The Philosophy of Information and Artificial Intelligence w/Dr. Luciano Floridi · 苏菲拉底
字幕 字幕位置
--:--
点击播放,这里会跟随视频显示当前句的中英字幕。

The Philosophy of Information and Artificial Intelligence w/Dr. Luciano Floridi

节目发布 2023-10-27 · Parker's Pensées
卢西亚诺·弗洛里迪 PParker Settecase
EDITED TRANSCRIPT · 依据现场录音编译整理,可划线生成便签
编者按:本文是播客《帕克的沉思》(Parker's Pensées)的一期访谈,主播帕克·塞蒂卡塞与信息哲学的开创者、耶鲁大学教授卢奇亚诺·弗洛里迪对谈,话题从信息的定义一路延伸到人工智能的行动者身份、延展心智、结构实在论与模拟假说。弗洛里迪早年治数理逻辑,师从苏珊·哈克,后来在牛津提出「信息哲学」这一领域,其入门读物《信息:牛津通识读本》是本次谈话的底本。本文依据现场录音编译整理。

从逻辑走向信息

主持人:您学术生涯的前半段是做逻辑的,数理逻辑。我查资料时发现,您曾跟随苏珊·哈克(Susan Haack)学习,对吗?她那本《逻辑哲学》(Philosophy of Logics)对我来说是改变一生的书,我读神学院的时候读的。您是从逻辑进的门,怎么后来成了「信息」这个人?

弗洛里迪:说来有趣,那本书当年被译成意大利文的时候,我还完全不懂英文。我是大一学生,英文几乎一个字不认识。我读它纯属偶然,它不在任何课程的书单上,只是典型的那种好奇心过剩的本科生,什么新书都想翻一翻。结果它彻底改变了我的思考方式。

其实当时有两本书。另一本是伦理学导论课的指定读物,摩尔(G. E. Moore)的《伦理学》。忽然之间,我发现了分析哲学这样一个世界:在这里你可以谈问题,而不是谈人。当然这话不完全对,因为后来分析哲学也变成了主要在谈罗素、维特根斯坦和弗雷格。但这是任何哲学潮流的宿命:起于哲学问题,终于哲学家的问题。可在那个年代,我们这些刚起步的小人物,还有那些大名字,处理的还是哲学问题。伦理这边有摩尔,逻辑与哲学逻辑那边有苏珊,你会发现,原来可以用如此清晰精确的工具去处理那些你觉得有趣又重要的问题。这就是起点。

于是我给苏珊写了一封信,贴上邮票寄出去,她当时在沃里克大学。我们开始通信,后来我拿到罗马大学的奖学金,去牛津跟她和迈克尔·达米特(Michael Dummett)一起工作。原本是个「出国六个月写本科论文」的项目,结果我到现在都没离开。

不过你问的是我怎么从那里走到信息的。这就得引入第三个人物,不是摩尔,也不是苏珊·哈克,而是卡尔·波普尔。我又一次走运,波普尔当时还在世,已经退休。我在写一些关于他的东西,作为研究生,照例写了一封贴邮票的信,没想到他真的回了。我们有过一段很短的通信,那封信我至今珍藏,他在信里完全不同意我的观点,不过那是另一个故事。你想,你只是个研究生,卡尔·波普尔亲自来反驳你,你会觉得自己终于算是入了这个圈子。

我为什么要提他?因为波普尔有一篇不算有名却极其重要的文章,如果我没记错,题目是《没有认知主体的认识论》(Epistemology Without a Knowing Subject)。这篇文章讲的是,你可以不从「谁在认知」出发去做认识论,而是(波普尔本人大概会对这个说法深感恐惧)从一种分析的角度,只看知识本身:知识的动态、知识的分布,也就是今天所谓认知逻辑的路数。你开始看到一块块的零件,然后突然意识到:这东西其实不该叫知识,因为它不为任何人所拥有。它叫信息。一旦你开始谈论某种「在外面」的东西,我可能有也可能没有,可能发送也可能不发送,可能去塑造也可能不去塑造,而知识正是由这种材料构成的,那么你很自然会想:这就是信息。

还有最后一个环节,比较私人,我可以讲吗?

主持人:请讲。

弗洛里迪:那要回到很久很久以前,少年时代的卢奇亚诺。我们在高中就学哲学,我最初的兴趣是宗教哲学,克尔凯郭尔那一类。大一的时候,我之所以想多学一点逻辑、逻辑与传播之类的东西,是因为我关心一个问题,这个问题我今天已经没法不带笑地说出口了:上帝如何向人类传递讯息。天使,Angelos,本来就是「信使」的意思。我当时想从香农的角度去理解上帝与人类的沟通,信不信由你:把上帝看作信源,人类看作接收者,中间有一条信道,这条信道如何运作,如果它真能运作的话,才能把那个讯息传过来,也就是「好消息」,福音(evangelion),词根是同一个。这是我做古典学训练时的功课。

把这些放在一起:宗教哲学里有信源与接收者,有奇迹;波普尔那里有没有认知主体的认识论;摩尔和苏珊那里有分析哲学;再加上古典语文学。故事的结尾是这样的:我当时在牛津的沃尔夫森学院做博士后,受邀去伦敦做一场讲座,题目不限。我特别想讲讲这些东西,后来它们被称为信息哲学。我记得我和一位女士坐在学院里一张望着河的小长凳上,我完全想不起她的名字了,真丢人,我一定把那可怜的姑娘烦死了。我说,我可以把这次讲座叫作「什么是信息哲学」,这个领域现在还不存在,但我觉得应该有。对方说,有意思,那你来讲讲看吧,看看我们能一起发现什么。事情就是这么开始的。

主持人:这个故事太棒了,我的听众也会喜欢,尤其是关于克尔凯郭尔的部分,我的听众里有一大批克尔凯郭尔信徒。您当年想用香农信息论来谈神圣启示,这一点特别让人着迷。

弗洛里迪:是的,从那以后我走了很长的路。不过我迟早会回到那个问题上去,那是一段很长的旅程,我终究会绕回关于上帝的问题,那才是真正的问题。

信仰的失落与不信的克制

主持人:这个问题可能太私人了。您做那个项目的时候,本人是相信上帝的吗?现在还信吗?

