学的形式论辩，而取决于神经科学界能否在实证研究中确切查明大脑产生意识的深层生物机制。

## Background 该视频由 YouTube 频道 Talks at Google 于 2015 年 12 月 4 日发布。讲座由 Google 内部的“奇点网络”（Singularity Network）主办，主讲人是加州大学伯克利分校的斯卢瑟哲学讲席教授约翰·塞尔（John Searle），主要探讨心灵哲学以及人工智能是否可能拥有意识的核心命题。 ## TL;DR 塞尔通过其著名的“中文屋”（Chinese Room）思维实验指出，单纯基于符号和句法规则操作的计算机程序永远无法跨越到语义理解，因此“模拟不等于复制”，计算本身绝非产生心智的充分条件。他严格区分了“认识论与本体论中的主客观”以及“观察者独立与观察者相对”的概念，论证了现代计算机的一切智能与计算本质上都是观察者主观赋予的，内在智能为零。面对未来学家 Ray Kurzweil 及现场观众关于超级智能叛乱、图灵测试与系统涌现论的质疑，塞尔坚决主张意识是大脑特有的生物学因果机制所产生的高阶现象。他强调，要创造有意识的机器，关键在于复制大脑底层的生物物理因果效力而非仅仅运行代码，并呼吁通过硬核神经生物学研究来破解意识的统一场之谜。 ## Atmosphere & style 整场讲座兼具深厚的学术思辨与犀利幽默的论战色彩，现场气氛活跃而充满智力张力。塞尔教授以其标志性的直率、不拘小节与略带自嘲的口吻侃侃而谈，动辄借扔下随身物品或生动比方瓦解科技狂热主义的幻觉。在与多年论敌 Ray Kurzweil 以及台下 Google 工程师的多轮交锋中，塞尔反应机敏、言辞锋利，展现出一股历经数十年辩论洗礼却依旧兴致勃勃的哲学大家风范，让观众深切感受到严谨分析哲学与前沿工程思维激烈碰撞的快感。 ## Bulletin - [01:38](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=98s) 约翰·希尔勒强调意识在本体论上是主观的，但这并不妨碍建立一门在认识论上完全客观的意识科学。 - [11:57](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=717s) “中文屋”思想实验表明符号计算仅停留在句法（Syntax）层面，而无法跨越到语义（Semantics），因而纯粹形式化的计算程序永远无法构成真正的人类理解。 - [20:36](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=1236s) 计算与机器智能属于“观察者相对”（Observer-relative）现象，脱离了有意识主体的解释，任何物理设备或落体运动都不具备内在的计算与思考能力。 - [29:47](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=1787s) 关于人造意识的实现，希尔勒指出必须在因果机制上“复制”（Duplicate）大脑产生意识的具体生物化学原理，而非仅仅在形式上进行计算“模拟”（Simulate）。 ## Key takeaways - [01:38](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=98s) 区分认识论客观与本体论主观：意识虽然具有主观的存在模式（依赖于主体体验），但依然可以成为客观科学严谨研究的对象。 - [05:22](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=322s) 观察者独立（如山川、分子）与观察者相对（如金钱、政府、AI 智能）有着根本不同，混淆两者的本体论地位会导致严重的认知谬误。 - [08:50](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=530s) “中文屋”思想实验证明，仅凭查阅规则手册匹配未知符号并通过图灵测试，操作者本身对符号背后的语义依然一无所知。 - [11:57](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=717s) 计算程序定义了一个完全独立于具体物理实现的句法等价类，但“语法不等于语义，模拟不等于复制”（Syntax is not semantics. And simulation is not duplication.）。 - [15:03](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=903s) 反驳“系统反驳”（The Systems Reply）：如果身处其中的人类无法理解语义，那么由纸箱、规则手册和房间构成的无意识系统更不可能涌现出理解。 - [18:51](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=1131s) 所谓“超级 AI 因厌倦被奴役而反叛”是科幻式的伪命题，是将人类特有的主观心理状态错误投射到无内在心智的机器上的无稽之谈。 - [20:36](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=1236s) 深蓝（Deep Blue）等国际象棋程序仅是在观察者相对意义下执行海量形式计算，在内在心理学意义上其智能完全为零。 - [29:47](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=1787s) 人类与动物本身就是生物机器，因此“机器能否思考”在生物学意义上答案是肯定的，思考是由大脑特定神经生物过程产生的物理现象。 - [35:24](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=2124s) 若将计算泛化为观察者相对的数学映射，任何遵循物理定律的运动（如物体自由落体运算 $s = \frac{1}{2}gt^2$）都可以被称为计算，这揭示了机器计算在本体论上的空洞性。 - [41:45](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=2505s) 回应 Ray Kurzweil 的反诘：大脑通过具体的神经递质和受体等电化学因果机制产生意识，而计算机算法与物理实现完全解耦，不能直接套用类比。 - [55:22](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=3322s) “计算消化披萨”思想实验表明，无论计算机对消化过程的数学模拟多么完美，也绝不可能消化掉真正的披萨；认知模拟同样不能等同于心智复现。 - [57:50](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=3470s) 图灵测试无法作为检验内在智能的有效标准，因为它的输入与输出完全依赖外部人类观察者的解释，容易产生严重的假阳性。 - [01:07:39](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=4059s) 在神经科学研究中，希尔勒更倾向于将意识视为一个预先存在的“统一场”（Unified field），感知输入只是在调制这个统一场，而非从零构建离散的意识积木。 - [01:09:34](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=4174s) 意识难题的终极裁决权不在哲学思辨，而在于未来神经生物学硬核实证研究能否精确定位清醒大脑与无意识大脑之间的物理因果机制差异。 ## Detailed walkthrough ### [00:01](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=1s) 讲座开场：Google 奇点网络与约翰·希尔勒的登场 在 YouTube 运营部门工作的 Googler 约翰·布拉卡利亚（John Bracaglia）担任主持人，代表其领导的内部组织“奇点网络”（Singularity Network）向听众问好。该组织专注于探讨人工智能领域的理性与思辨。布拉卡利亚隆重介绍了主讲嘉宾约翰·希尔勒（John Searle）——加州大学伯克利分校的斯卢瑟哲学讲席教授（Slusser Professor of Philosophy）。布拉卡利亚向听众梳理了希尔勒的学术成就，指出其在语言哲学、心灵哲学以及社会哲学领域做出了广泛而杰出的贡献，曾荣获让·尼科奖（Jean Nicod Prize）、国家人文奖章（National Humanities Medal）与心灵与大脑奖（Mind and Brain Prize）。在其众多知名概念中，尤以反驳强人工智能（Strong AI）的“中文屋”（Chinese Room）论证闻名遐迩。 在全场的掌声中，希尔勒走上讲台。他风趣地称赞道，回到 Google 的感觉非常棒，这里就像是一所“校外的大学”，甚至“有时候我觉得大学真正应该建成的样子就是这样”。希尔勒随即切入正题，表示现代人尤其是美国人，极度热衷于为技术进步喝彩。例如一旦有了自动驾驶汽车，大家根本不在乎这辆车到底有没有意识。但希尔勒强调，从理解技术的本质目的来看，有许多深层问题至关重要，而这正是他接下来要拆解的核心。 - 主持人：John Bracaglia（YouTube Operations，Singularity Network 负责人） - 主讲人：John Searle（UC Berkeley 哲学教授） - 获得的荣誉：Jean Nicod Prize、National Humanities Medal、Mind and Brain Prize ### [01:38](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=98s) 概念辨析：认识论与本体论的“客观与主观” 希尔勒指出，在进入当代思想争鸣之前，必须先厘清两个容易被混淆的概念——“客观性”（objectivity）与“主观性”（subjectivity）。在日常文化与科学探索中，人们极力推崇客观科学，但这两个词存在系统性的歧义，若不加分辨便会导致严重的思想灾难。它们在两个不同维度上游移：其一是认识论（epistemic，涉及知识的层面），其二是本体论（ontological，涉及存在的层面）。 ```mermaid flowchart TD accTitle: 认识论与本体论中的主客观双重维度 accDescr: 展示主客观概念在认识论知识层面与本体论存在层面的四种组合与区分 subgraph EPI["认识论维度 (Epistemic / 知识层面)"] EO["认识论客观 (如：伦勃朗生于1606年)"] ES["认识论主观 (如：伦勃朗是最伟大的画家)"] end subgraph ONT["本体论维度 (Ontological / 存在方式)"] OO["本体论客观 (山川、分子、板块：独立于感知)"] OS["本体论主观 (疼痛、痒：必须依赖主体体验)"] end EO -.->|"能够研究"| OS ``` 在认识论层面，两者的区别关乎知识断言（claims）的性质。如果声称“伦勃朗生于 1606 年”，这是一个客观事实，不以断言者或解读者的态度为转移，属于“认识论客观”；但若声称“伦勃朗是有史以来最伟大的画家”，则纯属个人意见，属于“认识论主观”。而在更底层的本体论层面，区分的标准则是存在的模式：山脉、分子和地质构造板块的存在不以任何人的意志为转移，具备“本体论客观”的存在模式；而疼痛、痒等感觉，只有在被主体体验到时才存在，属于“本体论主观”。 希尔勒分享了一段亲身经历：当年他初涉该领域时，纳闷脑科学家为何迟迟不解决意识问题，于是亲自前往加州大学旧金山分校（UCSF）质问神经生物学家团队：“你们这帮人为什么特么的不能搞清楚大脑究竟是如何产生意识的？我给你们付工资是让你们干什么的？”对方的回答却是：我们做的是客观科学，而你自己都承认意识是主观的，因此不可能存在关于意识的科学。希尔勒指出这完全是词义歧义导致的谬误。关于意识科学的可能性，希尔勒斩钉截铁地指出： > “The point is this: you can have an epistemically objective science of consciousness, even though consciousness is ontologically subjective.”（“核心在于：即使意识在本体论上是主观的，你依然可以拥有一门在认识论上完全客观的意识科学。”） ### [05:22](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=322s) 观察者独立性与制度现实的脆弱性 擦掉白板后，希尔勒引入了另一项关键区分：“独立于观察者的现象”（observer-independent）与“相对于观察者的现象”（observer-relative）。山川、分子和构造板块无论任何人怎么想都客观存在，属于前者；而人类生活中的许多核心事物，则只相对于观察者和使用者的态度才成立。例如钱包里的纸币之所以是钱，并非由其化学成分决定，而是完全取决于人类对待它的心理态度。 希尔勒列举道：金钱、财产、政府、婚姻、大学、Google、鸡尾酒会以及暑假，全都属于“相对于观察者”的范畴。这些现象均由人类意识创造，因此内含本体论主观性的成分。然而，人类依然可以对这些相对于观察者的领域建立认识论上客观的科学，经济学便是典型例子。但经济学家往往忘记这一点，以为经济学像物理学一样坚硬。希尔勒回顾自己当年学经济学时大为震惊：教授用讲授物理学“力等于质量乘以加速度”（$F=ma$）时一模一样的口吻，去传授经济学里的“边际成本等于边际收益”（$MC=MR$）。这完全是两码事——经济学研究的对象完全依赖于观察者的主观态度，“一旦主观态度发生改变——唰的一下——整个大厦就崩塌了，这在 2008 年金融危机中已展露无遗”。认知科学中的智力、认知、记忆、思维和知觉等核心概念同样兼具这两种含义，稍不留神就会混为一谈。 ### [08:50](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=530s) 认知科学的诞生与“中文屋”思维实验的缘起 希尔勒带听众回顾了一段思想史：几十年前认知科学诞生，最初是因为一批学者对心理学中的行为主义（Behaviorism）感到厌烦而发起。当时斯隆基金会（Sloan Foundation）出资赞助他们到处讲学，参与者们自嘲为“斯隆游侠”（Sloan Rangers），在各地主要是互相讲给对方听。希尔勒受邀前往耶鲁大学人工智能实验室演讲，当时他对人工智能一无所知，便掏了 16.95 美元加税买了一本耶鲁学者写的书。买完他心里还嘀咕：“这钱白扔了。” 读完后他发现书里提出了一套计算机能够理解故事的理论。研究者给计算机输入一段故事，然后提问，计算机能答出故事中未明写的隐含信息。书中的经典例子是：一个人进餐馆点了汉堡，侍者端上来时汉堡全烤焦了；此人勃然大怒，账单都没付就夺门而出。研究者问计算机：“这个人吃了汉堡吗？”计算机会正确回答：“没有。”耶鲁学者以此声称：这证明计算机真正理解了这个故事。 希尔勒在乘坐美联航飞往纽黑文的 30,000 英尺高空中灵光一闪：如果这套故事完全用中文写成，自己关在房间里依照英文规则去匹配符号并给出正确回答，而自己连一个汉字都不懂，这是否证明自己懂中文？希尔勒起初以为耶鲁的学者们肯定早已想过这个反驳，但这篇演说至少能应付差事。没想到对方不仅完全没预料到这一击，还纷纷跳出来反驳，更奇特的是反驳的理由各不相同。原本以为只是一周的学术交锋，结果演变成了长达 35 年的论辩。 希尔勒无奈地感叹道：“The argument has gone on longer than a week. It's gone on for 35 years. I mean, how often do I have to refute these guys?”（“这场辩论没有只维持一周，它持续了整整 35 年。我的意思是，我到底要驳斥这帮家伙多少次才算完？”）接着他向全场简要复述了经典“中文屋”模型的最简版本： 1. 自己被反锁在屋子里，屋内放满装有中文符号的盒子（在外界看来这叫“数据库”），以及一本用英文书写的符号操作规则手册（外界称之为“程序”）； 2. 外界递入一串中文符号（外界称之为“问题”）； 3. 他根据英文指令查阅规则，倒腾匹配这些符号，最后把另一组符号递出房间（外界视之为对问题的“回答”）。 整个过程纯粹是基于语法结构的符号重组，屋内的操作者对符号背后的语义（Meaning）一无所知。希尔勒讲到这里自嘲地笑称，希望大家对此感到厌烦，因为他自己早就讲烦了。 ### [11:57](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=717s) 中文屋的核心逻辑：语法不等于语义，模拟不等于复制 在中文屋思想实验中，约翰·塞尔（John Searle）设想外部专家编写的程序和规则极其完善，而自己在房间内对符号的挪动和匹配也熟练至极，以至于递出的回答与母语为中文的人毫无二致，轻而易举便通过了图灵测试（Turing test）。然而，即便如此，他依然连一个中文词都听不懂。塞尔指出，身处中文屋中的他绝不可能由此理解中文，因为他所扮演的无非是一套计算机系统，执行的只是一套计算机程序。程序纯粹是句法层面（syntactical elements）的运算规则。若用更严谨的技术语言来表述，所谓“相同已实现的程序”构成了一个完全独立于具体物理机制、独立于其实体实现的等价类。 如果执行这套程序的人根本无法借此理解问答的真实含义，那么任何基于执行程序运转的数字计算机也同样无法理解，因为计算机并不具备任何人类所没有的运算秘诀。计算机是彻底的句法语义设备，其运算完全在句法上被定义；而人类智能绝不仅仅需要句法，更需要语义（semantics），需要切实领会正在发生的一切。塞尔用自己面对英语和中文时的反应做对比：当别人用英语问他美国最长的河流是什么时，他清楚问题的含义，并回答密西西比河（或密西西比-密苏里河）；但当别人用中文问中国最长的河流是什么时，他看到的只是一堆毫无意义的符号，仅仅依靠规则手册查找匹配项，递出代表“长江”的符号。 ```mermaid flowchart LR accTitle: 中文屋与人类理解的对比 accDescr: 对比人类基于语义的理解与计算机仅依靠符号句法的处理机制。 subgraph Computer["中文屋 / 计算机处理机制"] A["输入：未知中文符号"] --> B["句法查找手册<br/>（纯规则匹配）"] B --> C["输出：正确中文符号"] end subgraph Human["真实人类智能"] D["输入：语言符号"] --> E["意识与语义理解<br/>（知晓现实指涉）"] E --> F["输出：基于认知的回答"] end ``` 约翰·塞尔在此总结了中文屋论证底层的两大支柱公理，他强调这两点显而易见且各自只需四个英文单词即可概括：“The bottom line is that the implemented computer program by itself is never going to be sufficient for human understanding because human understanding has more than syntax.” 他进而斩钉截铁地宣告：“Syntax is not semantics. And simulation is not duplication.”（语法不等于语义，模拟不等于复制。） - 计算机程序的等价类完全独立于物理实现，其运算完全由符号操作所定义。 - 计算机无论运算多么精密，都仅停留在符号句法层面，缺失对世界指涉的内在语义。 - 模拟某种认知过程（Simulation）从逻辑上根本不等同于复制该认知过程（Duplication）。 --- ### [15:03](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=903s) 反对意见与驳斥：系统反驳与荒谬的追问 在提出中文屋论证后的漫长岁月里，塞尔收到了大量反驳意见，但他直言这些反驳在荒谬程度上简直“令人叹为观止”。其中最流行的一种被称为“系统反驳”（The Systems Reply）。持该观点的人认为，坐在房间里的塞尔确实不懂中文，他充其量只是一个 CPU（中央处理器）；但塞尔加上整个房间、纸箱、草稿纸和规则符号，整个系统作为一个整体是理解中文的。塞尔对此回应道，当他第一次反问对方“难道是这间屋子懂中文吗”，对方竟然极其确信地回答“是的，屋子懂中文”。塞尔笑称自己虽然钦佩这种敢于语出惊人的勇气，但这纯属绝望之下的强辩：如果身处其中的人类无法跨越从句法规则到语义理解的鸿沟，没有意识的房间和纸箱更不可能跨越，除非假定房间里诞生了某种超越人类的独立意识。 另一种让塞尔哭笑不得的反驳来自公开辩论。有人质问他：假如外界用中文向中文屋提问“你懂中文吗”，而房间按照规则回答“懂，我理解中文”，那又该怎么说？塞尔现场演示了这种情境：房间收到一串未知符号（意思是“你懂中文吗”），他在手册里按形状检索规则，随后原样抄录并递出一串对应的新符号。这串符号在中文里翻译过来其实是在说：“你们怎么会问这么蠢的问题？难道看不出我懂中文吗？”然而身在其中的塞尔依然对其内容一无所知。房间之所以给出这个答案，完全是因为规则表格指示他画出这组形状，这丝毫不能证明任何真实的理解。 至于某些程序员辩解称，自然语言处理程序中往往有专门命名为“语义”（semantics）的代码模块，塞尔指出这根本是混淆概念：向系统添加所谓的语义模块，无非是写入了更多的计算机代码、实现了更多的句法操作，本质依然是纯句法。 > “The power of the syntactical definition of computation is you can implement it on electronic machines that can perform literally millions of computations in a very small amount of time.” —— 约翰·塞尔 - 计算的纯句法定义不是缺点而是其力量源泉，它使得机器能够在极短时间内执行天文数字级的运算（例如 Deep Blue 每秒据称可执行 2.5 亿次计算）。 - 无论向软件中堆叠多少层所谓的“语义代码”，在计算机实现层面增加的永远只是更多的句法指令。 --- ### [18:51](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=1131s) 信息时代的狂热与“超级AI叛乱”的荒谬幻觉 在转向讲座更核心的论题前，塞尔略作停顿喝了口水，并安抚台下听众，保证后续会有充裕的时间展开问答。他指出，许多人退而求其次地认为，即便计算机不懂中文，它至少在进行“信息处理”（information processing）和“计算”，因为这正是人类对机器的定义。然而，当下社会蔓延着对这一概念的严重泛化与误解。塞尔提到他为《纽约书评》（The New York Review of Books）撰写书评时读到的两本书：一本宣称人类步入了全新时代，并惊人地高呼“一切皆是信息”；而另一本则忧心忡忡地宣扬奇点威胁论。 这本充斥着危机感的著作宣称，计算机如今已经聪明到几乎与人类并驾齐驱，很快就会追平人类，并不可避免地演变为远超人类的超级智能。该书作者接着渲染道：届时人类必须当心，因为超级计算机很可能会厌倦被人类奴役和压迫的处境，进而奋起反抗推翻全人类；作者甚至宣称这可能是人类有史以来面临的最大挑战。塞尔直截了当地评价说，这种畅销书式的耸人听闻不仅浅薄，而且在哲学逻辑上完全站不住脚。 > “And pretty soon, they'll be just as smart as we are. And then I don't have to tell this audience the next step. They'll be much smarter than we are. And then look out because they might get sick of being oppressed by us. And they might simply rise up and overthrow us all.” —— 约翰·塞尔转述奇点恐慌论者的观点 - 塞尔认为将现实世界万物简化为“一切皆是信息”的泛信息论存在严重的哲学漏洞。 - 超级人工智能因“厌倦被压迫”而起义推翻人类的预言，是将人类特有的主观心理体验盲目投射给机器的无稽之谈。 --- ### [20:36](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=1236s) 两种智能：观察者独立与观察者相对 为了澄清人工智能领域的种种混乱，塞尔提出了一个根本性的哲学区分：“智能”（Intelligence）一词具有两种截然不同的含义——“观察者独立”（Observer-independent）与“观察者相对”（Observer-relative）。观察者独立的智能指代的是具有心理真实性（psychologically real）的客观现象。例如，塞尔比他养的伯恩山犬塔斯基（Tarski）更聪明。他风趣地插话道，自己养过四条狗，分别以现代逻辑学和哲学的先驱命名：弗雷格（Frege）、罗素（Russell）、路德维希（Ludwig，指维特根斯坦）以及塔斯基；尽管塔斯基体型太大塞不进汽车而未能来到现场，但它确实在与人类相同的心理学意义上拥有真正的智能，只是智力程度稍逊于人类。 与之形成鲜明对照的是计算机的所谓“智能”。计算机会处理信息、具有智能，但这完全是“观察者相对”的。人类设计并解释电子元件的高低电平状态，赋予其符号意义，使其表现得“仿佛在思考”。人们可以在认识论上对机器做出客观的陈述（epistemically objective claims），例如陈述深蓝（Deep Blue）在第几步走了哪只棋子，但这种智能归属完全取决于观察者的解读，机器本身内部根本不存在任何内在真实的智能。 ```mermaid flowchart TD accTitle: 智能的两种本体论分类 accDescr: 划分观察者独立与观察者相对两种智能形态。 subgraph OI["观察者独立（Observer-Independent）"] A["内在心理真实"] --> B["人类意识与思维"] A --> C["动物认知（如狗 Tarski）"] end subgraph OR["观察者相对（Observer-Relative）"] D["依赖外在解释"] --> E["深蓝国际象棋程序"] D --> F["数字计算机与算法"] end ``` 塞尔以 IBM 的深蓝击败国际象棋世界冠军加里·卡斯帕罗夫（Garry Kasparov）为例。媒体当时大肆宣传“深蓝击败卡斯帕罗夫，正如卡斯帕罗夫击败阿纳托利·卡尔波夫（Anatoly Karpov）一样”。但在塞尔看来，这种类比掩盖了根本差异：卡尔波夫与卡斯帕罗夫对弈时，双方都必须在主观上有意识地知道自己在下棋，知晓诸如“我以兵进王四开局”或“我的王后在棋盘左翼受到威胁”等语义事实；而深蓝对此一无所知，因为它在心理学意义上不具备任何知觉与意识。 > “In the intrinsic sense in which I have more intelligence than my dog, my computer has zero intelligence-- absolutely none at all.” —— 约翰·塞尔 - 认知科学中的一系列核心术语——记忆、知觉、决策、理性以及智能——都必须严格区分“心理学上的真实功能”与“观察者相对的机械行为”。 - 在人类与动物具备内在心理体验的严格意义上，现有一切计算机电路系统的内在智能均为零，它既非比人类多，亦非比人类少，而是彻底不存在。 ### [24:12](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=1452s) 观察者相对性与意识的不可替代性 John Searle 指出，机器中的所有“智能”完全取决于观察者：“All of the intelligence is in the eye of the beholder. It's all observer-relative.”（所有的智能都存在于观察者的眼中，它完全是观察者相对的）。在实用层面上，这通常毫无影响——如果有人造出了一辆能够自动驾驶的汽车，无论它是否具有意识、是否真正拥有字面意义上的智能，在工程应用上都无关紧要。这种区分只有在一种情况下至关重要：那就是当人们误以为将智能赋予一台完全没有内在智能的机器具有某种心理学意义时。 值得注意的是，人类用来将智能赋予计算机的心理过程本身必须依赖意识。也就是说，观察者相对性的赋予行为必须由有意识的主体来完成；但创造了这些观察者相对现象的意识本身，却绝非观察者相对的，它是一种内在的客观事实。 ```mermaid flowchart TD accTitle: 意识与观察者相对性的关系 accDescr: 展示人类内在意识如何赋予外部机器以观察者相对的计算与智能解释 subgraph Mind["内在心智领域 (Observer-Independent)"] A["人类与动物意识"] -->|"主观赋予与解释"| B["心理计算过程<br/>(例如：1 + 1 = 2)"] end subgraph Machine["机器与物理领域 (Observer-Relative)"] C["电子状态转换<br/>(口袋计算器 / 商业电脑)"] D["物理运动<br/>(物体自由落体运动)"] end A ==>|"赋予智能与计算意义"| C A ==>|"赋予算法解释: s = 1/2 gt²"| D ``` ### [25:28](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=1528s) 计算概念的历史演变与“算盘”的本质 Searle 进一步将这一区分推进到“计算”（Computation）本身。人们通常认为计算必然是计算机的内在特征，毕竟人类制造这些机器的目的就是为了计算。然而历史表明，“计算”一词经历了语义转变。Searle 回忆自己最初阅读 Alan Turing 的经典论文《Computing Machinery and Intelligence》（计算机器与智能）时的体会：图灵之所以没有将论文命名为《Computers and Intelligence》（计算机与智能），是因为在那个年代，“Computer”（计算者）指的是从事计算工作的人类个体，就像跑步者（Runner）或钢琴演奏家（Piano player）一样，指代的是执行该操作的人。 如今，这个词获得了系统的歧义性，既可以指独立于观察者的真实心理运算，也可以指观察者相对的机器物理过程。Searle 在黑板上随手演示了一个简单的加法计算：“一加一等于二”（$1 + 1 = 2$）。他强调，他在黑板上进行计算的过程是绝对内在且独立于观察者的，无论外界专家如何评说，他都是在有意识地进行真正的计算。反观袖珍计算器，当它完成同样的加减运算时，其过程纯粹是观察者相对的： > “Intrinsically all that goes on is a set of electronic state transitions that we have designed so that we can interpret computationally.” —— John Searle > （从本质上讲，内部所发生的不过是一组电子状态的转换，而这些转换是我们为了能够从计算角度对其进行解释而专门设计的。） 商业计算机中的计算是观察者相对的，但这并不意味着它是认识论上的主观臆断。Searle 调侃道，自己花了大量金钱购买 Apple 设备，就是为了让它运行更高级的程序，但这些程序和计算之所以成立，始终缺少了人类拥有的核心要素——意识。正是因为缺乏真正的内在心理现象，商业计算机才无法拥有内在的计算能力。只有当人们开始幻想“机械智能种族将会崛起并推翻人类”，或者给这些机器强加其他荒谬的心理学解释时，混淆这两者的危害才会爆发。 - **语言演化**：在 Turing 发表论文的时代，“Computer”仅指代执行数学计算的人类，如今该词演变为指代一类物理机械装置。 - **商业软硬件的本质**：更昂贵、更强大的硬件所具备的“更高智能”，本质上依然是使用者在观察者相对层面做出的属性赋予，并不具备任何内在的心理学意义。 ### [29:47](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=1787s) 陈旧的概念框架与“机器能否思考”的生物学解构 Searle 总结指出，当前哲学与科技界讨论心智与 AI 时使用的整个概念工具已经彻底过时了——诸如心理与物理的对立、社会与个体的对立，以及未能严格区分观察者相对与观察者独立的混淆术语，严重阻碍了人们的清晰思考。 针对经典的图灵之问“机器能否思考？”，Searle 直言这个提问从前提上就过时了。如果将“机器”定义为能够执行特定功能的物理系统，那么人类本身就是机器： > “Thinking is a biological process created in the brain by certain quite complex, but insufficiently understood neurobiological processes.” —— John Searle > （思考是由大脑中某些极其复杂、但尚未被充分理解的神经生物学过程所创造的一种生物学过程。） 因此，“机器能否思考？”的答案毫无疑问是肯定的，因为人类和动物作为生物机器每天都在思考。但面对派生出的核心问题，Searle 逐层进行了逻辑剖析： | 核心问题 | Searle 的判定 | 论证逻辑与核心依据 | |:--- |:--- |:--- | | **机器能否思考？** | **必然可以** | 人类与动物本身就是生物机器，思考是其物理器官产生的生物过程。 | | **人造机器能否思考？** | **原则上可以** | 类比于人造心脏泵血；只要人造器官复制了大脑同等的因果能力即可，但人类目前尚未完全弄清大脑产生意识的具体机理，因而无从下手。 | | **用非生物材料（非神经元）的人造机器能否思考？** | **完全开放的未知数** | 人造心脏无需肌肉组织也能泵血；人工大脑理论上也可能不需要脑组织，关键在于能否通过其他生物化学或物理材料“复制”（Duplicate）而非仅仅“模拟”（Simulate）大脑的因果效力。 | | **计算机能否思考？** | **取决于定义** | 任何能进行计算的有意识主体（如人类）既是计算机也能思考；但在观察者相对意义下的机器，由于缺乏内在意识，不能真正思考。 | Searle 特别强调了“模拟”（Simulate）与“复制”（Duplicate）之间的本质鸿沟。当前的计算机程序只是在形式系统上对思考过程进行模拟，这与制造出真正具备因果功能的人造大脑完全是两回事，正如电脑模拟天气绝不会带来真正的降雨一样。 ### [35:24](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=2124s) 观察者相对的计算：落体运动与万物皆为计算器 回到“计算机能否思考”的问题上，Searle 阐明：如果将计算机定义为“能够执行计算的主体”，那么任何有意识且能够进行基础算数运算的人类都是一台合格的计算机。Searle 推测，不仅人类这种计算者能够思考，而且任何具备思考能力的实体都必须具备执行简单计算的能力。 然而，一旦进入工程学与常规意义上的“观察者相对”语境，“计算”的外延就会发生灾难性的泛化。为了向听众展示这一结论，Searle 当场拿起一件物品让其自由下落，并指出它刚刚完成了一次著名物理函数的运算： $$s = \frac{1}{2}gt^2$$ 只要有人拿着足够精确的秒表记录时间，就可以推算出该物体下落的距离。Searle 强调，如果把这种能够映射数学公式的物理运动称为计算机，那么在观察者相对的意义下，世间万物都可以被称为计算机，因为任何物理系统只要被赋予了某种计算解释，都可以被视作在执行计算。这也彻底揭示了脱离观察者内在意识后，所谓“机器计算”在本体论上的空洞本质。 ### [36:16](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=2176s) 计算是观察者相对的，绝非心智的充分条件 John Searle 总结了他对计算本质的核心批判。他回顾了早年为“中文屋”论辩时的经典场面：他会抓起一支笔猛砸在桌上，声称这支笔就是一台数字计算机，只不过恰好运行着一个极其无聊的程序——“留在原地不动（stay there）”。令他感触颇深的是，从没有人能就此驳倒他，因为按照教材上的定义，这确实完全符合规范。在早期论战中，对手常常抛出一大堆技术术语来震慑他，比如指责他“漏掉了虚拟机与非虚拟机的区别”或“遗漏了转换器（transducers）”。Searle 坦承自己当时根本不知道什么是转换器或虚拟机，但学懂这些概念其实只要五分钟。 辩论的根本焦点在于：图灵所定义的“计算”本身，是否足以产生思考？对此，Searle 给出了坚定的否定回答。从观察者相对（observer-relative）的角度来看，计算根本不是自然界的固有事实，而是人类解释的产物。图灵机在自然界中是不存在的，它们只存在于我们对自然的诠释之中。单纯依靠形式化或句法（syntactic）定义的计算，永远不可能充分构成思维或任何认知过程。Searle 澄清自己对技术绝无偏见，他热爱且欢迎科技，也不认为技术的发展存在极限；他真正想做的是防止人们误解技术在哲学、心理学乃至科学上的真实含义。 > “Computation is not a fact of nature. It's a fact of our interpretation.” —— John Searle --- ### [38:41](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=2321s) Ray Kurzweil 发难：神经递质替换谬误与意识的测量难题 演讲结束后进入提问环节，首位提问者是著名未来学家 Ray Kurzweil。Kurzweil 表示自己与 Searle 已经就这个议题辩论了整整 18 年。他首先称赞了中文屋论证经久不衰的生命力，因为它切中了“计算这种确定性过程怎能产生思维”的直觉荒谬感。然而，Kurzweil 指出他在自己的著作《奇点临近》（The Singularity Is Near）中曾对 Searle 做出过一次反诘：既然神经元和大脑在某种层面上等同于机器，那么完全可以把 Searle 论证里的“计算机”替换成“人脑”，把“形式符号”替换成机制性的“神经递质浓度及相关机制”。 Kurzweil 当场宣读了这种替换后得出的荒谬结论：大脑只是在操纵毫无意义的神经递质浓度，大脑对此一无所知，因而人脑也不可能具备真正的理解力。以此为由，Kurzweil 提出了核心质问：Searle 坚称意识是像消化、泌乳、光合作用或有丝分裂一样的生物学过程，是大脑特定生物机制的产物，但我们究竟如何知道大脑是有意识的？Kurzweil 发问：“你如何知道我有意识？如果未来造出了一台反应极度逼真、看似具有意识的计算机，我们又该如何测量并确知它是真有意识，还是仅仅‘显得’有意识？” > “The neurotransmitter concentrations and related mechanisms themselves are quite meaningless. They only have the meaning we have attached to them. The human brain knows nothing of this. It just shuffles the neurotransmitter concentrations and related mechanisms. Therefore, the human brain cannot have true understanding.” —— Ray Kurzweil - **辩论历史**：Ray Kurzweil 指出两人就强人工智能与意识的交锋已有 18 年之久。 - **类比替换法**：Kurzweil 尝试证明，若将中文屋中对符号操作的贬低套用到生物神经机制上，同样会推导出人脑没有意识的荒谬推论。 - **意识判定难题**：对于未来在行为和语言响应上完美拟人的计算机，缺乏客观测量尺度去验证其主观状态。 --- ### [41:45](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=2505s) 大脑因果特异性与狗的意识常识 针对 Kurzweil 的替换论证，Searle 幽默地反驳称，随意偷换原词是政客惯用的把戏。他一针见血地指出计算机与人脑的本质差别：大脑是一个具体的因果机制（causal mechanism），通过极其复杂且尚未完全明晰的特定电化学过程产生意识。这种过程高度依赖具体的生物化学原理——例如可卡因之所以起作用，绝非它具备某种奇特的计算能力，而是因为它破坏了突触后受体对特定神经递质（去甲肾上腺素、多巴胺、GABA）的重吸收。相反，计算机的本质是抽象的数学过程，其运作与具体的物理实现（implementation）完全解耦；任何硬件只要能执行程序步骤即可，程序本身纯粹是形式化与句法层面的。