弗洛里迪:我大概是在本科一二年级的时候失去信仰的。那是四年制课程,一二年级的时候我意识到,信仰已经没有了。这不是某天早上醒来忽然说「我不信了」,它更像一座建筑的地基,被一点一点侵蚀,侵蚀,再侵蚀,然后某一天,轰的一声,整座楼塌了。但楼塌的那天并不是它真正被掏空的那天,那是一个漫长的过程。可以说从少年后期,到十七八岁的时候,我已经几乎完全失去了信仰。

但我要说,我并没有换上另一种信仰。我从来没有成为无神论者。我不想用一种信念去换另一种信念。科学哲学里有句名言:没有证据不等于有证据证明不存在(absence of evidence is not evidence of absence)。如果你足够严谨、足够认真,就应该承认这一点:我们可能永远不知道。所以我今天把自己描述为一个「灵性的不可知论者」(spiritual agnostic)。这是我能走到的最近的位置。

信息是什么

主持人:我不得不问这个问题,每个人都会问您,既然您开创了信息哲学,这是随身份而来的。信息到底是什么?分类的方法有很多种,在这些不同的定义之下,有没有一个核心的东西能把「信息」挑出来?

弗洛里迪:首先要说,这是一团乱麻。这是第一条,每个人都应该从这里出发。我见过很多人,尤其在哲学界,谈论信息的时候好像它是最不言自明的工具,所以人人都懂我在说什么。混乱正是从这里开始的,因为我们理解信息的方式确实有很多种。它们通常有一些共同之处,我一会儿再讲。既然你提到了那本小书,我可以毫无愧色地告诉大家,一切都在那本小书里了。

想象你在谈世界里的信息。可以是动物在森林里留下的足迹,可以是树桩上的年轮,它告诉你这棵树的年龄。这可以说是信息的本体论概念。既然黑胶唱片现在又复兴了,我们还可以说唱针读取的沟槽,用它来产生音乐,那也是一种物理痕迹,你可以指着它说:这就是信息。这是第一种意义:信息作为一种模式。所以很多人常说,我们这里有信息,因为有一个模式。

接着我们转到另一个概念,它们彼此相关。那个模式可能是有意义的,不是随机的,不像天上的云。比如黑胶唱片上被唱针读取的沟槽模式,或者我留在玻璃杯上的指纹,它们有意义,可能有后果。当我们谈论有意义的模式时,我们就开始谈论有意义的数据,这已经进入了语义信息的领域。你谈的不再是「外面有某种结构或模式」,而是关于某物的信息。它不再是「作为某物的信息」,而是「关于某物的信息」。这是我们平时说「信息」时通常采用的意思:我想要一些信息,我没有得到任何信息。今天信息就是一切,没有信息就没法做决定,我们说的就是这个意思:关于别的东西的内容。有一个指涉对象,有内容。内容如果不描述任何东西就没用,所以它是数据加意义加语法。数据有特定的结构。

如果你再往前走一步,加上真值,真或假,那你得到的就是一种语义信息,典型的事实性语义信息,维基百科、火车时刻表里的那种信息。

然后还有第三种,它没有真值,但有意义,有结构,有语义和语法,这就是我们称为「用于某物的信息」的那一类。想想音乐文件。我发给你一个文件,在某种意义上我发给你了一些信息,一定量的数据,有特定的意义和结构,可能压缩过,可能没有。你打开它,它能播放音乐吗?这就是「用于某物的信息」,用来播放音乐。一份蛋糕食谱,也是用于某物的信息。一个算法,你明白我的意思了。但它不见得有真假,你不会说「这份食谱是真的还是假的」。人们会用「真」这个词,但那是「地道」「正宗」的意思,不是哲学意义上的真:这真的是你祖母的食谱吗?是正宗的那份吗?

所以,如果我们谨慎一点:信息作为世界中的某物,信息关于某物(语义的、事实的),信息用于某物(更接近算法那一类)。有了这三种,在理解我们谈论的是什么这件事上,已经是一大步。

树的年轮会说话吗

主持人:我很喜欢这个区分。在书里就是「作为实在的信息」「关于实在的信息」「用于实在的信息」。它对我帮助很大,因为很多人,尤其在通俗层面上,会把它们混在一起,然后说「一切都是信息」,可他们指的是「关于实在的信息」,然后说一切都是那种信息。我想问,「关于实在的信息」有一种特定的意向性,它是关于世界中某个东西的,或许是表征性的。可当我想到「作为实在的信息」,想到指纹或者树的年轮,这类信息有意向性吗?树的年轮是「关于」这棵树的年龄的吗?

弗洛里迪:不,不是真的。那里只有一种我们可以利用的相关性。我们可以把世界中的某物 A 具有某个特征 F,与另一物 B 具有另一个特征 G 关联起来。那个树桩具有那些年轮的特征,与这棵树具有某个年龄相关联,我们利用这种相关性来获取关于这棵树的信息。但如果说「这棵树想给我们传递一个讯息」,那就有点滑稽了。

想象一下,明天你在一个岛上发现了一种未知的古代语言,属于一个已经消亡的文明。我们发现那些符号显然具有结构,这是信息的第一个标志:东西在重复,有规律性,有语法。你看见了语法,看见了模式,就开始觉得这些模式不是自然的,是有人放在那里的。你推断那里一定有意义,有语义,他们在给我们发讯息,或者说他们在广播一个讯息,你需要破解那套代码。

我们得小心,有时候我们太爱打比方了,说树的年轮是一种「密码」。这些都是比喻,要谨慎。树什么都没说,也不想告诉我们任何事。这一点在谈 AI 的时候也很有用:不管你得到什么,那仍然只是模式。AI 并没有想说什么,没有计划,没有意图。凭着那些相关性,你可以在这一头说出树的年龄,但别把树本身和意向性混为一谈。我们对那个模式做了什么,那是我们的事。

主持人:这很有帮助。往深处再挖一点。我发了一条动态说要和您对谈,为了引起兴趣提了一句「环境信息」,结果一位科学哲学家被戳到了。她的意思是,树轮年代学家要做大量的工作才能从年轮中推导出信息,所以年轮不等同于信息。我脑子里想的是塞尔(John Searle)的「观察者相对性」论证,我认同它,我认为需要有一个观察者来解释信息。就像您说的,树本身没有意向性。可是树轮年代学这整门学问,从年轮里找出树的年龄,似乎预设了那里有信息可供关联,至少是这样吧?