因此，要制造有意识的机器，必须复制大脑生物原理的因果效力（causal powers），而不仅仅是模拟其计算。 ```mermaid flowchart TD accTitle: 因果机制与形式计算的对比 accDescr: 对比生物大脑的因果特异性机制与计算机硬件无关的形式计算特征 subgraph BrainMechanism["生物大脑 (因果特异性机制)"] B1["特异性电化学过程<br/>(神经递质、突触受体)"] --> B2["产生高阶生物特性<br/>(主观意识/感知体验)"] B3["物理实现不可替代<br/>(生物化学原理是核心)"] -.-> B1 end subgraph ComputerSystem["计算系统 (基质无关的形式系统)"] C1["抽象数学/符号操作<br/>(纯粹句法驱动)"] --> C2["输出行为模拟结果<br/>(不产生内在心智)"] C3["物理硬件任意可替<br/>(纸带、芯片或桌上的笔)"] -.-> C1 end ``` 谈到如何判定意识，Searle 指出这在现实生活中根本不是难题。他以自己的爱犬塔斯基（Tarski）和手上的智能手机为例：他确信狗有意识而手机没有。这种判断并非基于行为主义——狗见了他会摇尾巴、舔他的手，而人类朋友绝不会这么做；他之所以坚信狗有意识，是因为狗拥有与人类相似的感知解剖结构（眼睛、耳朵、皮肤）来介导输入刺激与输出行为。虽然白蚁有大约 10 万个神经元是否足以产生意识属于尚待专家解答的事实问题，但在人类日常生活中，“他人拥有意识”并不是一种假设或理论，而是如同“假定地板是坚硬的”一样的背景预设（background presupposition）。 > “The brain is a specific biological organ that operates on specific principles. And to create a conscious machine, we've got to know how to duplicate the causal powers of those principles.” —— John Searle - **递质生物学实例**：Searle 引用可卡因对突触间隙去甲肾上腺素（norepinephrine）、多巴胺（dopamine）及 GABA 重吸收的干扰，证明心智依赖底层化学因果，而非算法逻辑。 - **神经元规模玩笑**：提到白蚁拥有 10 万个神经元时，Searle 调侃自己“过一个狂欢周末就会损失 10 万个神经元”。 - **哲学术语投射**：Searle 戏谑地提到，虽然知道在座各位都有意识，但要探究大家是否正承受“晚期资本主义下后工业时代人类的焦虑（angst of post-industrial man under late capitalism）”，则需要深入交流才知晓。 --- ### [46:03](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=2763s) 技术成就与哲学实质的界限 现场观众随后提出一个延伸假想：既然现在人类尚未完全理解大脑，但假定未来不仅能完整模拟大脑，还能通过传感器和运动输出端与现实世界进行交互，这与 Searle 所说的产生意识的大脑还有何区别？Searle 回应道，在某些应用场景中，两者确实没有任何表观区别，而且这种区别对于实用目的并不重要。 Searle 举例提到自动驾驶：只要有一套算法程序能像人一样、甚至比人更好地驾驶汽车，那就是一项极好的科技成果。他分享了自己家庭的一段趣事——他的父亲曾是 AT&T 的电气工程师，毕生最大的失望就是 Searle 竟然跑去当了哲学家，而没有像他期望的那样进入贝尔实验室（Bell Labs）或麻省理工学院（MIT）。因此，他发自内心为工程和技术进步喝彩。但技术上的成功并不等同于解决了心理学和哲学的本质问题；能够完美驾车并不等于系统拥有了意识体验。在许多至关重要的领域，机器是否具备真实的“意识”依然具有不可替代的决定性意义。 > “The question is, what is the philosophical, psychological, scientific significance of that? And if you think, well, that means you've created consciousness, you have not. You have to have more to create consciousness.” —— John Searle ### [47:37](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=2857s) 终结者式的恐慌、Deep Blue 与模拟 vs. 复制 John Searle 驳斥了人们普遍担忧的“终结者机器起义推翻人类”的科幻末日论。他认为这绝非真正的现实威胁，因为机器根本不具备意识或任何有意识的心理状态。为了说明这种恐惧的荒谬，他打了个生动的比方：“这简直就像在说，鞋柜里的鞋子可能会站起来，踩得我们满地找牙。毕竟我们踩了它们好几个世纪，它们为什么不反抗呢？”这种假想在现实中根本不值得担忧。 现场观众随即提出一个更为严谨的问题：关心的不是系统的输出结果，而是机器内部是否存在与大脑类似的“因果同构性”（causal similarity）。Searle 承认这是一个事实层面的科学问题，但从现有证据看，计算机运作与大脑运作的同构性“并不怎么大”。以象棋程序为例，早期 AI 曾试图采访大师以探寻其思维过程并用计算机复现，但众所周知的 Deep Blue 并不是靠人类心理学机制下棋的。Deep Blue 每秒能够计算 2.5 亿个棋局位置。从博弈论角度看，国际象棋因具有“完全信息”（perfect information）和有限可能步数，在本质上是个很微不足道的游戏；之所以让人觉得难，纯粹是因为其面临巨大的“指数级组合爆炸”问题。 ```mermaid flowchart TD accTitle: 计算模拟与强AI复制的区别 accDescr: 图表展示了计算模拟（在各个领域建模过程）与强AI（宣称模拟即复制认知）的本质分歧。 subgraph SimulationStage["计算模拟 (Simulation)"] direction TB A["现实世界过程<br/>(天气/生物/心理)"] --> B["计算机数学模型<br/>(算法与符号推演)"] B --> C["有效研究工具<br/>(揭示推论与实验设计)"] end subgraph StrongAIClaim["强 AI 的谬误 (Duplication)"] direction TB B -.->|"错误等同"| D["真正的物理/心理复现<br/>(Duplication)"] D --> E["宣称计算本身产生意识与理解"] end ``` IBM 成功克服了指数搜索的难题，但该成就在心理学层面上毫无趣味，且诚实的 AI 研究人员也不会将其标榜为人类认知模拟的胜利。Searle 总结道，大脑实际过程在何种程度上与计算模拟重合是一个有趣的哲学与心理学课题，模拟也确实在许多领域大显身手；但弱 AI 的模拟绝不等同于强 AI 所坚称的“复制”（duplication）：“强 AI 宣称模拟不仅是模拟，而是复制。而这一点我们可以予以驳斥。” - Deep Blue 的核心能力：每秒可计算 2.5 亿个象棋局势走法（positions per second）。 - 博弈论特征：国际象棋具备完全信息（perfect information）和有限可能性，挑战仅在于指数复杂度。 - 模拟的限度：计算机模拟各类物理与生物过程是常规且有价值的科学手段，但符号计算过程不等于大脑生物过程。 --- ### [50:03](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=3003s) “你怎么证明你懂英语？”与拒绝认识论陷阱 一位听众提问道：“你能向我证明你理解英语吗？”Searle 戏谑地回答：“是的，但我懒得证明。”随后他用纯正的英国口音讲起自己在牛津大学基督堂学院（Christ Church）的旧事，笑称那是一所相当势利的学院，当时那里许多人就怀疑他到底懂不懂英语；他打趣说自己从来不声称自己讲的是 English，而是讲一种让不少英国人听到就浑身起鸡皮疙瘩的“美洲方言”（a dialect of American）。 提问者紧接着追问：既然你说你懂英语，我怎么知道你不是一个正在运行的计算机程序？Searle 指出这与此前 Ray 提出的问题如出一辙，并警告人们绝对不要掉入“认识论路线”（the epistemic route）的思维陷阱。认识论路线会诡辩称：“我们拥有的‘电脑有意识’的证据，和我们拥有的‘你有意识’的证据是一样多的。”Searle 直斥这种逻辑的脱节：“不，事实并非如此。”确认一个人具备意识并不需要玄奥艰深的神经生物学假说，我们能清晰辨认出物理结构： > “我不需要精妙的神经生物学理论也能断定，那就是你的眼睛，你是通过你的嘴巴在说话，而你的提问是一个表达提问意图的有意识意愿。” Searle 进一步对提问者打趣道：“相信我，如果你是 Google 本地制造出来的机器，那 Google 的技术进展可真比我想象的还要先进得多——但显然，你并不是机器。” --- ### [51:49](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=3109s) 常识性定义：什么是意识？ Google 的主持人 John Bracaglia 插播了一个来自内网提问系统 Dory 的高票问题：“在整场讲座中，你所使用的‘意识’的具体定义是什么？请尽可能具体阐述。” Searle 直截了当地指出，人们总常说意识难以定义，但他认为意识其实相当容易界定。科学上之所以尚无最终的科学定义，纯粹是因为目前还没有一套完整的科学理论支撑。如同科学史上的其他概念，常识性定义负责指明“研究目标”：比如水在常识中是“一种装在瓶子里、透明、无色、无味的液体”，而经过深入科学研究后才揭示出其本质是 H₂O。当前我们对意识的把握，正处在类似于把水看作无色透明液体的常识阶段。 Searle 给出了神经生物学家实际开展科研时所采纳的运行定义（working definition）： > “意识由所有那些感觉、感知或觉知的状态所组成，它们始于你早晨从无梦睡眠中醒来的那一刻，并在一天中持续进行，直到你再次入睡，或者以其他方式陷入他们所说的无意识状态。” 在这一界定下，梦境也是一种意识形态。意识的核心秘密和精髓，在于对任何一种意识状态而言，“处在那种意识状态中总有其特定的感受”（there's something it feels like to be in that conscious state）。正因如此，意识永远具备“主观本体论”（subjective ontology）。这正是神经科学家们试图攻克的问题：大脑究竟是如何产生这种主观体验的？它又是如何存在并发挥功能的？ --- ### [53:43](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=3223s) 进化概率、平行宇宙与意识的真正价值 现场听众提出了一个悲观边界的假设：即便假定人脑这种“思考机器”是靠漫长的进化过程塑造出来的，人类目前也可能根本没有足够的时间来造出思考机器——各种小概率事件（如小行星撞击灭绝恐龙等灾难）可能在我们的宇宙中阻断这一进程；提问者进一步指出，如果接受平行宇宙理论，也许在其他宇宙中早有某种人工意识产生，所以这件事在哲学上“也许根本不重要”。 对于“时间可能不够”的担忧，Searle 幽默回应：“关于我们可能没时间了这一点，嗯，我确实赶时间；但我认为我们应该拼尽全力去尝试。”人类科学史曾断言生命本身是无法参透的神秘现象，过去的“机械论”（mechanists）与“活力论”（vitalists）激烈争辩，而现在这些对立已被现代生物学超越，人类取得了不可思议的进展。 针对“意识也许无足轻重”的哲学虚无态度，Searle 给出了坚定的现实主义反驳： > “这可是我切身生活的地方。它对我至关重要。” Searle 阐明，世界上有许多事物无论在宇宙尺度上是否微不足道，但对人类而言却关乎根本——例如民主、性爱、文学、美食等；别人怎么看无关紧要，但所有这些体验都真真切切地以不同程度在个体生命中占据重要位置。 --- ### [55:22](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=3322s) 披萨消化实验：模拟绝非生化复制 有观众以人工心脏为类比发问：人工心脏可以使用完全不同于肉身心脏的材料与路径来达成功能、甚至走向复制；那么在尝试用计算机模拟大脑的过程中，我们是否有可能“歪打正着”，意外创造出真正的意识或实现复制？ Searle 直言不讳：“要相信这一点，你必须相信奇迹的存在。”他随即抛出了经典的“计算消化披萨”的思想实验来说明概念层面的混淆： ```mermaid flowchart LR accTitle: 披萨消化模拟的思想实验 accDescr: 展示了完美的披萨消化数学模拟与实际物质消化之间的鸿沟。 subgraph Simulation["计算机模拟"] A["精确算法与模型"] --> B["完美的披萨消化模拟"] B --> C["输出：图像 / 数据 / 数学图表"] end subgraph PhysicalWorld["物理生物世界"] D["真实披萨"] --> E["物理化学消化与吸收"] E --> F["真实的营养与胃肠生化反应"] end C -.->|"无法取代"| E ``` 计算机几乎能对任何精确描述的过程做模拟，人们完全可以写出一套对“消化披萨”进行完美建模的程序。但没有人会在写完程序后，买一块真正的披萨塞进计算机机箱里，指望电脑把披萨消化掉——模拟出来的只是图表、图像或数学公式。Searle 澄清自己完全支持利用计算机辅助科学研究，它在推演公理假说、设计验证实验上是极佳的生物学工具。但若妄想仅凭认知行为的“计算模拟”就能奇迹般替代大脑底层的真实“生物化学因果机制”，在逻辑和物理常理上根本讲不通。 --- ### [57:02](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=3422s) 图灵测试的缺陷与观察者相关的主观智能 针对 Dory 平台提出的第二个问题——“鉴于人类很容易受蒙骗并过度高估机器智能，能否提出一个比图灵测试（Turing test）更好、更不容易产生假阳性（false positive）的通用智能测试标准？”Searle 再次回到他的根本立场。 Searle 指出，走出这一泥潭的第一步，是必须彻底区分两种概念： 1. **真实的、内在的、独立于观察者的智能**（genuine intrinsic observer-independent intelligence）； 2. **观察者相对的智能**（observer-relative intelligence）。 观察者相对的智能完全取决于“旁观者的眼光”（in the eye of the beholder）。在这种定义下，人们可以把“智能”投射给任何物体。为了直观展现这一谬误，Searle 拿起身边的道具进行现场演示：如果只要能计算数学公式就算有智能，那么随手扔下一个物体，它在下落时都在“计算”重力加速度函数 $s = \frac{1}{2}gt^2$，那它就拥有了巨大的智能；他又指向另一个日常工具，打趣说它能够区分每一根毛发，难道它也拥有惊人的高超智能？将观察者赋予的功能当作物体的内在心理现实，正是图灵测试最容易产生假阳性判断的根源所在。 ### [57:50](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=3470s) 图灵测试的根本缺陷与因果能力复制 约翰·塞尔（John Searle）继续阐明为何智能无法通过单纯的外部行为来断定。他打趣道，就算是在自己这颗头发稀疏的脑袋里，思考也是确实发生的；但由于智能是“观察者相关”（observer-relative）的，评判它必须先确立标准。在塞尔看来，图灵测试存在致命缺陷：其输入和输出之所以能被视为问题与答案，完全依赖于外部人类的解释。塞尔给出了一个决定性的断言：“我今天整个讨论的一个底线就是图灵测试失败了。它根本无法为你提供一个关于智能的测试。”（“One bottom line of my whole discussion today is that the Turing test fails. It doesn't give you a test of intelligence.”） 现场一位观众提出质疑，认为塞尔似乎把人类拥有意识、狗拥有意识、且意识源自特定生物组织当成了不可动摇的教条，这种论调听起来非常像“智能设计论”（intelligent design）：因为进化与创造宇宙过于复杂，就断定它不可能由无生命的惰性物质演化而来。观众反问，从变形虫到狗之间，究竟在哪里画出意识的分界线？既然人类甚至狗的意识都是进化过程中“涌现”（emergent）出来的，为什么意识就不能从一个高度分布式、联网、具备海量计算能力甚至与生物系统接驳的计算系统中涌现出来？ 塞尔幽默地反驳道，奇迹总是有可能的，但你不能因此问“你怎么知道化学反应不会把这把梳子变成有意识的梳子”。他强调大脑产生意识的机制非常具体，任何创造意识的系统都必须复制这些因果能力（causal powers），而不仅仅是进行模拟。为了形象地说明这一点，塞尔借用了空气动力学的类比：制造飞行器并不需要羽毛，但必须复制鸟类在地球大气层中克服重力的因果能力；飞机之所以能飞，正是因为它利用了同样的物理原理——伯努利定律（Bernoulli's principle）去克服重力。单纯去模拟输入输出函数的某些形式化结构，绝不可能产生真正的意识。 ```mermaid flowchart LR accTitle: 因果复制与形式模拟的对比 accDescr: 展现鸟与飞机复制因果能力，而计算机仅模拟形式结构的区别 subgraph Replication["因果能力复制 (Duplication)"] A["鸟类的飞行机制"] -->|"遵循伯努利原理克服重力"| B["飞机"] C["大脑生物机制"] -->|"复制因果能力"| D["真实意识"] end subgraph Simulation["形式结构模拟 (Simulation)"] E["输入/输出形式结构"] -->|"仅仅运行程序/计算"| F["无语义的模拟（非意识）"] end ``` --- ### [01:00:31](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=3631s) 机器自省、复杂度与“中文屋”的形式推导 针对观众提出的“机器可以自我改进，变形虫能演化成狗，为何复杂计算系统不能涌现意识”的追问，塞尔澄清自己从未否定变形虫能演化成高等生物，但单细胞的变形虫本身缺乏复制大脑因果机制的生理结构。谈到“系统复杂度”时，塞尔指出复杂度同样是观察者相关的——没有度量标准（metric），复杂度就毫无意义，甚至大脑产生意识的机制最终可能非常简单。他重申自己的理论从不排斥“计算机具有意识”，因为人类本身就是有意识的生物计算机；但他强调：“如果你认为自己命悬一线、必须造出一台能产生意识的机器，你绝不会坐在控制台前开始用某种编程语言写程序。那完全是南辕北辙。”（“If you're thinking about your life depends on building a machine that creates consciousness, you don't sit down your console and start programming things in some programming language. It's the wrong way to go about it.”） 另一位观众提出了一个极具技术色彩的思想实验：如果把谷歌翻译（Google Translate）的代码反汇编（disassembly），让塞尔在中文屋里用纸笔去一步步执行所有汇编指令，塞尔要么耗费数千年计算，要么把指令反编译为英文并进行深度优化；在此过程中，塞尔必然会理解各种变量和子程序之间的深层关系，难道这不能算作一种对中文的“理解”从中涌现吗？ 塞尔非常赞赏这个切入点，并坦然承认：如果让人类在物理上执行一套能够回答中文问题的庞大程序，确实需要耗费数百万年时间。但塞尔随即指出，中文屋思想实验本身只是为了辅助说明背后的逻辑推导。他将核心论证归纳为清晰的公理化三段论： - **公理一**：计算机程序完全是形式化的或句法性的（formal / syntactical）；严格来说，“同一套被实现的程序”这一概念定义了一个纯粹由形式或句法所规定的等价类。 - **公理二**：人类心灵具有语义（semantics）。 - **公理三**：纯粹的句法本身不足以构成语义（中文屋思想实验正是这一公理的直观展现）。 - **推论**：如果计算机仅由其句法操作来定义，那么计算机操作本身永远不足以产生真正的理解。 --- ### [01:04:21](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=3861s) 生物学认知机制与神经科学对意识的研究 面对观众关于“能否通过实验区分纯句法长期运行涌现的结果与真正语义”的追问，塞尔坚决反对走认识论怀疑主义（epistemic route）的路子。他指出，我们不能因为“无法从外部区分思考机器与非思考机器”就陷入虚无，因为现代神经科学已经积累了压倒性的生物机制证据。 塞尔分享了自己刚接触脑科学时的趣事：为了弄懂大脑，他买回了所有大学一年级的基础教科书，因为初级教材最容易读懂。书中提到猫的色觉（color vision）与人类完全不同，视觉体验大相径庭。塞尔当时第一反应是：“天哪，这些科学家难道变成过猫吗？他们解决了‘其他猫的心灵问题’（other cats' mind problem）？他们知道当一只猫是什么感觉吗？”但答案其实非常平实：科学家完全了解猫的色觉，因为他们直接检测了猫眼底的感光受体（color receptors）。同样，人们完全清楚狗能听到人类听不见的高频声波，这一切都不是靠猜想，而是通过观察生物感知仪器的生理构造得出的客观结论。 活动主持人约翰·布拉卡利亚（John Bracaglia）提示这将是最后一个提问环节，塞尔兴致盎然地打趣回应：“行啊，我完全可以跟你们聊上一整下午，我太喜欢这套东西了。”（“I'm prepared to go all afternoon. I love this kind of crap.”） 最后一位观众提问：塞尔在演讲开头曾提到早年神经科学家对意识研究避之不及，但如今许多神经科学家正在通过向被试呈现处于知觉阈值（threshold of perceptibility）边缘的刺激，并对比阈上与阈下的脑部响应差异来进行研究；这种研究路径是否走在正确的轨道上？塞尔对此给出了高度积极的评价。他感慨自己活了这么大岁数，亲眼见证的最大幸事之一就是神经科学界对意识的研究如今已经变成了一门蓬勃发展的显学。他回忆早年去加州大学旧金山分校（UCSF）劝说科学家研究意识时，对方根本不想听一个哲学家唠叨，一位著名的神经科学家甚至直言相告：“在我的学科里，对意识感兴趣是可以的，但你得先拿到终身教职才行。”（“In my discipline, it's OK to be interested in consciousness, but get tenure first.”）而如今时代彻底变了，年轻学者完全可以凭借研究意识来获得终身教职；塞尔坚信，当神经科学家真正揭示大脑制造意识的确切生物机制时，那将是人类历史上无比巨大的智力突破。 ### [01:07:39](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=4059s) 神经相关物研究的两条路径：“积木法”对决“统一场” 面对观众关于寻找意识的神经相关物（Neural Correlates of Consciousness）是否可行以及该提出何种问题的提问，约翰·塞尔（John Searle）用他标志性的直率与幽默开场：“I happened to be interested in this crap.” 他向听众推荐了自己在《神经科学年鉴》（*Annual Review of Neuroscience*）上发表的一篇题为《意识》（Consciousness）的论文，并指出当今的脑科学研究主要分为两大阵营。第一种阵营采取的是所谓“积木法”（building block approach），研究者试图通过寻找单一感知经验的具体神经相关物来切入——例如大脑看到红色物体或听到中央C（middle C）音符时的对应脑区活动。持有这种观点的学者坚信，只要能彻底破解大脑如何产生“红色的主观体验”，就等同于攻克了整个意识之谜；这就如同当年破译 DNA 一样，只要掌握了底层核心机制与通用原理，便无需逐一去解释每一种复杂的表现型（phenotype）。 然而，塞尔对这种积木式研究持保留态度，认为其至今未取得实质性突破。他主张另一种更有前景的研究方向，即应当将意识理解为一个“统一场”（unified field）。在这一理论框架下，感知活动本质上并不是在凭空制造意识，而是在调制既有的意识场。塞尔指着现场观众举例说明：当他看见某位听众衣服上的红色时，这一知觉并没有从无到有地生成一段意识，而是修改了他原本就处于激活状态的意识场，使他在原本的体验中增添了此前没有的“红色的感觉”。尽管科学史上的大多数成就都建立在“原子论式”（atomistic）的方法论之上——先从小尺度微观事物着手，再层层推演至宏观系统——但塞尔强调，这种还原论在意识问题上遭遇了瓶颈。人类 experiences（如看红或听音）必须依附于一个预先存在的统一意识场，研究者若不首先弄清大脑如何生成这个总体的意识背景，单纯研究局部知觉的神经对应物就如同缘木求鱼。 ```mermaid flowchart TD accTitle: 意识神经机制研究的两大进路 accDescr: 展示积木法与统一场进路在解释大脑如何形成主观感知时的根本差异。 subgraph BB["积木法（Building Block Approach）"] B1["寻找个别感知的神经相关物<br/>(如看到红色、听中央C)"] --> B2["破解单一感官体验的成因"] B2 --> B3["推导全局意识的一般规律<br/>(类比DNA解析表现型)"] end subgraph UF["统一场进路（Unified Field Approach）"] F1["大脑大尺度运作机制"] --> F2["预先建立起统一的意识场"] F2 --> F3["外部知觉输入<br/>(如红衬衫、声音)"] F3 --> F4["修改既有意识场<br/>(而非凭空生成意识)"] end ``` - 约翰·塞尔发表在《神经科学年鉴》（*Annual Review of Neuroscience*）上的关键论文篇名为《意识》（"Consciousness"），论述了研究意识神经生物学基础的核心框架。 - 积木法假定意识是由各种离散的知觉拼图拼接而成的；而统一场假定意识是连续的、先验的整体背景场。 - 塞尔总结道：“Think of perception not as creating consciousness, but as modifying the conscious field.” --- ### [01:09:34](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=4174s) 意识之谜的硬科学解法与讲座闭幕 探讨统一场进路的真正难点在于，科学家必须回答大脑的大尺度神经结构究竟是如何协同产生主观意识的，而这一领域目前仍充满未知。塞尔指出，现代脑成像技术揭示了一个令人困惑的现象：在磁共振成像（MRI）扫描中，处于清醒意识状态的大脑与无意识状态的大脑在物理影像上看起来惊人地相似。两者之间必然存在决定性的神经机制差异，但当前的科学手段尚未能精确定位这种差异到底在哪里。塞尔坦言自己近期正专注于其他哲学课题，但他始终期待神经科学家能够拿出一份确切的答案，讲清楚清醒大脑与无意识大脑之间究竟存在什么本质不同，从而从物理机制上彻底说明意识的存在。 塞尔强调，自己在此提出的推论属于“神经生物学猜想”（neurobiological speculation）。意识难题的终极答案绝不会来自纯粹的哲学论证或思辨辩论，而必须依赖于实验室里扎实艰辛的硬核实证研究——去真正揭示大脑中生成意识的生物学机制到底是什么，以及它们究竟如何具体运作。讲座接近尾声时，主持人约翰·布拉卡利亚（John Bracaglia）走上讲台向塞尔致敬并宣布活动结束。塞尔对 Google 的邀请表示感谢，全场报以热烈掌声。 - 塞尔指出核磁共振等技术尚未能清晰界定清醒与昏迷/无意识之间的决定性脑状态：“The problem is in an MRI, that conscious brain looks a lot like the unconscious brain.” - 塞尔阐明哲学的局限与硬科学的终极裁判权：“I mean, I'm going to be answered not by a philosophical argument, but by somebody who does the hard research of figuring out exactly what are the mechanisms in the brain the produce consciousness and exactly how do they work.” - 主持人约翰·布拉卡利亚在结语中致意：“John, it's been an immense, immense honor to be here with you today.” ## Notable quotes > [01:38](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=98s) The point is this: you can have an epistemically objective science of consciousness, even though consciousness is ontologically subjective. — John Searle > [11:57](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=717s) Syntax is not semantics. And simulation is not duplication. — John Searle > [36:16](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=2176s) Computation is not a fact of nature. It's a fact of our interpretation. — John Searle > [29:47](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=1787s) Thinking is a biological process created in the brain by certain quite complex, but insufficiently understood neurobiological processes. — John Searle > [41:45](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=2505s) The brain is a specific biological organ that operates on specific principles. And to create a conscious machine, we've got to know how to duplicate the causal powers of those principles. — John Searle > [20:36](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=1236s) In the intrinsic sense in which I have more intelligence than my dog, my computer has zero intelligence-- absolutely none at all. — John Searle > [57:50](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=3470s) One bottom line of my whole discussion today is that the Turing test fails. It doesn't give you a test of intelligence. — John Searle > [01:07:39](https://www.youtube.com/watch?v=rHKwIYsPXLg&t=4059s) Think of perception not as creating consciousness, but as modifying the conscious field. — John Searle ## Mentioned John Searle, John Bracaglia, Ray Kurzweil, Google, Singularity Network, Talks at Google, University of California, Berkeley, Jean Nicod Prize, National Humanities Medal, Mind & Brain Prize, Chinese room, Sloan Foundation, Alan Turing, Turing test, Deep Blue, Garry Kasparov, Anatoly Karpov, The New York Review of Books, The Singularity Is Near, Apple, AT&T, Bell Labs, MIT, University of California, San Francisco, Christ Church, Oxford, Google Translate, Annual Review of Neuroscience ## Video description John Searle is the Slusser Professor of Philosophy at the University of California, Berkeley. His Talk at Google is focused on the philosophy of mind and the potential for consciousness in artificial intelligence. This Talk was hosted for Google's Singularity Network. John is widely noted for his contributions to the philosophy of language, philosophy of mind and social philosophy. Searle has received the Jean Nicod Prize, the National Humanities Medal, and the Mind & Brain Prize for his work. Among his notable concepts is the "Chinese room" argument against "strong" artificial intelligence. [Source](https://www.youtube.com/watch?v=rHKwIYsPXLg)

[1:38] John Searle: The point is this: you can have an epistemically objective science of consciousness, even though consciousness is ontologically subjective. [11:57] John Searle: Syntax is not semantics. And simulation is not duplication. [36:16] John Searle: Computation is not a fact of nature. It's a fact of our interpretation. [29:47] John Searle: Thinking is a biological process created in the brain by certain quite complex, but insufficiently understood neurobiological processes. [41:45] John Searle: The brain is a specific biological organ that operates on specific principles. And to create a conscious machine, we've got to know how to duplicate the causal powers of those principles. [20:36] John Searle: In the intrinsic sense in which I have more intelligence than my dog, my computer has zero intelligence-- absolutely none at all. [57:50] John Searle: One bottom line of my whole discussion today is that the Turing test fails. It doesn't give you a test of intelligence. [67:39] John Searle: Think of perception not as creating consciousness, but as modifying the conscious field.

[11:57] John Searle: The implemented computer program by itself is never going to be sufficient for human understanding because human understanding has more than syntax. [2:23] John Searle: You can have an epistemically objective science of consciousness, even though consciousness is ontologically subjective. [32:01] John Searle: The question, can you build an artificial brain that can think, is like the question, can you build an artificial heart that pumps blood. [18:51] John Searle: In the strict sense, in the observer-independent sense in which you and I have intelligence, there is zero intelligence in the computer. [29:47] John Searle: I'm a machine. You're a machine. And my guess is only machines could think. [67:39] John Searle: Think of perception not as creating consciousness, but as modifying the conscious field. [29:47] John Searle: Thinking is a biological process created in the brain by certain quite complex, but insufficiently understood neurobiological processes. [51:49] John Searle: Consciousness consists of all those states of feeling or sentience or awareness that begin in the morning when you awake from a dreamless sleep.

[0:01] JOHN BRACAGLIA: My name is John Bacaglia. [0:03] And I'm a Googler working in YouTube operations. [0:05] I also lead a group called the Singularity Network, [0:09] an internal organization focused on discussions [0:11] and rationality in artificial intelligence. [0:14] I'm pleased to be here today with Mr. John Searle. [0:19] As a brief introduction, John Searle [0:21] is the Slusser Professor of Philosophy at the University [0:23] of California-Berkeley. [0:25] He is widely noted for his contributions [0:27] to the philosophy of language, philosophy of mind, [0:30] and social philosophy. [0:32] John has received the Jean Nicod Prize, the National Humanities [0:35] Medal in the Mind and Brain prize for his work. [0:38] Among his noble concepts is the Chinese room argument [0:42] against strong artificial intelligence. [0:45] John Searle, everyone. [0:46] [APPLAUSE] [0:47] JOHN SEARLE: Thank you Thank you. [0:53] Many thanks. [0:55] It's great to be back at Google. [0:58] It is a university outside of a university. [1:00] And sometimes, I think, this is what a university [1:02] ought really to look like. [1:04] Anyway, it's just terrific to be here. [1:06] And I'm going to talk about some-- well, [1:08] I'm going to talk about a whole lot of stuff. [1:10] But, basically, I want to start with talking [1:12] about the significance of technological advances. [1:16] And America, especially, but everybody, really, [1:19] is inclined to just celebrate the advances. [1:23] If they got a self-driving car who the hell [1:25] cares about whether or not it's conscious. [1:28] But I'm going to say there are a lot of things that [1:32] matter for certain purposes about the understanding [1:35] of the technology. [1:36] And that's really what I'm going to talk about. [1:38] Now to begin with, I have to make a couple rather [1:41] boring distinctions because you won't really [1:43] understand contemporary intellectual life if you don't [1:46] understand these distinctions. [1:49] In our culture, there's a big deal about objectivity [1:52] and subjectivity. [1:54] We strive for an objective science. [1:57] The problem is that these notions are systematically [2:00] ambiguous in a way that produces intellectual catastrophes. [2:04] They're ambiguous between a sense, which is epistemic, [2:10] where epistemic means having to do with knowledge-- [2:13] epistemic-- and a sense, which is ontological, [2:19] where ontological means having to do with existence. [2:23] I hate using a lot of fancy polysyllabic words. [2:25] And I'll try to keep them to a minimum. [2:27] But I need these two, epistemic and ontological. [2:31] Now the problem with objectivity and subjectivity [2:34] is that they're systematically ambiguous-- [2:42] I'll just abbreviate subjectivity-- [2:44] between an epistemic sense and an ontological sense. [2:48] Epistemically, the distinction is between types [2:52] of knowledge claims. [2:54] If I say, Rembrandt died in 1606, well-- no, [2:58] he didn't die then. [2:59] He was born then. [3:00] I'd say Rembrandt was born in 1606. [3:03] That is to say, it's a matter of objective fact. [3:06] That's epistemically objective. [3:09] But if I say Rembrandt is the greatest [3:11] painter that ever lived, well, that's a matter of opinion. [3:14] That is epistemically subject. [3:18] So we have epistemic objectivity and subjectivity. [3:22] Underlying that is a distinction in modes of existence. [3:26] Lots of things exist regardless of what anybody thinks. [3:30] Mountains, molecules, and tectonic plates [3:33] have a mode of existence that is ontologically objective. [3:38] But pains and pickles and itches, [3:40] they only exist insofar as they are experienced by a subject. [3:45] They are ontologically subjective. [3:48] So I want everybody to get that distinction [3:50] because it's very important because-- well, [3:53] for a lot of reasons, but one is lots of phenomena that [3:57] are ontologically subjective admit of an account which [4:02] is epistemically objective. [4:05] I first got interested in this kind of stuff. [4:07] I thought, well, why don't these brain guys solve [4:10] the problem of consciousness. [4:12] And I went over UCSF to their neurobiology gang [4:16] and told them, why the hell don't [4:17] you guys figure out how the brain causes consciousness? [4:20] What am I paying you to do? [4:22] And their reaction was, look, we're doing science. [4:26] Science is objective. [4:29] And you, yourself, admit that consciousness is subjective. [4:33] So there can't be a science of consciousness. [4:36] Now you'll all recognize that's a fallacy of ambiguity. [4:39] Science is indeed epistemically objective [4:44] because we strive for claims that [4:47] can be established as true or false, [4:49] independent of the attitudes of the makers [4:51] and interpreters of the claim. [4:53] But epistemic objectivity of the theory [4:59] does not preclude an epistemically objective account [5:04] of a domain that's ontologically subjective. [5:07] I promised you I wouldn't use too many big words, [5:09] but anyway there are a few. [5:10] The point is this. [5:11] You can have an epistemically objective science [5:14] of consciousness, even though consciousness [5:18] is ontologically subjective. [5:20] Now that's going to be important. [5:22] And there's another distinction. [5:25] Since not everybody can see this, [5:26] I'm going to erase as I go along. [5:29] There's another distinction which is crucial. [5:32] And that's between phenomena that are observer-independent. [5:40] And there I'm thinking of mountains and molecules [5:44] and tectonic plates, how they exist regardless [5:47] of what anybody thinks. [5:49] But the world is full of stuff that [5:51] matters to us that is observer-relative. [5:56] It only exists relative to observers and users. [6:01] So, for example, the piece of paper in my wallet is money. [6:08] But the fact that makes it money is not a fact of its chemistry. [6:11] It's a fact about the attitudes that we have toward it. [6:15] So money is observer-relative. [6:19] Money, property, government, marriage, universities, Google, [6:22] cocktail parties, and summer vacations [6:25] are all observer-relative. [6:30] And that has to be distinguished from observer-independent. [6:34] And notice now, all observer-relative phenomenon [6:38] are created by human consciousness. [6:41] Hence, they contain an element of ontological subjectivity. [6:47] But you already know that you can have, [6:49] in some cases, an epistemically objective [6:53] science of a domain that is observer-relative. [6:57] That's why you can have an objective science of economics [7:01] even though the phenomena studied by economics [7:04] is, in general, observer-relative, [7:06] and hence contains an element of ontological subjectivity. [7:11] Economists tend to forget that. [7:12] They tend to think that economics [7:14] is kind of like physics, only it's harder. [7:19] When I studied economics, I was appalled. [7:22] We learned that marginal cost equals marginal revenue [7:25] in the same tone of voice that in physics we [7:28] learned that force equals mass times acceleration. [7:30] They're totally different because the stuff in economics [7:35] is all observer-relative and contains an element [7:39] of ontological subjectivity. [7:41] And when the subjectivity changes-- ffft-- [7:44] the whole thing collapses. [7:45] That was discovered in 2008. [7:48] This is not a lecture about economics. [7:50] I want you to keep all that in mind. [7:52] Now that's important because a lot of the phenomena that [7:57] are studied in cognitive science, [8:00] particularly phenomena of intelligence, cognition, [8:03] memory, thought, perception, and all the rest of it [8:06] have two different senses. [8:08] They have one sense, which is observer-independent, [8:14] and another sense, which is observer-relative. [8:17] And, consequently, we have to be very careful [8:21] that we don't confuse those senses because many [8:25] of the crucial concepts in cognitive science [8:28] have as their reference phenomena [8:32] that are observer-relative and not observer-independent. [8:37] I'm going to get to that. [8:39] OK, everybody up with us so far? [8:41] I want everything to sound so obvious you [8:43] think, why does this guy bore us with these platitudes? [8:45] Why doesn't he say something controversial? [8:50] Now I'm going to go and talk about [8:52] some intellectual history. [8:54] Many years ago, before any of you were born, [8:57] a new discipline was born. [9:00] It was called cognitive science. [9:03] And it was founded by a whole bunch of us [9:05] who got sick of behaviorism in psychology, effectively. [9:08] That was the reason for it. [9:09] And the Sloan Foundation used to fly us around to lecture, [9:16] mostly to each other. [9:17] But anyway, that's all right. [9:19] We were called Sloan Rangers. [9:20] And I was invited to lecture to the Artificial Intelligence Lab [9:25] at Yale. [9:26] And I thought, well, christ, I don't know anything [9:28] about artificial intelligence. [9:30] So I went out and bought a book written by the guys at Yale. [9:34] And I remember thinking, $16.95 plus tax-- money wasted. [9:41] But it turned out I was wrong. [9:45] They had in there a theory about how computers could understand. [9:49] And the idea was that you give the computer a story. [9:53] And then you ask the computer questions about the story. [9:57] And the computer would give the correct answer to the questions [10:00] even though the answer was not contained in the story. [10:04] A typical story. [10:05] A guy goes into a restaurant and orders a hamburger. [10:07] When they brought him the hamburger, [10:09] it was burned to a crisp. [10:11] The guy stormed out of the restaurant [10:13] and didn't even pay his bill. [10:14] Question, did the guy eat the hamburger? [10:17] Well, all of you computers know the answer to that. [10:19] No, the guy didn't eat the hamburger. [10:21] And I won't tell you the story where the answer is yes. [10:24] It's equally boring. [10:28] Now, the point was this proves that the computer really [10:33] understands the story. [10:35] So there I was on my way to New Haven on United Airlines [10:39] at 30,000 feet. [10:40] And I thought, well, hell, they could give me [10:42] these stories in Chinese. [10:44] And I could follow the computer program for answering stories. [10:47] And I don't understand a word of the story. [10:50] And I thought, well, that's an objection [10:53] they must have thought of. [10:55] And besides that won't keep me going [10:57] for a whole week in New Haven. [11:00] Well, it turned out they hadn't thought of it. [11:02] And everybody was convinced I was wrong. [11:06] But interestingly they all had different reasons [11:08] for thinking I was wrong. [11:10] And the argument has gone on longer than a week. [11:13] It's gone on for 35 years. [11:15] I mean, how often do I have to refute these guys? [11:17] But anyway, let's go through it. [11:22] The way the argument goes in its simplest version [11:24] is I am locked in a room full of Chinese-- well, [11:29] they're boxes full of Chinese symbols and a rule book [11:32] in English for manipulating the symbols. [11:34] Unknown to me, the boxes are called a database, [11:38] and the rule book is called a program. [11:40] In coming in the room, I get Chinese symbols. [11:44] Unknown to me, those are questions. [11:46] I look up what I'm supposed to do. [11:48] And after I shuffle a lot of symbols, [11:49] I give back other symbols. [11:51] And those are answers to the questions. [11:54] Now we will suppose-- I hope your bored [11:56] with this, because I am. [11:57] I mean, I've told this story many times. [12:01] We will suppose that they get so good at writing the program, [12:04] I get so good at shuffling the symbols, [12:06] that my answers are indistinguishable [12:08] from a native Chinese speaker. [12:10] I pass the Turing test for understanding Chinese. [12:13] All the same, I don't understand a word of Chinese. [12:16] And there's no way in the Chinese room [12:18] that I could come to understand Chinese because all I am [12:23] is a computer system. [12:24] And the rules I operate are a computer program. [12:28] And-- and this is the important point-- [12:30] the program is purely syntactical. [12:33] It is defined entirely as a set of operations [12:37] over syntactical elements. [12:39] To put it slightly more technically, [12:41] the notion same implemented program [12:43] defines an equivalence class that [12:45] is specified completely independently of any physics [12:48] and, in particular, independent of the physics [12:51] of its realization. [12:53] The bottom line is if I don't understand [12:56] the questions and the answers on the basis of implementing [13:00] the program, then neither does any other digital computer [13:03] on that basis because no computer [13:06] has anything that I don't have. [13:09] Computers are purely syntactical devices. [13:13] Their operations are defined syntactically. [13:15] And human intelligence requires more than syntax. [13:19] It requires a semantics. [13:20] It requires an understanding of what's going on. [13:25] You can see this if you contrast my behavior in English [13:30] with my behavior in Chinese. [13:31] They ask me questions in English. [13:35] And I give answers in English. [13:37] They say, what's the longest river in the United States? [13:39] And I say, well, it's the Mississippi, [13:41] or the Mississippi-Missouri, depending [13:44] on if you count that as one river. [13:46] They ask me in Chinese, what's the longest river in China? [13:50] I don't know what the question is or what it means. [13:52] All I got are Chinese symbols. [13:54] But I look up what I'm supposed to do with that symbol, [13:57] and I give back an answer, which is the right answer. [14:00] It says, it's the Yangtze. [14:02] That's the longest river in China. [14:03] I don't know any of that. [14:04] I'm just a computer. [14:06] So the bottom line is that the implemented computer program [14:12] by itself is never going to be sufficient [14:15] for human understanding because human understanding has [14:18] more than syntax. [14:20] It has a semantics. [14:22] There are two fundamental principles [14:24] that underlie the Chinese room argument. [14:27] And both of them seem to me obviously true. [14:30] You can state each in four words. [14:32] Syntax is not semantics. [14:35] And simulation is not duplication. [14:40] You can simulate-- you're going to have [14:42] plenty of time for questions. [14:43] How much time we got, by the way? [14:45] I want to-- [14:45] JOHN BRACAGLIA: We'll leave time for questions at the end. [14:48] JOHN SEARLE: I want everybody that [14:49] has a question to have a chance to ask the question. [14:51] Anyway, that's the famous Chinese room argument. [14:55] And it takes about five minutes to explain it. [14:57] Now you'd be amazed at the responses I got. [15:03] They were absolutely breathtaking [15:05] in their preposterousness. [15:08] Now let me give you some answers. [15:11] A favorite answer was this. [15:14] You were there in a room. [15:16] You had all those symbols. [15:18] You had a box. [15:19] You probably had scratch paper on which to work. [15:22] Now, it wasn't you that understood. [15:26] You're just a CPU, they would say with contempt, the Central [15:30] Processing Unit. [15:31] I didn't know what any of these words meant in those days. [15:34] CPU, it's the system that understands. [15:41] And when I first heard this, I mean, [15:43] the room understands Chinese, I said to the guy. [15:45] And he said, yes, the room understands Chinese. [15:49] Well, it's a desperate answer. [15:51] And I admire courage. [15:53] But it's got a problem. [15:55] And that is the reason I don't understand [15:57] is I can't get from the syntax to the semantics. [16:00] But the room can't either. [16:01] How does the room get from the syntax of the computer [16:05] program of the input symbols to the semantics [16:09] of the understanding of the symbols? [16:11] There's no way the room can get there [16:13] because that would require some consciousness [16:15] in the room in addition to my consciousness. [16:18] And there is no such consciousness. [16:20] Anyway, that was one of many answers. [16:22] One of my favorites was this. [16:24] This was in a public debate. [16:25] A guy said to me, but suppose we ask you, [16:30] do you understand Chinese? [16:33] And suppose you say, yes, I understand Chinese. [16:36] Well? [16:37] Well, OK, let's try that and see how far we get. [16:42] I get a question that looks like this. [16:48] Now, this will be in a dialect of Chinese some of you [16:52] won't recognize. [16:55] Unknown to me, that symbol means, [16:58] do you understand Chinese? [17:00] I look up what I'm supposed to do. [17:02] And I give them back a symbol that's [17:03] in the same dialect of Chinese. [17:06] And it looks like that. [17:08] And that says, why do you guys ask me such dumb questions? [17:12] Can't you see that I understand Chinese? [17:18] I could go on with the other responses and objections, [17:23] but I think they're all equally feeble. [17:25] The bottom line is there's a logical truth. [17:29] And that is that the implemented computer program [17:32] is defined syntactically. [17:34] And that's not a weakness. [17:36] That's the power. [17:37] The power of the syntactical definition of computation [17:42] is you can implement it on electronic machines that [17:47] can perform literally millions of computations [17:51] in a very small amount of time. [17:53] I'm not sure I believe this, but it always [17:55] says it in the textbooks, that Deep Blue [17:58] can do 250 million computations in a second. [18:02] OK, I take their word for it. [18:04] So it's not a weakness of computers. [18:07] Now, another argument I sometimes got [18:09] was, well, in programs, we often have [18:12] a section called the semantics of natural understanding [18:15] programs. [18:16] And that's right. [18:17] But, of course, what they do is they put in more computer [18:20] implementation. [18:21] They put in more syntax. [18:23] Now, so far, so good. [18:25] And I think if that's all there was to say, [18:28] I've said all of that before. [18:30] But now I want to go on to something much more [18:32] interesting. [18:33] And here goes with that. [18:35] Now how we doing? [18:36] I'm not-- everybody seems to understand [18:39] there's going to be plenty