弗洛里迪:也可以反过来说:我们一直在找,直到找到某种相关性能帮我们做这件事。这不是科幻,完全可能发现那些年轮其实跟年龄无关。当然那些年轮里确实有关于这个个体的某种东西,跟指纹一样:这棵树有那样的结构,是因为它生活在特定的环境里,慢慢积累,沉淀成它特有的年轮。所以年龄,或者说生长的时间,与生长环之间有直接的相关性,而我们能够从年轮逆向推回年龄。这里没有魔法,没有任何奇怪的东西。

再举一个例子,考古学里读一个遗址,读地层,说这一层属于人类的某个年代,一万年前,那一层六千年前。为什么?因为那些器物,那些地质事件,让我们能够断代。所以那里确实有某种东西帮助我们获取信息。这确实棘手,所以才会有这么多混乱。因为我们使用「作为实在的信息」这个说法时,几乎像是信息本来就在那里,只等着被发现,好像你是把信息从地里挖出来似的。这就把它和另一种情形混淆了。

想象一个不同的例子:我们收到一个来自另一个星系的无线电信号,它已经旅行了几百万年。这不是科幻,是可能发生的,我们没法超光速旅行,但某一天收到这样一个信号是可能的,那将是人类历史的转折点。设想我们收到了那个信号,那种重构是完全不同的性质。那不是读树轮,也不是在玻璃杯上找我的指纹。有人在某处广播了它。这时我们回到了真正的信源、讯息和接收者,这更接近香农。而树没有在发送任何东西,是我们能够解释那个特定的模式,借助它推回树的年龄。或者说,某人出现在犯罪现场,因为指纹在那里,当然然后福尔摩斯会告诉你,不,他不在场,是别人把玻璃杯放在那里的,如此等等。

指纹、背包与抽象层次

主持人:那我想问,情况会不会改变,如果我是有意把指纹留在玻璃杯上的?比如我被绑架了,我把手指按上去,就是为了让人找到,说「帕克来过这里」。这会不会把信息从「环境的」变成别的什么,它现在有意向性了吗,因为我打算通过我独一无二的指纹来传达讯息?

弗洛里迪:我们改变的东西,我想书里也有,是我们观察这条信息所通过的界面。计算机科学里叫「抽象层次」(level of abstraction),我们不必那么花哨,就叫界面好了。比如你用警方的界面去看,他们在找一个可能被绑架的人。或者:我在森林里迷了路,开始一路留下衣服的碎片,这是一个场景。另一个场景,让我把它设计得更聪明一点:我的背包有个洞,我在森林里一路掉东西。第一种情形,我迷路了,想被人找到,所以我有意留下痕迹。第二种情形,从一个不带任何目的的界面去看,两个现象本质上是一模一样的。我只知道有人在林子里走,背包的碎片一路留下来。他是想被找到,还是迷了路、背包破了个洞而已?「有意」是我们在解释时加上去的。这没问题,但我们要能把它和眼前实际有的东西区分开来。

希望我没有把事情搞得太乱。科学哲学里有个说法,数据「不充分决定」(underdetermine)理论。换句话说,你可以有很多很多数据点,而这些数据点常常可以用不止一种方式来重构。我在林子里掉东西的故事也是一样:我是想被找到,还是背包破了?如果警方认为我迷路了,所有人都会把它解释为求救,但那可能只是个意外。我碰巧走运,在不知情的情况下留下了痕迹,然后人们找到了我。

世界是二进制的吗

主持人:我想我大致理解了,如果把信息分析为数据加意义,那么我们可以深入到数据,说数据由比特构成,比特就是一个单位。在二进制的语境里我懂,可在物理的语境里,「作为实在的信息」里,我就不确定了。物理世界里的一个比特是什么?

弗洛里迪:首先,「比特」又是一个不幸的难词。它要么是我们用来编码信息的代码的一部分,想想电报,点、划、点、划。世界是由点和划构成的吗?当然不是。但我们可以把整部《哈姆雷特》变成点和划,从大西洋的这一头发到那一头,这是当年的好日子。所以比特是一种代码,极少是你在世界里找到的东西。世界并不倾向于二进制,至少我们体验到的世界不是。我们通常把它体验为连续的、模拟的,任何东西之间总有别的东西。有句名言:自然不作飞跃(natura non facit saltus)。不作飞跃,翻译成二十一世纪的话,就是自然不是二进制的,它不是从零跳到一,不是「是」或「否」,不是电路里的高电平低电平,不是光盘上有孔没孔。

所以,如果我们谈的不是代码,而是外面那个作为实在的信息,它通常是模拟的。想象一张地图,它是连续的,是线条,比如地铁图,你可以顺着它从一点到下一点。我们经历的这场非同寻常的革命,如果你想找一个英雄,可以说是莱布尼茨,但其实更早,霍布斯就有了,这场革命在于:我们找到了一种办法,把世界上的每一种信息,包括连续统,包括波,都编码成离散的元素,那就是你的代码。而且我们有了能自动处理这种代码的机器。今天我们不仅有能自动处理代码的机器,这些机器还能通过学习来处理它,那就是 AI。这个演进是惊人的,因为在几个历史上令人赞叹的步骤里,你从模拟的、连续的世界走到了数字的、二进制的世界,再走到一台能越来越高效地处理它的机器,以至于这台机器通过统计,还能在原始的模拟世界里找到底层的模式。想想英语,所有说出来的英语,机器从这门语言最深的结构之井里挖出模式,供我们拿来做有用的工作,比如把这个翻译成那个。

这是一段几乎堪称奇迹的旅程,请允许我用这个危险的类比。在我看来,从第一台计算机到 ChatGPT,这段旅程真正非凡。而你如果想「哦,我们正在创造某种形式的智能」,那几乎等于没搞懂这场变革有多大。那等于是在贬低这一步步之中所倾注的非凡的人类智慧,转而认为只要点一下某个你自己都不太明白的黑箱,轰的一声,智能就冒出来了。事情不是这样运作的。

行动与智能的分离

主持人:我从来没这样想过。我之前听您的一个讲座,现在全都串起来了。您说,AI 的意义在于,我们不再需要用智能去停车,自动驾驶汽车,也不再需要用智能去下棋。这是一件大事:您把智能从任务里剔除出去了。这恰恰显示了我们有多聪明,才能做到这一点。而当你说「不,那东西一定是智能的,因为它在做我们做的事」,你其实是在贬低程序员这些人的成就。

弗洛里迪:是的。如果你用「非凡的、强大的、善意或不善意的智能」这种框架来思考,你就属于过去,你还是一个坏意义上的「现代」思想家,现代有好的意义,也有坏的意义。你还困在弗兰肯斯坦那里,困在对我们自身的复制里,它可能接管世界也可能不会,可能善良也可能不善良,弗兰肯斯坦,魔像(Golem)。这种思路没有抓住那场远为有趣的革命。

你刚才的说法是一种表述方式:人类历史上第一次,我们做到了把成功的行动(agency)与智能解耦。这以前从未发生过。我们从来没能在零智能的条件下,朝着一个目标成功地做成一件事,而且还能不断改进。所以我常对这里的学生说:想象在零智能的条件下停车,别拿你父母的车试,他们不会高兴的。想象在零智能的条件下做饭,洗碗,下棋,管理仓库,或者采草莓,我说的都是真实的工业应用,在零智能的条件下造一辆汽车。这在过去是不可能的,那会是一团糟,一场灾难,成本失控。而今天,此时此刻,不是「将来某一天」,而是好几年以来,我们已经有了能成功做到这些的机器,而且越做越好,因为它们从自己的输出中学习,把输出变成输入,性能不断提升,直到停车停得完美,草莓采得最高效。

行动与智能的这场离婚,是历史性的。请允许我再往前看一步:这正是问题所在,挑战所在,机会所在,但也是问题所在。当我们谈论 AI 伦理时,真正的要害就在于我们正在制造并不断扩大的这道行动与智能之间的鸿沟。在这道鸿沟里,你能找到所有的难题:隐私、偏见、版权与知识产权、对资源的控制、剥削,往下数,假信息,深度伪造,全都在这里。它们都源于这样一个事实:过去这几年,我们已经能在零智能的条件下把事情做得越来越成功。这才是我们应该更好地理解、更好地设计的东西。

河流、狗、人与第四种行动者

主持人:我现在正在上一门人工智能哲学的课,一直想找出某种区分。我知道计算机科学那边的 AI 人士不喜欢「人工智能」这个词,因为他们认为那是真正的智能。我倒觉得这个词很完美:它就是人工的,它是程序员意图的一种表征。但我想问,您会不会把机关陷阱和自动驾驶汽车归为一类?想想青铜时代的陷阱,有人踩上某个机关,一支飞镖射进胸口。这也是没有智能的行动吗,只是形式更粗陋?

弗洛里迪:不。这里也需要做几个区分。我们现在拥有的这种人工智能(其实已经有一段时间了),是统计学类型的,而不是过去的演绎类型的,它强大得多。它的特点是学习的能力,我说的行动就是这种行动。机关陷阱不会学习。就像地震也是一种行动,但不是我们说的那种,因为地震不会学习。它不会说:我真是一股邪恶的力量,下次我要造成更大的破坏,我要等所有人都睡着了再来。当然不会。

接下来的话,对任何在听的人来说,是完全不正统的,别在家里尝试。我因此受过批评,《人工智能伦理学》(The Ethics of AI)出版前,有一位审稿人看到我做这个区分,说这完全是胡说八道。幸好我有足够的回应,说明他们没弄懂我在说什么。那么来看:想象行动是有程度的,从很弱到很强。最高的程度大概是全能,想想上帝,无所不能,无所不学,无所不知。先把上帝放一边,只看这个星球上的情形。

第一种,与世界的互动。只要你在世界里造成了差别,你就已经是一个行动者。我们有时说「大气作用者」,就是这个意思:雨,地震,一条河。一条从山上一路流到海里的河是一个行动者,无论好坏,因为它满足最低的条件:在世界里造成差别,造成改变。这是互动。

第二种,介于中间的东西,比如一只普通的动物,一条狗。狗会学习,也有意图,它想活下去,会躲开危险,不喜欢一条更大的狗靠近。所以既有学习,也有我们姑且称为意向性的东西。

第三种,也是最高的程度,是人。人有以上全部,还有超越以上全部的能力,或者说有意识地控制以上全部的能力:你可以说「我要自杀」,狗不会;你可以说「我不必害怕这个东西」;你可以说「我要把手指放在蜡烛火焰上三十秒」,我们小时候干过多少这种蠢事,就因为「我想这么做」。这种有意识的控制,是一种额外的行动形式。

这就是三种:河,狗,我。从我们走出洞穴以来,地球上的生命就是这样。而今天有了第四个玩家。它不像河,因为它不只是互动,它还能学习。它不像狗,因为它没有意图,不会说「我今天不想吃这个」,不会因为想出去散步而叫。它没有那种意向性,当然更没有有意识的自我控制和规划,比如「两年后我想去探索亚马逊雨林」。所以这里有第四种行动者,我们过去从未遇到过,我们没有与之相称的概念工具。

结果是,有些人把它仅仅看作又一种互动形式:它是个工具,跟电钻没什么两样,跟汽车没什么两样,河那一类的行动,它做它做的事,我们控制它。这是坏主意。至少我的旧车是这样,新车会学习了,旧车不会。另一些人说,不,这是一种新形式,它几乎像一条正在长大的狗,像一个五岁的孩子。我们拥有的技术和科学,应该足以告诉我们至少今天不是这样。我认为这是胡扯,不管怎么说,我们谈的是我的 iPhone,或者 ChatGPT,它当然不像我们。

所以我们必须承认,我们有了一个新玩家,游戏变得更复杂了一点。我们造了它,我们控制它,或者至少我们有控制它的责任,我们可以开机关机。它能做惊人的事,能从自己的表现中学习。它不像河,不像狗,不像人。我们需要的是哲学家做一点工作,去理解这种新的行动形式。这是我觉得激动人心的开口。任何在听的研究者,你手上有一个了不起的题目,这就是二十一世纪。如果你试图把这种行动形式还原为「其实它是某种智能」,那你只是不肯承认某种全然新鲜的事情已经发生。「我的模型是旧的,我的诠释学还是现代的,我只有这三个范畴,所以它必须是一、二或三;既然它不是河,它就一定像狗或者像人。」打开思路,想想第四种。它完全合理:我们设计了这种解决问题、执行任务的人造物。

主持人:行动者这个问题实在太难想了,这恰恰给您那个论点加了个惊叹号:我们需要更好的概念来理解这个新东西。我在想,也许可以借用群体行动者的文献。ChatGPT 是一个群体行动者吗?我在用 3.5,可还有 4,这两者之间是什么关系?而且如果我和您同时在用它,把它叫作「一个」行动者还有意义吗?