of time for questions. [18:41] I insist on a good question period. [18:42] So let me take a drink of water, and we [18:44] go to the next step, which I think is more important. [18:51] A lot of people thought, well, look, [18:54] maybe the computer doesn't understand Chinese, [18:59] but all the same, it does information processing. [19:05] And it does, after all, do computation. [19:09] That's what we define the machine to do. [19:12] And I had to review a couple of books recently. [19:15] One book said that we live in a new age, [19:19] the age of information. [19:22] And in a wonderful outburst, the author [19:25] said everything is information. [19:29] Now that ought to worry us if everything is information. [19:33] And I read another book. [19:34] This was an optimistic book. [19:37] I reviewed-- this for "The New York Review of Books"-- [19:39] a less optimistic book by a guy who said computers are now [19:44] so smart they're almost as smart as we are. [19:48] And pretty soon, they'll be just as smart as we are. [19:52] And then I don't have to tell this audience the next step. [19:55] They'll be much smarter than we are. [19:58] And then look out because they might get [20:02] sick of being oppressed by us. [20:05] And they might simply rise up and overthrow us all. [20:10] And this, the author said modestly-- [20:11] I guess this is how you sell books-- [20:13] he said this may be the greatest challenge that humanity [20:17] has ever faced, the upcoming revolt [20:22] of super-smart computers. [20:25] Now, I want to say both of these claims are silly. [20:29] I mean, I'm speaking shorthand here. [20:31] There'll be plenty of chance to answer me. [20:34] And I want to say briefly why. [20:36] The notion of intelligence has two different senses. [20:42] It has an observer-independent sense [20:46] where it identifies something that is psychologically real. [20:50] So I am more intelligent than my dog Tarski. [20:56] Now, Tarski's pretty smart, I agree. [20:58] But overall, I'm smarter than Tarski. [21:02] I've had four dogs, by they way-- [21:03] Frege, Russell, Ludwig, and Tarski. [21:08] And Tarski, he's a Bernese mountain dog. [21:10] I'm sorry I didn't bring him along, [21:12] but he's too big for the car. [21:15] Now, he's very smart. [21:17] But he does have intelligence in the same sense that I do. [21:21] Only he happens to have somewhat less than I do. [21:24] Now, my computer is also intelligent. [21:28] And it also processes information. [21:31] But-- and this is the key point-- it's observer-relative. [21:36] The only sense in which the computer has intelligence [21:41] is not in an intrinsic, but it's in an observer-relative sense. [21:46] We can interpret its operations in such a way [21:49] that we can make-- now, watch this terminology-- we can make [21:52] epistemically objective claims of intelligence [21:57] even though the intelligence in question [21:59] is entirely in the eye of the beholder. [22:03] This was brought home forcefully to me [22:06] when I read in the newspapers that IBM had designed [22:10] a computer program which could beat the world's leading chess [22:15] player. [22:16] And in the same sense in which Kasparov beat Karpov [22:22] so we were told Deep Blue beat Kasparov. [22:24] Now that ought to worry us because for Karpov and Kasparov [22:28] to play chess, they both have to be conscious [22:31] that they're playing chess. [22:33] They both have to know such things [22:35] as I opened with pawn to king four, [22:38] and my queen is threatened on the left-hand side [22:41] of the board. [22:42] But now notice, Deep Blue knows none on that [22:46] because it doesn't know anything. [22:48] You can make epistemically objective claims [22:52] about Deep Blue. [22:53] It made such and such a move. [22:55] But the attributions of intelligent chess playing, [22:58] this move or that move, it's all observer-relative. [23:01] None of it is intrinsic. [23:04] In the intrinsic sense in which I have more intelligence [23:08] than my dog, my computer has zero intelligence-- absolutely [23:14] none at all. [23:15] It's a very complex electronic circuit [23:18] that we have designed to behave as if it were thinking, [23:22] as if it were intelligent. [23:24] But in the strict sense, in the observer-independent sense [23:28] in which you and I have intelligence, [23:30] there is zero intelligence in the computer. [23:34] It's all observer-relative. [23:36] And what goes for intelligence goes [23:38] for all of the key notions in cognitive science. [23:42] The notions of intelligence, memory, perception, [23:45] decision-making, rationality-- all those [23:48] have two different senses, a sense [23:51] where they identify psychologically real phenomena [23:55] of the sort that goes on in you and me and the sort [23:58] where they identify observer-relative phenomena. [24:02] But in the intrinsic sense in which [24:03] you and I have intelligence, the machinery we're talking about [24:07] has zero intelligence. [24:09] It's no question of its having more or less. [24:12] It's not in the same line of business. [24:14] All of the intelligence is in the eye of the beholder. [24:17] It's all observer-relative. [24:21] Now, you might say-- and I would say-- so, for most purposes, [24:24] it makes no difference at all. [24:26] I mean, if you can design a car that can drive itself, [24:29] who cares if it's conscious or not? [24:32] Who cares if it literally has any intelligence? [24:35] And I agree. [24:35] For most purposes, it doesn't matter. [24:38] For practical purposes, it doesn't matter whether or not [24:41] you have the observer-independent [24:43] or the observer-relative sense. [24:46] The only point where it matters, if you [24:49] think there's some psychological significance to the attribution [24:54] of intelligence to machinery which [24:56] has no intrinsic intelligence. [24:58] Now, notice the intelligence by which we-- [25:04] the mental processes by which we attribute intelligence [25:08] to the computer require consciousness. [25:12] So the attribution of observer-relativity [25:14] is done by conscious agents. [25:16] But the consciousness is not itself observer-relative. [25:21] The consciousness that creates the observer-relative phenomena [25:25] is not itself observer-relative. [25:28] But now let's get to the crunch line then. [25:30] If information is systematically ambiguous [25:35] between an intrinsic sense, in which you and I have [25:38] information, and an observer-relative sense, [25:41] in which the computer has information, [25:44] what about computation? [25:47] After all, computation, that must surely [25:50] be intrinsic to the computer. [25:53] That's what we designed and built the damn things to do, [25:56] was computation. [25:58] But, of course, the same distinction applies. [26:01] And I want to take a drink of water [26:03] and think about history for a moment. [26:05] When I first read Alan Turing's article, [26:13] it was called "Computing Machinery and Intelligence." [26:18] Now why didn't he call it "Computers and Intelligence"? [26:21] Well, you all know the answer. [26:22] In those days, "computer" meant "person who computes." [26:27] A computer is like a runner or a piano player. [26:31] It's some human who does the operation. [26:33] Nowadays nobody would think that because the word [26:36] has changed its meaning. [26:37] Or, rather, it's acquired the systematic ambiguity [26:41] between the observer-relative sense [26:46] and the observer-independent sense. [26:49] Now we think that a computer names [26:51] a type of machinery, not a human being who actually carries out [26:55] computation. [26:56] But the same distinction that we've been applying, [26:59] the same distinction that we discovered [27:01] in all these other cases, that applies [27:04] to computation in the literal or observer-independent [27:09] sense in which I will now do a simple computation. [27:14] I will do a computation using the addition function. [27:21] And here's how it goes. [27:22] It's not a very big deal. [27:24] One plus one equals two. [27:26] Now, the sense in which I carried out a computation [27:30] is absolutely intrinsic and observer-independent. [27:33] I don't care what anybody says about me. [27:35] If the experts say, well, you weren't really computing. [27:38] No, I was. [27:39] I consciously did a computation. [27:41] When my pocket calculator does the same operation, [27:46] the operation is entirely observer-relative. [27:49] Intrinsically all that goes on is a set of electronic state [27:53] transitions that we have designed so that we [27:56] can interpret computationally. [27:58] And, again, to repeat, for most purposes, it doesn't matter. [28:02] When it matters is when people say, [28:05] well, we've created this race of mechanical intelligences. [28:12] And they might rise up and overthrow us. [28:14] Or they attribute some other equally implausible [28:18] psychological interpretation to the machinery. [28:21] In commercial computers, the computation [28:24] is observer-relative. [28:26] Now notice, you all know that doesn't mean [28:29] it's epistemically subjective. [28:30] And I pay a lot of money so that Apple [28:33] will make a piece of machinery that will implement programs [28:37] that my earlier computers were not intelligent enough [28:40] to implement. [28:41] Notice the observer-relative attribution [28:43] of intelligence here. [28:45] So it's absolutely harmless unless you [28:47] think there's some psychological significance. [28:50] Now what is lacking, of course, in the machinery, which [28:54] we have in human beings which makes [28:56] the difference between the observer [28:59] relativity of the computation in the commercial computer [29:04] and the intrinsic or observer independent computation [29:08] that I have just performed on the blackboard, what's lacking [29:11] is consciousness. [29:13] All observer-relative phenomena are [29:17] created by human and animal consciousness. [29:20] But the human and animal consciousness [29:22] that creates them is not itself observer-relative. [29:26] So there's an intrinsic mental phenomena, the consciousness [29:30] of the agent, which creates the observer-relative phenomena, [29:35] or interprets the mechanical system in an observer [29:38] relative fashion. [29:40] But the consciousness that creates observer relativity [29:43] is not itself observer-relative. [29:45] It's intrinsic. [29:47] Now, I wanted to save plenty of time for discussion. [29:51] So let me catch my breath and then give [29:54] a kind of summary of the main thrust of what [29:57] I've been arguing. [29:59] And one of things I haven't emphasized [30:00] but I want to emphasize now, and that [30:02] is most of the apparatus, the conceptual apparatus, [30:07] we have for discussing these issues is totally obsolete. [30:12] The difference between the mental and the physical, [30:15] the difference between the social and the individual, [30:20] and the distinction between those features which [30:27] can be identified in an observer-relative fashion, [30:33] such as computation, and those which [30:36] can be identified in an observer-independent fashion, [30:39] such as computation. [30:40] We're confused by the vocabulary which doesn't make [30:44] the matters sufficiently clear. [30:46] And I'm going to end this discussion [30:48] by going through some of the elements of the vocabulary. [30:51] Now, let me have a drink of water and catch my breath. [31:00] Let's start with that old question, [31:03] could a machine think? [31:05] Well, I said the vocabulary was obsolete. [31:09] And the vocabulary of humans and machines [31:11] is already obsolete because if by machine is [31:16] meant a physical system capable of performing [31:19] certain functions, then we're all machines. [31:22] I'm a machine. [31:23] You're a machine. [31:25] And my guess is only machines could think. [31:29] Why? [31:30] Well that's the next step. [31:32] Thinking is a biological process created in the brain [31:37] by certain quite complex, but insufficiently understood [31:43] neurobiological processes. [31:45] So in order to think, you've got to have a brain, [31:49] or you've got to have something with equivalent causal powers [31:52] to the brain. [31:53] We might figure out a way to do it in some other medium. [31:55] We don't know enough about how the brain does it. [31:58] So we don't know how to create it artificially. [32:01] So could a machine think? [32:04] Human beings are machines. [32:06] Yes, but could you make an artificial machine [32:08] that could think? [32:09] Why not? [32:10] It's like an artificial heart. [32:12] The question, can you build an artificial brain that [32:14] can think, is like the question, can you [32:16] build an artificial heart that pumps blood. [32:19] We know how the heart does it, so we [32:22] know how to do it artificially. [32:24] We don't know how the brain does it, so we have no idea. [32:27] Let me repeat this. [32:27] We have no idea how to create a thinking [32:31] machine because we don't know how the brain does it. [32:34] All we can do is a simulation using [32:36] some sort of formal system. [32:38] But that's not the real thing. [32:39] You don't create thinking that way, [32:41] whereas the artificial heart really does pump blood. [32:44] So we had two questions. [32:46] Could a machine think? [32:48] And could an artificially-made machine think? [32:52] Answer the question one is obviously yes. [32:54] Answer to question two is, we don't know yet, [32:57] but there's no obstacle in principle. [33:00] Does everybody see that? [33:01] Building an artificial brain is like building [33:04] an artificial heart. [33:05] The only thing is no one has begun to try it. [33:09] They haven't begun to try it because they have no idea how [33:12] the actual brain does it. [33:13] So they don't know how to imitate actual brains. [33:16] Well, OK, but could you build an artificial brain that [33:21] could think out of some completely different materials, [33:24] out of something that had nothing [33:25] to do with nucleo-proteins, had nothing to [33:28] with neurons and neurotransmitters [33:30] and all the rest of it. [33:32] And the answer is, again, we don't know. [33:34] That seems to me an open question. [33:37] If we knew how the brain did it, we [33:39] might be able to define-- I mean, [33:42] be able to design machines that could do it using [33:45] some completely different biochemistry in a way [33:49] that the artificial heart doesn't use [33:52] muscle tissue to pump blood. [33:54] You don't need muscle tissue to pump blood. [33:57] And maybe you don't need brain tissue to create consciousness. [34:00] We just are ignorant. [34:01] But notice there's no obstacle in principle. [34:04] The problem is no one has begun to think about how you would [34:08] build a thinking machine, how you'd build a thinking [34:13] machine out of some material other than neurons [34:16] because they haven't begun to think about how we might [34:21] duplicate and not merely simulate [34:23] what the brain actually does. [34:26] So the question, could a machine think, [34:28] could an artificial machine think, [34:30] could an artificial machine made out [34:32] of some completely different materials, [34:34] could those machines think? [34:37] And now the next question is the obvious one. [34:39] Well, how about a computer? [34:41] Could a computer think? [34:43] Now, you have to be careful here. [34:45] Because if a computer is defined as anything that can carry out [34:49] computations, well, I just did. [34:53] This is a computation. [34:55] So I'm a computer. [34:57] And so are all of you. [34:59] Any conscious agent capable of carrying out [35:02] that simple computation is capable, is both a computer [35:07] and capable of thinking. [35:09] So my guess is-- and I didn't have a chance [35:11] to develop this idea-- is that not only can computers [35:15] think-- you and me-- but my guess [35:18] is that anything capable of thinking [35:20] would have to be capable of carrying out [35:22] simple computations. [35:24] But now what is the status of computation? [35:28] Well, the key element here is the one I've already mentioned. [35:33] Computation has two senses, an observer-independent sense [35:36] and an observer-relative sense. [35:39] In the observer-relative sense, anything [35:43] is a computer if you can ascribe a computational interpretation. [35:47] Watch. [35:47] I'll show you a very simple computer. [35:50] That computer just computed a well-known function. [35:54] s equals one-half gt squared. [35:57] And if you had a good-enough watch, [35:59] you could actually time and figure out [36:02] how far the damn thing fell. [36:05] Everybody sees. [36:05] It's elementary mathematics. [36:07] So if this is a computer, then anything [36:10] is a computer because being a computer [36:13] in the observer-relative sense is not [36:16] an intrinsic feature of an object, [36:18] but a feature of our interpretation of the physics [36:21] of the phenomenon. [36:22] In the old Chinese room days, when [36:23] I had to debate these guys, at one point, I'd take my pen, [36:27] slam it on a table, and say that is a digital computer. [36:32] It just happens to have a boring computer program. [36:36] The program says stay there. [36:38] The point is nobody ever called me on this [36:42] because it's obviously right. [36:43] It satisfies a textbook definition. [36:46] You know, in the early days, they [36:48] tried to snow me with a whole lot of technical razzmatazz. [36:51] "You've left out the distinction between the virtual machine [36:54] and the non-virtual machine" or "you've [36:57] left out the transducers." [36:58] You see, I didn't know what the hell a transducer was, [37:00] a virtual machine. [37:01] But it takes about five minutes to learn those things. [37:04] Anyway, so now we get to the crucial question in this. [37:07] If computers can think, man-made computers [37:10] can think, machines can think, what about computation? [37:14] Does computation name a machine, a thinking process? [37:20] That is, is computation, as defined by Alan Turing, [37:24] is that itself sufficient for thinking? [37:26] And you now know the answer to that. [37:28] In the observer-relative sense, the answer is no. [37:32] Computation is not a fact of nature. [37:35] It's a fact of our interpretation. [37:37] And insofar as we can create artificial machines that [37:42] carry out computations, the computation by