智能是一片群岛

主持人:我们不想说它是智能的,可 AI 那边的人会说,智能是一个功能性概念,就是解决问题而已。

弗洛里迪:关于智能,如果到头来只是一个词汇问题,那我什么词汇都接受,你可以叫它「天才」,我不在乎。问题在于这个词汇选择的后果。比如,如果你随后开始担心它会接管宇宙,那就傻了。如果我把这支笔叫作智能的,因为我用得越多,看到的墨水越少,这很「智能」,好吧,从明天起,「智能笔」,无所谓。这里面有很多是广告,是炒作,是推销产品。让我们哲学一点:你得对词小心,因为词有后果,有蕴涵。而「智能」,我们并没有定义。顺便说一句,好奇的人可以看看有多少个学科给智能下了多少种定义,简直是个笑话。

所以真正的问题变成:既然没有智能的定义,我们在谈什么?我们确实有一些标准,你想知道这些标准在这里是否得到了体现。比如,因为有更重要的事情,停下正在玩的游戏。你在和电脑下棋,谁在赢无所谓,忽然火警响了,快跑!电脑会继续下,它会赢了那盘棋,然后和房子一起烧掉。所以智能也是一种灵活性。有社交智能,有音乐智能,有那种直觉的智能,让你把一个坏消息推迟告诉朋友,因为今天不是合适的日子。

所以当人们问我「那到底什么是智能」,我说:智能是我们每天可能犯下的所有愚蠢的对立面。坐飞机把护照忘在家里是蠢的,带上护照就是智能的。关车门时把手指夹在门里是蠢的,不夹手指就是智能的。认为二加二等于二十二是蠢的,认为等于四就是智能的。今天对那个朋友说了那句话是蠢的(「帕克,你难道不知道劳拉今天过得很糟吗?」),不说就是智能的。你读了报纸却不会概括,是蠢的。我们称为智能的东西的种类几乎无边无际,因为它是「做人」的另一种说法,是拥有一段成功的心智生活。这不是一个可以像「水等于 H₂O」那样钉死的定义,那是科学,是好科学,但那是在我们拥有规整的词汇、能处理清晰精确的概念的时候。而在这里,我们有的是一片群岛。你可以把这片群岛叫作马尔代夫,但没有哪一座岛是马尔代夫,那样说很傻,马尔代夫是整片群岛。

再打一个比方。这就像有人在找意大利和法国之间的边界:它在法国,还是在意大利?它是边界。「可它总得在某个地方吧?」不,把一边和另一边区分开来的,正是那条既不在这边也不在那边的线,因为你一跨过去,就在另一边了。所以你得在心智上灵活一点,智能一点。于是「这台机器智能吗」这个问题,要么是琐碎的,要么是提错了。琐碎,因为它取决于你说的智能是什么:如果是会算数,那我的袖珍计算器就是智能的,它算数比我认识的任何人都强,了不起。提错了,因为那片群岛根本不适用于它,那是在比苹果和橘子。我觉得 AI 哲学里大多数这类讨论,不是琐碎就是无谓。

主持人:我觉得您谈信息的三种方式在这里特别有用,我在 AI 讨论里很少看到有人用它们。它或许也能压一压炒作,因为您等于在说:你在混淆范畴,你把这些算法里「用于实在的信息」当成了「作为实在的信息」,或者别的什么。

弗洛里迪:完全正确。混淆这些范畴会制造无穷无尽的困惑,以及对某种东西的担忧,坦白说,就像担心僵尸一样。我不担心僵尸。我不担心僵尸意义上的 AI。我担心的是人类使用 AI。

主持人:这就是赤裸裸的偷换概念。这是一个很好的工具,能显示我们在「智能」或者「AI」这个词上玩了歧义。

大脑是模拟的还是数字的

主持人:我想聊点更出格的东西。我觉得您能帮我思考查尔默斯和克拉克的延展心智(extended mind)论题。您提到波普尔那篇《没有认知主体的认识论》,您认为那其实不是知识,是信息。那我想问,此刻我心里的想法,不管心智是什么,它们更像模拟的还是数字的?更像黑胶唱片还是 CD?然后我还有后续问题。

弗洛里迪:当然,大脑是一台模拟的引擎,至少因为归根到底,一切都取决于它得到多少血液。过去有很多东西被表征为数字的,神经元要么放电,要么不放电。然而现在我们发现,比如放电是有波的,这些波也决定我们的神经生命过程。所以在某种程度上,这有点像在问音乐是数字的还是模拟的。这是两种代码,你可以把音乐编码为模拟的,也可以编码为数字的。每次编码,你可能丢失也可能不丢失一些东西,可能是有损的,你不断转译下去,最后几乎面目全非。

你一定看过那个实验。我妻子是神经科学家,她是这里的神经科学教授,之前在牛津当神经科学系主任,我们一起来了耶鲁,她现在是这里神经科学中心的主任。她的一个朋友,也算我的间接朋友,在伯克利,做了一个研究:通过读取大脑来重建人们正在听的音乐。结果是平克·弗洛伊德的歌。不知道你听过没有,他们得先告诉你那是平克·弗洛伊德,你才勉强能认出来。但确实令人印象深刻。然后人们说,天哪,AI 会读心了。

这有多混乱?首先,这有点像个把戏,我们能做这件事已经有一段时间了,只是现在做得好得多。其次,声波进入一个模拟的大脑,大脑收集这些波,你记录大脑的波,这基本上就是从一种代码翻译到另一种代码,再到另一种,再到另一种。当然损耗巨大,从实际的声音,到耳膜的振动,到大脑对这些振动的编码,再到输出的东西。你之所以知道那是平克·弗洛伊德,是因为你知道输进去的是什么。如果你事先不知道,你根本无法解读输出。这就是翻译。如果你知道如今的神经科学有多少是这样做的,就不会觉得奇怪。

回到你的问题:我们谈的是一个模拟的世界还是数字的世界?我认为这个问题本身就放错了位置。我们拥有的是来自世界的信号,我们可以把它们编码为数字的或模拟的,而且可以随时切换。如果你觉得这太玄,记住光:光子,取决于框架,看起来像点,或者像波。问光「究竟」是波(连续的)还是由一颗颗子弹般的点构成的,说明你没弄懂自己在说什么。它是这个或者那个,或者两者都是,取决于你放在你和光之间的那个界面。

主持人:那么和延展心智的关联在这里:我有一本笔记本,里面全是我关于您的工作的思考,它引导着我,像我的活动记忆。它们似乎是我思想的表征。可我想问,笔记本里的笔记算模拟的还是数字的?我会觉得是数字的,虽然听起来奇怪,因为我们会说「模拟的笔记本」。

弗洛里迪:这真的取决于解释。除非你先问清楚这个问题是为了什么,否则没有办法独立地回答是或否。

心智是环岛,不是道路

主持人:查尔默斯和克拉克会说,也许我们的心智延展进了笔记本,因为我们的思想在那里。他们举了一位阿尔茨海默病患者的例子,他用笔记本来引导自己在世界中行动,他们说,在一种非常真实的意义上,这就是他的心智。我不清楚他们说的是哪种意义,没说是比喻的、类比的还是单义的,但他们想让你接受的结论是:你的心智不只在你的头骨里,它延展到你的思想所到之处。