itself [37:46] is never going to be sufficient for thinking [37:48] or any other cognitive process because the computation is [37:52] defined purely formally or syntactically. [37:56] Turing machines are not to be found in nature. [37:59] They're to be found in our interpretations of nature. [38:03] Now, let me add, a lot of people think, ah, [38:05] this debate has something to do with technology [38:07] or there'll be advances in technology. [38:10] I think that technology's wonderful. [38:12] And I welcome it. [38:13] And I see no limits to the possibilities of technology. [38:17] My aim is this talk is simply to get across, [38:20] you shouldn't misunderstand the philosophical, psychological, [38:24] and, indeed, scientific implication of the technology. [38:28] Thank you very much. [38:29] [APPLAUSE] [38:41] JOHN BRACAGLIA: Thank you, John. [38:42] JOHN SEARLE: I'm sorry I talk so fast, [38:44] but I want to leave plenty of time for questions. [38:46] JOHN BRACAGLIA: We'll start with one question from Mr. Ray [38:49] Kurzweil. [38:49] RAY KURZWEIL: Is this on? [38:50] [INTERPOSING VOICES] [38:55] RAY KURZWEIL: Well, thanks, John. [38:58] I'm one of those guys you've been debating this issue for 18 [39:02] years, I think. [39:05] And I would praise the Chinese room for its longevity [39:10] because it does really get at the apparent absurdity [39:14] that some deterministic process like computation [39:17] could possibly be responsible for something like thinking. [39:20] And you point out the distinction [39:22] of thinking between its effects and the subjective states, [39:29] which is a synonym for consciousness. [39:32] So I quoted you here in my book "Singularity is Near," [39:41] at the equivalence of neurons and even brains with machines. [39:46] So then I took your argument why a machine and a computer [39:51] could not truly understand what it's doing and simply [39:56] substituted human brain for computers, [39:59] since you said they were equivalent, [40:01] and neurotransmitter concentrations and related [40:04] mechanisms for formal symbols, since basically [40:09] neurotransmitter concentrations, it's [40:10] just a mechanistic concept. [40:14] And so you wrote, with those substitutions, [40:17] the human brain succeeds by manipulating [40:20] neurotransmitter concentrations and other related mechanisms. [40:24] The neurotransmitter concentrations [40:26] and related mechanisms themselves [40:28] are quite meaningless. [40:30] They only have the meaning we have attached to them. [40:32] The human brain knows nothing of this. [40:34] It just shuffles the neurotransmitter concentrations [40:37] and related mechanisms. [40:39] Therefore, the human brain cannot have true understanding. [40:43] So-- [40:44] [LAUGHTER] [40:46] JOHN SEARLE: There's something interesting variations, again, [40:49] on my original. [40:50] RAY KURZWEIL: But the point I'd like to make, [40:54] and that I'd be interested in your addressing, [40:57] is the nature of consciousness because, I mean, [41:00] you said today, and you wrote, the essential thing [41:04] is to recognize that consciousness [41:06] is a biological processes like digestion, lactation, [41:10] photosynthesis, or mitosis. [41:12] We know that brains cause consciousness [41:14] with specific biological mechanisms. [41:18] But how do we know that a brain is conscious? [41:21] How do you know that I'm conscious? [41:22] And how do you-- [41:24] JOHN SEARLE: [INAUDIBLE] [41:25] RAY KURZWEIL: And how do we know if a computer was conscious? [41:27] We don't have a computer today that [41:29] seems conscious, that's convincing in its responses. [41:32] But my prediction is we will. [41:35] We can argue about the time frame. [41:37] And when we do, how do we know if it's conscious of [41:40] it just seems conscious? [41:43] How do we measure that? [41:45] JOHN SEARLE: Well, there are two questions here. [41:47] One is, if you do a substitution of words that I didn't use [41:52] and the words I did use, can you get these observed results? [41:55] And, of course, you can do that. [41:57] That's a well-known technique of politicians. [42:01] But that wasn't the claim. [42:02] What is the difference between the computer and the brain? [42:06] In one sentence, the brain is a causal mechanism [42:10] that produces consciousness by a certain rather complex [42:16] and still imperfectly understood neurobiological processes. [42:19] But those are quite specific to a certain electrochemistry. [42:23] We just don't know the details. [42:24] But we don't know if you mess around in the synaptic cleft, [42:29] you're going to get weird effects. [42:31] How does cocaine work? [42:33] Well, it isn't because it's got a peculiar computational [42:36] capacity. [42:37] Because it messes with the capacity [42:39] of the postsynaptic receptors to reabsorb quite specific [42:43] neurotransmitters, norepinephrine-- [42:45] what are the other two? [42:46] God, I'm flunking the exam here. [42:49] Dopamine. [42:50] Gaba is the third. [42:51] Anyway, the brain, like the stomach or any other organ, [42:57] is a specific causal mechanism. [43:01] And it functions on specific biochemical principles. [43:04] The problem of the computer is it [43:06] has nothing to do with the specifics [43:09] of the implementation. [43:11] Any implementation will do provided [43:13] it's sufficient to carry out the steps in the program. [43:16] Programs are purely formal or syntactical. [43:19] The brain is not. [43:21] The brain is a specific biological organ [43:23] that operates on specific principles. [43:26] And to create a conscious machine, [43:28] we've got to know how to duplicate the causal powers [43:30] of those principles. [43:32] Now, the computer doesn't in that way [43:36] work as a causal mechanism producing higher level [43:40] features. [43:41] Rather, computation names an abstract mathematical process [43:46] that we have found ways to implement in specific hardware. [43:50] But the hardware is not essential to the computation. [43:54] Any system that can carry out the computation [43:57] will be equivalent. [43:59] Now, the second question is about how do [44:01] you know about consciousness. [44:02] Well, think about real life. [44:04] How do I know my dog Tarski is conscious [44:08] and this thing here, my smartphone, is not conscious? [44:13] I don't have any doubts about either one. [44:15] I can tell that Tarski is conscious not [44:19] on behavioristic grounds. [44:20] People say, well, it's because he behaves like a human being. [44:22] He doesn't. [44:23] See, human beings I know when they [44:25] see me don't rush up and lick my hands and wag their tails. [44:29] They just don't. [44:30] My friends don't do that. [44:32] But Tarski does. [44:33] I can see that Tarski is conscious [44:35] because he's got a machinery that's [44:37] relatively similar to my own. [44:40] Those are his eyes. [44:41] These are his ears. [44:41] This is his skin. [44:43] He has mechanisms that mediate the input stimuli to the output [44:50] behavior that are relatively similar to human mechanisms. [44:53] This is why I'm completely confident that Tarski's [44:56] conscious. [44:57] I don't know anything about fleas and termites. [45:01] You know, your typical termite's got 100,000 neurons. [45:04] Is that enough? [45:05] Well, I lose 100,000 on a big weekend. [45:08] So I don't know if that's enough for consciousness. [45:11] But that's a factual question. [45:12] I'll leave that to the experts. [45:14] But as far as human beings are concerned [45:17] there isn't any question that everybody in this room [45:19] is conscious. [45:20] I mean, maybe that guy over there is falling asleep, [45:22] but there's no question about what the general-- it's not [45:27] even a theory that I hold. [45:29] It's a background presupposition. [45:30] The way I assume that the floor is solid, [45:32] I simply take it for granted that everybody's conscious. [45:35] If forced to justify it, I could. [45:37] Now, there's always a problem about the details [45:41] of other minds. [45:42] Of course, I know you're conscious. [45:45] But are you suffering the angst of post-industrial man [45:49] under late capitalism? [45:50] Well, I have a lot of friends who claim they do. [45:53] And they think I'm philistine because I don't. [45:55] But that's tougher. [45:56] We'd have to have a conversation about that. [45:58] But for consciousness, it's not a real problem [46:01] in a real-life case. [46:03] AUDIENCE: So you've said that we haven't [46:05] begun to understand how brains work or build [46:08] comparable machines. [46:09] But imagine in the future we do. [46:12] So we can run a simulation, as you put it, of a brain. [46:16] And then we interface it with reality [46:18] through motor output, sensory input. [46:21] What's the difference between that [46:23] and a brain, which you say you know [46:25] is producing consciousness? [46:26] In [46:27] JOHN SEARLE: In some cases, there's no difference at all. [46:30] And the difference doesn't matter. [46:32] If you've got a machine-- I hope you [46:34] guys are, in fact, building it because the newspapers [46:36] say you are. [46:39] If you've got a program that'll drive [46:41] my car without a conscious driver, that's great. [46:45] I think that's wonderful. [46:46] The question is not, what can the technology do? [46:49] My daddy was an electrical engineer for AT&T. [46:52] And his biggest disappointment was [46:53] I decided to be a philosopher, for God's sake, [46:56] instead of going to Bell Labs and MIT as he had hoped. [47:02] So I have no problem with the success of the technology. [47:06] The question is, what does it mean? [47:08] Of course, if you've got a machine [47:10] that can drive a car as well as I, [47:13] or probably better than I can, then so much [47:16] the better for the machinery. [47:17] The question is, what is the philosophical psychological [47:21] scientific significance of that? [47:23] And if you think, well, that means you've created [47:25] consciousness, you have not. [47:27] You have to have more to create consciousness. [47:29] And for a whole lot of things, consciousness [47:31] matters desperately. [47:33] In this case of this book that I reviewed, [47:35] where the guy said, well, they got machines [47:37] that are going to rise up and overthrow us all, [47:40] it's not a serious possibility because the machines [47:43] have no consciousness. [47:44] They have no conscious psychological state. [47:47] It's about like saying the shoes might get up out of the closet [47:51] and walk all over us. [47:52] After all, we've been walking on them for centuries, [47:55] why don't they strike back? [47:56] It is not a real-life worry. [47:58] Yeah? [47:59] AUDIENCE: The difference that I'm interested in-- sorry, [48:01] the similarity I'm interested in is not necessarily [48:04] the output or the outcome of the system, [48:06] but rather, that is, it has the internal causal similarity [48:11] to the brain that you mentioned. [48:13] JOHN SEARLE: Yeah, that's a factual question. [48:16] The question is, to what extent are the processes that [48:20] go on in the computer isomorphic to processes [48:24] that go on in the brain? [48:25] As far as we know, not very much. [48:30] I mean, the chess-playing programs [48:32] were a good example of this. [48:33] In the early days of AI, they tried [48:36] to interview great chess players and find out what their thought [48:40] processes were and get them to try [48:42] to duplicate that on computers. [48:44] Well, we now know how Deep Blue worked. [48:47] Deep Blue can calculate 250 million chess positions [48:50] in one second. [48:52] See, chess is a trivial game from a games theoretical point [48:55] of view because you have perfect information. [48:58] And you have a finite number of possibilities. [49:00] So there are x number of possibilities [49:01] of responding to a move and x number [49:03] of possibilities for that move. [49:05] It's interesting to us because of the exponential problem. [49:09] And it's very hard to program computers that [49:13] can go very many steps in the exponents, but IBM did. [49:18] It's of no psychological interest. [49:20] And to their credit, the people in AI [49:22] did not claim it as a great victory for-- at least the ones [49:25] I know didn't claim it as a victory for AI [49:27] because they could see it had nothing [49:28] to do with human cognition. [49:30] So my guess is it's an interesting philosophical [49:34] question-- or psychological question-- to what extent [49:38] the actual processes in the brain [49:41] mirror a computational simulation. [49:43] And, of course, to some respect, they do. [49:45] That's why computational simulations [49:47] are interesting in all sorts of fields [49:49] and not just in psychology, because you [49:51] can simulate all sorts of processes that are going on. [49:54] But that's not strong AI. [49:56] Strong AI says the simulation isn't just a simulation. [50:00] It's a duplication. [50:01] And that we can refute. [50:03] AUDIENCE: Could you prove to me that you understand English? [50:06] JOHN SEARLE: Yeah, I wouldn't bother. [50:09] (SPEAKING WITH BRITISH ACCENT) When I was in Oxford, [50:11] many people doubted that I did. [50:17] I happened to be in a rather snobbish college called [50:20] Christ Church. [50:22] And, of course, I don't speak English. [50:24] I never pretended to. [50:25] I speak a dialect of American, which makes many English people [50:29] shudder at the thought. [50:36] AUDIENCE: So you've said you understand English, [50:39] but how do I know you're not just a computer program? [50:42] JOHN SEARLE: Well, it's the same question as Ray's. [50:45] And the answer is all sorts of ways. [50:48] You know, if it got to a crunch, you might ask me. [50:53] Now I might give a dishonest answer. [50:55] Or I might give an honest answer. [50:56] But there's one route that you don't want to go. [50:59] And that's the epistemic route. [51:01] The epistemic route says, well, you [51:04] have as much evidence that the computer is conscious [51:07] as that we have that you are conscious. [51:10] No, not really. [51:11] I mean, I could go into some detail [51:13] about what it is about people's physical structure that [51:18] make them capable of producing consciousness. [51:20] You don't have to have a fancy theory. [51:22] I don't need a fancy theory of neurobiology [51:25] to say those are your eyes. [51:26] You spoke through your mouth. [51:30] The question was an expression of a conscious intention [51:34] to ask a question. [51:35] Believe me, if you are a locally produced machine, [51:40] Google is further along than I thought. [51:44] But clearly, you're not. [51:49] JOHN BRACAGLIA: We're going to take a question from the Dory. [51:54] JOHN SEARLE: Is he next? [51:55] JOHN BRACAGLIA: We had some people-- [51:57] AUDIENCE: Almost. [51:58] JOHN BRACAGLIA: We had some people [51:59] submit questions ahead of time. [52:01] JOHN SEARLE: OK. [52:02] JOHN BRACAGLIA: So we're going to read those as well. [52:06] JOHN SEARLE: OK. [52:06] All right. [52:07] Right. [52:08] JOHN BRACAGLIA: So the first question from the Dory [52:10] is, what is the definition of consciousness you've been using [52:12] for the duration of this talk? [52:14] JOHN SEARLE: OK. [52:14] Here goes. [52:14] JOHN BRACAGLIA: Please be as specific as possible. [52:15] JOHN SEARLE: It is typically said that consciousness [52:18] is hard to define. [52:19] I think it's rather easy to define. [52:22] We don't have a scientific definition [52:24] because we don't have a scientific theory. [52:26] The commonsense definition of any term [52:29] will identify the target of the investigation. [52:32] Water is a clear, colorless, tasteless liquid. [52:35] And it comes in bottles like this. [52:36] That's the commonsense definition. [52:38] You do science and you discover it's H2O. [52:41] Well, with consciousness, we're in [52:42] the clear, colorless, liquid, tasteless sense. [52:46] But here it is. [52:47] Consciousness consists of all those states [52:50] of feeling or sentience or awareness [52:53] that begin in the morning when you awake [52:55] from a dreamless sleep. [52:57] And they go on all day long until you fall asleep [53:00] again or otherwise become, as they would say, unconscious. [53:04] On this definition, dreams are a form of consciousness. [53:08] The secret, the essence, of consciousness [53:11] is that for any conscious state, there's [53:13] something it feels like to be in that conscious state. [53:17] Now, for that reason, consciousness always [53:18] has a subjective ontology. [53:21] Remember, I gave you that subjective-objective bit. [53:25] It always has a subjective ontology. [53:27] That's the working definition of consciousness. [53:30] And that's the one that's actually [53:31] used by neurobiological investigators trying to figure [53:35] out how the brain does it. [53:36] That's what you're trying to figure out. [53:38] How does the brain produce that? [53:39] How does it exist in the brain? [53:41] How does it function? [53:43] AUDIENCE: I'd like to propose a stronger [53:45] bound on your observation that we do not [53:48] know how to build a thinking machine today. [53:51] Even if we knew how to build it, because, I mean, [53:54] our thinking machine was built by the process of evolution, [53:57] I'd like to propose-- well, what do you [53:59] think about stating that, actually, we [54:02] may not have the time? [54:04] And that it actually may not matter. [54:06] The reason we may not have the time is the probabilities that [54:10] need to happen, like the asteroid falling and wiping [54:12] the dinosaurs and whatnot, may not happen in the universe [54:15] that we live been. [54:16] But if you subscribe to the parallel universes theory, [54:20] then there is some artificial consciousness somewhere else. [54:23] JOHN SEARLE: Yeah. [54:24] OK, about we may not have the time, well, I'm in a hurry. [54:30] But I think we ought to try as hard as we can. [54:32] It's true. [54:33] Maybe some things are beyond our capacity [54:35] to solve in the life of human beings on Earth. [54:38] But let's get busy and try. [54:40] There was a period when people said, well, we'll never [54:43] really understand life. [54:45] And while we don't fully understand it, [54:47] but we're pretty far along. [54:48] I mean, the old debate between the mechanists [54:50] and the vitalists, that doesn't make any sense to us anymore. [54:53] So we made a lot of