弗洛里迪:我读它的方式是,这是一个不错的比喻。超出比喻的任何东西,我都会非常批判。首先,我从来没见过、没摸过、没踢过一个心智。如果你开始把心智想成脑子里的一种液体,流出来把周围的东西都弄湿,那是个坏主意。我们有心智生活。对于觉得这说法奇怪的人,想象有人说:世界由对象、行动和对象的属性构成。有汽车,汽车可以是红的,可以跑,很好。那「心智的」在哪里?按这个类比,「心智的」基本上是一个副词。我们不是「有」一个心智,我们是「有心智地」活着。请原谅这个过度简化。如果你对心智持一种副词的看法,而且像我一样认为真正重要的是一种二阶的本体论,认为事物不是先有东西、再有东西的属性、再有东西之间的互动,而是关系构成了事物,那么「有心智地活着」这件事,就在你、在我、在新生儿或者别的什么人身上得到了肉身化,整幅图景几乎是颠倒过来的。我觉得这在哲学上更有说服力,因为它不会凭空造出一个叫「心智」的本体。

那很危险。换一个比方,我希望它更直观,其实是同一个故事:民主。我生活在一个民主之中。其实你生活在一栋房子里,这个「民主」是什么东西?哦,你生活在一个行动和互动都是民主式的地方,你民主地生活着,现在我明白了。或者:某人跑得很快,那个「快」在哪里?请指给我看。如果你从一种基于事物的本体论出发去看,而你实际谈论的是存在的方式、关系性的方式,混淆就会导致一些很好笑的说法,比如「我的心智在外面」。我理解那个比喻,但更有意思的说法是:在一段有心智地度过的生活里,我当然是这张网络中心的节点,但心智性不只是这个节点,它是我与网络的全部关系。

最后一个比方,免得有人觉得这家伙完全在胡说。想想环岛,或者你更喜欢十字路口也行,环岛是很英国的说法。十字路口和道路,问人们哪个先有?当然是道路,十字路口是道路碰巧相遇的地方。我的心智是什么?我的心智是一个十字路口,一个环岛,是所有以特定方式互动的东西汇聚的地方:我过去的记忆,我说英语的方式,我的生活,我的经验,我的感受,我身体的当下状态,所有这些关系,所有这些存在方式,都汇聚在这里。而且它非常具体:过环岛的时候别不停车不看路就冲过去,它完全是实在的。可如果先想环岛(先有心智),再想有心智的存在,那就是先放十字路口,再修路把它们连起来。这说不通。

主持人:这也许是跑进兔子洞了,我们可以不谈。但我脑子里有那些实体本体论者,他们会说:关系是事物之间的关系,如果你有关系而没有事物,那我们到底在谈什么?这就像鸡和蛋。

鸡与蛋留给形而上学家

弗洛里迪:同意。这里需要两步。首先,科学哲学里有一种做法叫结构实在论(structural realism),你可以用信息的方式来做,试图同时解决鸡和蛋的问题。但那有点专门,听众可以去查「信息结构实在论」(informational structural realism),那是一条可走的路。

主持人:我们做过一整期节目谈这个,感兴趣的听众可以去听。

弗洛里迪:那我们就保持简单直接,不谈科学哲学,谈日常经验。谈论实在至少有两种方式。一种是:我有通道,我看,我检查,也许借助工具,借助科学,借助理论,但我有办法告诉你世界是什么样子,也许可以改进,但那种可能性是开放的。另一种是:别管外面是什么,你能看出我要说什么了,外面的东西在向我们发送信号,我们得到的是信号,信号构建了我们自己的本体论。我的本体论,此刻和你谈话,我手里的这支笔,是这个特定的接收者处理那些特定信号的结果。这就是我生活于其中的实在,它完全是真实的。信号是真实的,一切都是具体的。我不是在说「没有什么是真实的」,那种东西留给孩子的游戏,留给哲学家玩的游戏。真实的生活,你去买菜再回来的那种生活,我们谈的是哪种实在?形而上学的那种,实在自身,本体(noumenon)?还是我通过这个特定界面(在认识论意义上,就是我的身体,我的文化,我的语言,等等)与世界互动的结果?把后者叫作本体论。

有了这个区分,我就到鸡和蛋了:鸡和蛋的问题是形而上学家的问题。形而上学家想告诉你世界自身是什么,独立于我、独立于任何人、独立于任何特定界面,仿佛他拥有上帝之眼的视角。经典做法,他得告诉你基础的构件是什么,什么在先,什么在后:环岛和路,事物和事物之间的关系,节点和关系。他们不知道,他们一直在绕圈子,当然了,鸡和蛋嘛。

主持人:而且如果我们采取一种康德式的图景,你等于是在试图逃出自己的心智,你在用自己的语言、自己的范畴去做超越的事。

弗洛里迪:别去做形而上学,去做本体论,意思是:我们能给这个「认识论层面的本体论」建立的最好模型是什么?建模的美妙之处在于,你可以决定。你可以说,我们决定从这里开始建模,因为这样更合理,更有成效,更有解释力。我们决定从线开始,而不是从点开始。然后你在白板上画:这是一条线,一个关系,关系连接 A 和 B。我刚刚就做了。你就这样跳出了那个把戏。这只是在说:我想怎样给鸡和蛋的关系建模?我打算按时间顺序来,从蛋开始,再到鸡,这更合理。在真实的生活里我并不做形而上学,那是你的问题,不是我的。所以步骤是:认识论层面的本体论,而不是形而上学的进路;而在认识论层面的本体论上,我们意识到必须在哲学上给它建模才能理解它,而建模就是在设计你所需要的东西,你想从哪里开始都行。

信息一元论是一次信仰之跃

主持人:有些人想拿这个认识论层面的本体论,再把它推回形而上学,说:信息是一种中立一元论(neutral monism),你看我们用信息和我们接收到的信号在认识论上能做多少事,那就干脆说本体界(noumenal realm)就是信息吧。您怎么看?