progress. [54:55] There was another half to your question. [54:57] AUDIENCE: It may not matter because all universe-- [54:59] JOHN SEARLE: Oh, yeah. [54:59] Maybe conscious doesn't matter. [55:00] Well, it's where I live. [55:02] It matters to me. [55:04] AUDIENCE: Philosophically speaking. [55:06] JOHN SEARLE: Yeah, but the point is [55:07] there are a lot of things that may or may not [55:10] matter which are desperately important to us-- [55:12] democracy and sex and literature and good food [55:15] and all that kind of stuff. [55:17] Maybe it doesn't matter to somebody, [55:18] but all those things matter to me in varying degrees. [55:22] AUDIENCE: Your artificial heart analogy that you mentioned. [55:24] I think you included the idea that it's possible, [55:26] just like with the artificial heart, [55:28] that we use different materials and different approaches [55:30] to simulate a heart and, in some ways, [55:34] go beyond just-- come closer to duplication, [55:37] that we might, in theory, be able to do the same thing [55:40] with an artificial brain. [55:41] I'm wondering if you think it's possible [55:43] that going down the path just trying [55:46] to do a simulation of a brain accidentally creates [55:50] a consciousness or accidentally creates duplication, [55:52] even if we don't intend to do it with exact same means [55:55] as a brain is made. [55:57] JOHN SEARLE: I would say to believe in that, you [55:59] have to believe in miracles. [56:00] You have to-- now think about it. [56:03] We can do computer simulations of just about anything [56:07] you can describe precisely. [56:08] You do a computer simulation of digestion. [56:12] And you could get a computer model [56:14] that does a perfect model of digesting pizza. [56:17] For all I know, maybe somebody in this building has done it. [56:20] But once you've done that, you don't rush out and buy a pizza [56:23] and stuff it in the computer because it [56:25] isn't going to digest a pizza. [56:26] What it gives you is a picture or a model [56:29] or a mathematical diagram. [56:31] And I have no objection to that. [56:33] But if my life depended on figuring out [56:36] how the brain produces consciousness, [56:38] I would use the computer the way you [56:40] use a computer in any branch of biology. [56:42] It's very useful for figuring out [56:45] the implications of your axioms, for figuring out [56:47] the possible experiments that you could design. [56:50] But somehow or other that the idea [56:52] that the computer simulation of cognitive behavior [56:56] might provide the key to the biochemistry, [56:59] well, it's not out of the question, [57:01] it's just not plausible. [57:02] JOHN BRACAGLIA: Humans are easily fooled and frequently [57:04] overestimate the intelligence of machines. [57:06] Can you propose a better test of general intelligence [57:09] than the Turing test, one that is less likely to relate [57:12] false positives? [57:14] JOHN SEARLE: Well, you all know my answer [57:16] to that is the first step is to distinguish [57:18] between genuine intrinsic observer-independent [57:21] intelligence and observer-relative intelligence. [57:26] And observer-relative intelligence [57:27] is always in the eye of the beholder. [57:30] And anything will have the intelligence [57:32] that you're able to attribute to it. [57:34] I just attributed a great deal of intelligence to this object [57:38] because it can compute a function, [57:40] s equals one-half squared. [57:42] Now this object has prodigious intelligence [57:45] because it discriminates one hair from-- I [57:48] won't demonstrate it, but in any-- take [57:50] my word for it that it does, even in a head that's [57:53] sparse with hair. [57:54] So because intelligence is observer-relative, [57:57] you have to tell me the criteria by which we're [58:00] going to judge it. [58:01] And the problem with the Turing test-- [58:03] well, it's got all sorts of problems, [58:06] but the basic problem is that both the input and the output [58:12] are what they are only relative to our interpretation. [58:15] You have to interpret this as a question. [58:17] And you have to interpret that as an answer. [58:20] One bottom line of my whole discussion [58:23] today is that the Turing test fails. [58:24] It doesn't give you a test of intelligence. [58:27] AUDIENCE: So you seem to take it as an article of faith [58:30] that we are conscious, that your dog is conscious, [58:33] and that that consciousness comes [58:35] from biological material, the likes of which we can't really [58:39] understand. [58:40] But forgive me for saying this, that [58:42] makes you sound like an intelligent design theorist who [58:45] says that because evolution and everything [58:50] in this creative universe that exists [58:53] is so complex, that it couldn't have [58:55] evolved from inert material. [58:58] So somewhere between an amoeba and your dog, [59:00] there must not be consciousness. [59:02] And I'm not sure where you would draw that line. [59:05] And so if consciousness in human beings [59:08] is emergent, or even in your dog at some point [59:10] in the evolutionary scale, why couldn't it [59:13] emerge from a computation system that's [59:16] sufficiently distributed, networked, and has the ability [59:19] to perform many calculations and maybe is even hooked [59:23] into biologic systems? [59:25] JOHN SEARLE: Well, about could it emerge, miracles are always [59:28] possible. [59:30] How do you know that you don't have [59:32] chemical processes that will turn this [59:35] into a conscious comb? [59:37] How do I know that? [59:38] Well, it's not a serious possibility. [59:40] I mean, the mechanisms by which consciousness [59:43] is created in the brain are quite specific. [59:46] And remember, this is the key point. [59:48] Any system that creates consciousness [59:51] has to duplicate those causal powers. [59:53] That's like saying, you don't have to have feathers in order [59:57] to have a flying machine, but you have to duplicate [60:00] and not merely simulate the causal power of the bird [60:04] to overcome the force of gravity in the Earth's atmosphere. [60:07] And that's what airplanes do. [60:08] They duplicate causal powers. [60:10] They use the same principle, Bernoulli's principle, [60:12] to overcome the force of gravity. [60:15] But the idea that somehow or other you might do it [60:18] just by doing a simulation of certain formal structures [60:22] of input-output mechanisms, of input-output functions, [60:26] well, miracles are always possible. [60:28] But it doesn't seem likely. [60:29] That's not the way evolution works. [60:31] AUDIENCE: But machines can improve themselves. [60:33] And you're making the case for why an amoeba could never [60:35] develop into your dog over a sufficiently long period [60:38] of time and have consciousness. [60:40] JOHN SEARLE: No, I didn't make that case. [60:40] No, I didn't make that case. [60:41] [INTERPOSING VOICES] [60:44] JOHN SEARLE: Amoeba don't have it. [60:46] AUDIENCE: You're refuting that consciousness [60:48] could emerge from a sufficiently complex computation system. [60:53] JOHN SEARLE: Complexity is always observer-relative. [60:57] If you talk about complexity, you [60:58] have to talk about the metric. [61:02] What is the metric by which you calculate complexity? [61:04] I think complexity is probably irrelevant. [61:06] It might turn out that the mechanism is simple. [61:10] There's nothing in my account that [61:12] says a computer could never become conscious. [61:16] Of course, we're all conscious computers, as I said. [61:19] And the point about the amoeba is not [61:21] that amoebas can't evolve into much more complex organisms. [61:25] Maybe that's what happened. [61:27] But the amoeba as it stands-- a single-celled organism-- [61:30] that doesn't have enough machinery to duplicate [61:33] the causal powers of the brain. [61:35] I am not doing a science fiction project to say, well, [61:41] there can never be an artificially created [61:43] consciousness by people busy designing computer programs. [61:47] Of course, I'm not saying that's logically impossible. [61:50] I'm just saying it's not an intelligent project. [61:53] If you're thinking about your life depends [61:55] on building a machine that creates consciousness, [61:57] you don't sit down your console and start programming things [62:00] in some programming language. [62:02] It's the wrong way to go about it. [62:05] AUDIENCE: If we gave you a disassembly of Google Translate [62:09] and had you implement the Chinese room experiment, [62:12] either it would take you thousands [62:14] of years to run all the assembly instructions on pen and paper, [62:19] or else you'd end up decompiling it into English [62:22] and heavily optimizing it in that form. [62:26] And in the process, you'd come to learn a lot [62:28] about the relationships between the different variables [62:30] and subroutines. [62:31] So who's to say that an understanding of Chinese [62:35] wouldn't emerge from that? [62:37] JOHN SEARLE: Well, OK, I love this kind of question. [62:40] All right. [62:40] Now, let me say, of course, when I did the original thought [62:43] experiment, anybody will point out to you if you actually [62:47] were carrying out the steps in a program [62:49] for answering questions in Chinese, [62:51] well, we'd be around for several million years. [62:54] OK, I take their word for it. [62:56] I'm not a programmer, but I assume [62:58] it would take an enormous amount of time. [63:00] But the point of the argument is not the example. [63:08] The example is designed to illustrate [63:11] the point of the argument. [63:12] The point of the argument can be given [63:14] in the following derivation. [63:16] Programs are formal or syntactical. [63:19] That's axiom number one. [63:22] That's all there is to the program. [63:24] To put it slightly more pretentiously, [63:26] the notion same implemented program [63:29] defines an equivalence class specified entirely [63:32] formally or syntactically. [63:34] But minds have a semantics, and-- and this [63:37] is the whole point of the example-- the syntax by itself [63:41] is not sufficient for the semantics. [63:43] That's the point of the example. [63:44] The Chinese room is designed to illustrate axiom three, that [63:49] just having the steps in the program is not by itself [63:54] sufficient for a semantics. [63:57] And minds have a semantics. [63:58] Now, it follows from those that if the computer is defined [64:02] in terms of its program operations, [64:05] syntactical operations, then the program operations, [64:08] the computer operations by themselves [64:09] are never sufficient for understanding [64:11] because they lack a semantics. [64:13] But, of course, I'm not saying, well, you [64:16] could not build a machine that was both a computer [64:19] and had semantics. [64:20] We are such machines. [64:21] AUDIENCE: You couldn't verify experimentally [64:23] what the difference might be between semantics [64:25] and what would emerge from thousands [64:28] of years of experience with a given syntactical program. [64:33] JOHN SEARLE: I think you can-- I don't inherit this. [64:35] He does. [64:36] I think you don't want to go the epistemic route. [64:39] You don't want to say, well, look [64:41] you can't tell the difference between the thinking machine [64:45] and the non-thinking machine. [64:48] The reason that's the wrong route to go [64:50] is we now have overwhelming evidence [64:54] of what sorts of mechanisms produce what [64:57] sorts of cognition. [64:58] When I first got interested in the brain, [64:59] I went out and bought all the textbooks. [65:01] By the way, if you want to learn a subject, [65:03] that's the way to do it. [65:04] Go buy all the freshman textbooks [65:05] because they're easy to understand. [65:08] One of these textbooks, it said cats have different color [65:13] vision from ours. [65:15] Their visual experiences are different from ours. [65:18] And I thought, christ, have these guys been cats? [65:21] Have the other cats mind problem? [65:23] Do they know what it's like to be a cat? [65:26] And the answer is, of course, they know completely [65:29] what's the cat's color vision is because they can [65:31] look at the color receptors. [65:32] And cats do have different color vision from ours [65:35] because they have different color receptors. [65:37] I forget the difference. [65:38] You can look them up in any textbook. [65:40] But if in real life we're completely [65:42] confident that my dog can hear parts of the auditory spectrum [65:47] that I can't hear. [65:47] He can hear the higher frequencies that I can't hear. [65:51] And cats have a different color vision from mine [65:54] because we can see what the apparatus is. [65:57] We got another question? [65:58] You're on. [65:58] JOHN BRACAGLIA: This will be our final question. [66:00] JOHN SEARLE: OK. [66:01] I'm prepared to go all afternoon. [66:02] I love this kind of crap. [66:04] AUDIENCE: So at the beginning of your talk, [66:06] you mentioned an anecdote about neuroscientists [66:08] not being interested in consciousness. [66:10] And, of course, by this time, a number of neuroscientists [66:13] have studied it. [66:13] And so they'll present stimuli that [66:16] are near the threshold of perceptibility [66:18] and measure the brain responses when it's above or below. [66:21] What do you think about that? [66:22] Is that on the right track? [66:23] What would you do differently? [66:24] JOHN SEARLE: No, I think one of the best things that's [66:27] happened in my lifetime-- it's getting a rather [66:29] long lifetime-- is that there is now [66:31] a thriving industry of neuroscientific investigations [66:36] of consciousness. [66:37] That's how we will get the answer. [66:40] When I first got interested in this, [66:42] I told you I went over to UCSF and told those guys get busy. [66:46] The last thing they wanted to hear [66:47] was being nagged by some philosopher, I can tell you. [66:50] But one guy said to me-- famous neuroscientist [66:54] said-- in my discipline, it's OK to be [66:58] interested in consciousness, but get tenure first. [67:02] Get tenure first. [67:05] Now, there has been a change. [67:07] I don't take credit for the change, [67:09] but I've certainly been urging it. [67:10] You can now get tenure by working on consciousness. [67:13] Now, neuroscience has changed, that now there's [67:16] a thriving industry in neuroscience [67:18] of people who are actually trying to figure out [67:20] how the brain does it. [67:22] And when they figure that out-- and I don't see any obstacle [67:25] to figuring that out-- it will be [67:27] an enormous intellectual breakthrough, [67:29] when we figure out how exactly does the brain [67:31] create consciousness. [67:33] AUDIENCE: But in particular, that [67:34] approach they're using now-- I use the example of presenting [67:37] stimuli that are near the threshold of perceptibility [67:39] and looking for neural correlates, [67:41] do you think that's going to be fruitful? [67:43] What particular questions would you ask to find out? [67:46] JOHN SEARLE: I happened to be interested in this crap. [67:48] And if you're interested in my views, [67:50] I published an article in the "Annual Review of Neuroscience" [67:54] with a title "Consciousness." [67:56] It's easy to remember. [67:57] You can find it on the web. [67:58] And what I said is, there are two main lines [68:01] of research going on today. [68:03] There are guys who take what I call the building block [68:06] approach. [68:06] And they try to find the neuronal correlate [68:09] of particular experiences. [68:10] You see a red object. [68:12] Or you hear the sound of middle C. [68:14] What's the correlate in the brain? [68:15] And the idea they have is if you can figure out [68:18] how the brain creates the experience of red, [68:21] you've cracked the whole problem. [68:23] Because it's like DNA. [68:24] You don't have to figure out how every phenotype is [68:27] caused by DNA. [68:28] If you get the general principles, that's enough. [68:31] Now, the problem is they're not making much progress [68:34] on this what I call the building block approach. [68:36] It seems to me a much more fruitful approach [68:38] is likely to be think of consciousness as coming [68:42] in a unified field. [68:44] Think of perception not as creating consciousness, [68:47] but as modifying the conscious field. [68:49] So when I see the red in this guy's shirt, [68:52] it modifies my conscience field. [68:54] I now have an experience of red I didn't have before. [68:57] Most people-- and the history of science [69:00] supports them-- use the building block approach [69:03] because most of the history of science [69:05] has proceeded atomistically. [69:08] You figure out how little things work, [69:10] and then you go to big things. [69:12] They're not making much progress with consciousness. [69:15] And I think the reason is you need to figure out [69:17] how the brain creates the conscious field [69:20] in the first place because particular experiences, [69:23] like the perception of red or the sound of middle C, [69:26] those modify that conscious field. [69:29] They don't create a conscious field from nothing. [69:31] They modify an existing conscious field. [69:34] Now, it's much harder to do that because you [69:37] have to figure out how large chunks of the brain [69:40] create consciousness. [69:41] And we don't know that. [69:42] The problem is in an MRI, that conscious brain looks a lot [69:47] like the unconscious brain. [69:50] And there must be some differences there. [69:52] But at this point-- and I haven't been working on it. [69:54] I've been working on other things. [69:56] But I want somebody to tell me exactly what's [69:59] the difference between the conscious brain [70:02] and the unconscious brain that accounts for consciousness. [70:04] We're not there yet. [70:06] However, what I'm doing here is neurobiological speculation. [70:10] I mean, I'm going to be answered not [70:12] by a philosophical argument, but by somebody [70:14] who does the hard research of figuring out exactly what are [70:18] the mechanisms in the brain the produce consciousness [70:21] and exactly how do they work. [70:23] JOHN BRACAGLIA: John, it's been an immense, immense honor [70:25] to be here with you today. [70:26] Thank you so much for your time. [70:27] And thank you for talking to Google. [70:29] JOHN SEARLE: Well, thank you for having me. [70:31] [APPLAUSE]