弗洛里迪:这里是有自由的。你的本体论承诺(ontological commitment),借用蒯因的术语,可以走得远得多。你可以说:我有一个认识论层面的本体论,它运转得很好,好到我能把机器人放到火星上,所以这是理解那个终极的本体信源发给我的信号的一种坚实的方式,我以最好的方式解释了它们,而且如此成功。这种工具性的成功,是我作为人类的本体论运转的方式,完全没问题。那我是否必须走下一步,说「世界本身就是这样」?你想说也可以,但那是一次信仰的行动。没有任何东西迫使你这么说。作为一个好的科学家、学者、哲学家,我们该做的是留在我们的理解所担保的范围之内。如果我们想在有理由相信的东西之上再说「还有更多」,那本质上是一个无摩擦的承诺,是或否,随便谁都可以玩。通常一个承诺是被要求的,而在这里,它是百分之百自愿的,对一个并不被要求的东西做出本体论承诺。我可以做吗?随时可以。我必须做吗?不。

想象这样一件事:收音机发送信号,发送音乐,我在这头接收音乐,我把信号解释为音乐。我还想相信这些信号表征了收音机的内在本性。你有这么想的自由。它们是吗?不太可能。收音机发送信号。有人说,不不不,收音机发送的是一种特定的信号,关于它自身本性的信号。有意思的想法,也许吧,可我们怎么知道?这就是我的康德式警报开始变红的地方。这也没什么错,就像讲故事,谁都可以,玩得开心。但我还是宁愿克制,那样好得多。

哲学家的问题与哲学问题

主持人:最后一个问题,您已经非常慷慨了。这是替我的听众问的,他们酷爱模拟假说。我看到这里有一条线:如果我们接受某种先验观念论,我们接收着这些信号,说不出物自体是什么,这是否意味着我们生活在一个模拟之中?即便不是计算机模拟,我们看到的这个世界,生活世界(Lebenswelt),「对我而言的世界」,本体论的世界,不管用什么术语,它是不是一个被模拟的世界,也许是我们自己的心智模拟出来的?

弗洛里迪:这种讨论恰恰是给哲学抹黑的东西。这就是哲学系被关掉的原因,因为人们觉得,你们原来干的是这个?那我们不需要你们。有意思,但让我认真对待它一会儿。这就像说:我们生活在一个满是鬼魂的世界里,你摸不到它们,看不到它们,甚至无法证明它们在那里,可它们在那里。是或否?这是什么样的问题?前提本身就取消了对它给出任何理性回答的可能。我们是否生活在未来某人生成的模拟之中?前提取消了任何合理回答的可能。这只是一个把戏,又一个哲学家爱玩的把戏。他们不该玩,至少不该公开玩,因为当社会看到我们在这种蠢事上浪费时间,人们的评判是我们自找的。如果哲学就是这个,人们会说:别花我纳税人的钱,周六周日自己玩去。

让我换一种方式说。有很多问题,我称之为「绝对问题」:它们被提出来的时候,对合理答案的空间不施加任何约束。如果你通过提出那个问题,就已经取消了给出任何有意义答案的机会,那你一开始为什么要问?这就是我们走错的那一步,所有那些思想实验,模拟,源头上都有一个混淆,是从笛卡尔的恶魔开始的,甚至更早,柏拉图的隐身戒指。柏拉图和笛卡尔是可笑的人吗?不是。他们在做思想实验,把它作为达到目的的手段,而不是目的本身。他们并不是在问「我是否可能一直被一个恶魔欺骗」「我是否生活在一个恶魔制造的世界里」,母体(Matrix),模拟,永远是同一个故事。笛卡尔没有问那个愚蠢的问题,因为他是天才。他问的是:我怎样才能对这辆叫作「人类知识」的车做最极端的路测?它在什么样的极端天气下会崩溃?最极端的条件是:想象一个恶魔。它连这个测试都能通过吗?至少「我思」能通过。那么即便恶魔在场,我也能开这辆车。我希望大家明白这个类比。别忘了,笛卡尔是个工程师。他说的是:让我测试「人类知识」这个人造物,施加压力直到它崩溃。当我火力全开,恶魔上阵,它崩溃了吗?没有,还剩下「我思」。宾果,我可以从这里重建一切。这是经典,我们至今还在读它,就是因为这一点,而不是因为他在纳闷自己是不是生活在一个恶魔让他觉得眼前有五个光点的世界里。那完全没用。如果那不是你用来测试理论的工具,而是游戏本身的终点,那你玩错了游戏。那个游戏是浪费时间。

这就回到我们开头说的:那不是一个哲学问题,那是一个哲学家的问题。没有人在乎哲学家的问题,所有人都在乎哲学问题。如果这期节目只留下一句话,请记住这个区分:我在问一个哲学问题,还是一个哲学家的问题?延展心智,虚拟实在,模拟,AI 接管世界,这些会被写进「哲学如何令人尴尬」那一章,不了,谢谢。我不想看到那样一本书。我够尖刻了吗?

主持人:太精彩了。弗洛里迪教授,感谢您付出这么多时间,也感谢您允许我把这些问题,包括哲学家的问题,都过了一遍。

本期讲者
卢西亚诺·弗洛里迪意大利裔哲学家,信息哲学与数字伦理的奠基者,著有《信息哲学》《AI 伦理学》。长期任教牛津大学,2023 年起任耶鲁大学数字伦理中心创始主任。
Parker Settecase美国播客《Parker's Pensées》主持人,神学院背景,节目以哲学与神学访谈为主,本期正在修读人工智能哲学课程。
章节 · 点击跳转视频
0:00 开场与嘉宾介绍 ▶ 正在看
1:52 从逻辑到信息:一门学科的起点 ▶ 正在看
10:17 失去信仰与灵性的不可知论 ▶ 正在看
12:44 信息的三种含义 ▶ 正在看
17:35 年轮、界面与比特:信息在哪里 ▶ 正在看
30:28 AI:能动性与智能的解耦 ▶ 正在看
36:25 第四种能动者:河流、狗、人之外 ▶ 正在看
43:31 智能无定义:群岛与国界 ▶ 正在看
49:11 大脑编码与延展心智 ▶ 正在看
55:26 心智是副词:关系本体论 ▶ 正在看
1:00:38 信号、本体论与形而上学的边界 ▶ 正在看
1:07:42 模拟假设与「哲学家的问题」 ▶ 正在看
本期论点
本期回应
53:37
世界本身无所谓模拟或数字,两者只是我们对世界信号的两种可互换编码 实在由我们构成世界有它自己的样子吗?
1:01:41
外部世界向我们发送信号,我们拥有的实在是接收器处理这些信号的产物 实在由我们构成世界有它自己的样子吗?
1:11:22
笛卡尔的恶魔与柏拉图的隐身戒指是检验知识的手段,当成目的就玩错了游戏 这问题本身不成立世界有它自己的样子吗?
20:14
人工智能输出的仍然只是模式,它并不打算说什么,没有计划也没有意图 还差得远机器能真的理解吗?
48:00
「这台机器到底有没有智能」要么是毫无意义的问题,要么就是问错了 问法错了机器能真的理解吗?
31:42
把AI说成在创造某种智能,等于贬低了投入这场技术转变的人类智慧 算人的机器做出的新东西,创造力算谁的?
40:30
人工智能是人类从未遇到过的第四种能动性,现有概念工具不足以把握它 没有先例AI 带来的变化,以前有过先例吗?
56:42
心智不是一个实体而是一个副词,人并没有一颗心,而是「有心智地」活着 心身分不开心灵和身体是两回事吗?
其他论点
6:45
构成知识的原料不为任何人所拥有,因此它该被称为信息,而不是知识
13:35
动物足迹、树木年轮这类物理模式本身就是信息,而不只是关于世界的信息
16:09
食谱和乐谱这类「为了某事的信息」有语义和句法却没有真值,是独立的第三类
35:01
AI伦理的真正关键在于能动性与智能之间那道不断拉大的裂缝
37:51
能动性有程度之分,凡在世界中造成差别的东西都已经是行动者
57:08
并非先有事物再有属性与互动,而是关系构成了事物
1:09:38
「我们是否生活在模拟之中」这个提问的前提本身摧毁了理性回答的可能
01开场与嘉宾介绍
0:00
hey welcome back to another episode of Parker's penes I'm your host Parker setticase and this is a podcast where we explore all the deepest ideas in philosophy theology nature and life I love thinking about cool stuff so come think with me today's episode is very very special I have with me Dr Luciano FL we're going to be talking about uh information which is really fun the philosophy of information he is a Pioneer I think he's invented the philosophy of information in a sense in a sense it's always been here the whole time that's one of his arguments as well it's really cool uh you guys can check out his his short little book information a very short introduction he's got a lot of other books but uh this is a great intro to it I've been blessed by it as well today we're going to be talking about information maybe some AI but just what what is information that's going to be a really fun one before we jump in I want to thank everyone who's making this podcast happen over on patreon if you guys like
嘿,欢迎回到《Parker's Pensées》的又一期节目,我是主持人 Parker Settecase,这是一档播客,我们在这里探讨哲学、神学、自然和人生中最深刻的那些思想。我特别喜欢思考有意思的东西,所以来和我一起思考吧。今天这期非常非常特别,我请到了 Luciano Floridi 博士,我们要聊的是「信息」,这真的很有意思——信息哲学。他是一位开创者,我觉得从某种意义上说,是他发明了信息哲学。当然从某种意义上说,信息一直都在那儿,一直都在——这也是他的论点之一,非常酷。大家可以去看看他那本小书《信息简论》(Information: A Very Short Introduction),他还写过很多别的书,但这本是很棒的入门读物,我自己也从中获益良多。今天我们要聊的是信息,可能还会聊聊人工智能,但核心是:信息到底是什么?这一定会很有意思。在正式开始之前,我想感谢所有在 Patreon 上支持这档播客的朋友。如果你们喜欢这个节目,可以在 Patreon 上支持,
便签引用
0:52
the show support it on patreon or YouTube members this one this episode's also support uh brought to you by Saddleback leather you guys know I love Saddleback leather stuff look uh in the description for link to Saddleback I have it all over my office it's amazing stuff I love it check them out and if you use my link then uh you'll support the podcast so please do that no not too much Comm modification today because I want to jump in with Dr F so let's just pull him right in Dr F thanks thanks so much for coming on the podcast thank you thank you for the invitation um before we get in so I already I pronounce the podcast incorrectly and I do that intentionally because I'm an American swine and I I think it'd be a little pretentious if I just jumped in from my Chicago accent to Pon but also my my last name is Italian my grandfather or my grandfather was Sicilian but my dad was adopted by him he's Italian you're you're Italian can you pronounce my last name for me you can you can you help us with
或者成为 YouTube 会员。这一期还由 Saddleback Leather 赞助播出。你们知道我有多喜欢Saddleback Leather 的东西,看一下简介里有 Saddleback 的链接,我办公室里到处都是他们家的东西,特别棒,我太喜欢了,去看看吧。如果你用我的链接,就等于支持了这档播客,所以拜托了。今天广告就不多说了,因为我想赶紧和 Floridi 博士开始。那我们就直接把他请进来吧。Floridi 博士,非常感谢你来上这档播客。谢谢,谢谢你的邀请。嗯,在我们开始之前——我刚才故意把播客的名字念错了,我一直是故意这么念的,因为我是个美国土包子,我觉得要是我用芝加哥口音突然切换成法语的 pensées,会显得有点装。不过我的姓其实是意大利语,我祖父——我祖父是西西里人,我爸爸是被他收养的,他是意大利人。你也是意大利人,你能帮我念一下我的姓吗?你能帮我们念一下吗?我得先找到你的姓才行,
便签引用
02从逻辑到信息:一门学科的起点
1:52
this I need to find your son name first of all but yeah it's uh right down in right down in the picture it's right that there it is uh septica C it there we go it means seven houses yeah that's right set the can can you do it one more time for me okay here we go uh that's why the invitation I can see that okay set the C oh it's beautiful that's so good thank you um well Dr FL you you were you spent like half of your academic life as as a logic guy mathematical logic and I just realized um I was listening to one of your interviews and you said you or maybe I was looking up your your info about you and you studied with Susan hack is that right I did yes so so her book philosophy of Logics is like epic so wonderful it life changing for me a while back when I was in seminary beautiful amazing you you got in through logic and somehow you became the the information guy can you can you help us with that how'd that happen yeah so um well that book uh was um at a time was translated into Italian at a time I
首先。不过没错,就在下面——就在画面里,就在那儿,就是它——Settecase。就是它,这个名字的意思是「七栋房子」。对,没错。Sette-case。你能再念一遍吗?好,来了——嗯,这就是为什么我发出这个邀请。我看到了,好——Settecase。哦,太美了,太棒了,谢谢你。嗯,Floridi 博士,你有大概半个学术生涯是搞逻辑的,数理逻辑。我刚刚才意识到——我在听你的一个访谈时,你说过,或者可能是我在查你的资料时看到的,说你师从 Susan Haack,是这样吗?确实是的。她那本《逻辑哲学》(Philosophy of Logics)简直是经典,太精彩了,在我念神学院的时候改变了我的人生。太美了,太棒了。你是从逻辑入行的,后来却成了「信息那个人」,你能给我们讲讲吗?这是怎么发生的?好的,嗯——那本书当时被翻译成了意大利语,那时候我
便签引用
2:53
didn't have any English that was a balancing so I was a a first year student undergraduate and uh my English was like I could hardly all their uh so I read that um accidentally it was not part of any course but you know the the usual curious uh investigative uh undergrad that wants to know more about this and that just came out um and uh it it just made a revolution in my way of thinking um there were two books one actually was a set Book for a course on ction to ethics which was ethics by EG Moore and all of a sudden I discovered this world of analytic philosophy where you could actually talk about problems not about people well that is not entirely true because then anal philosopy became mostly about Russell weakness and Fredy uh but it's the destiny of any you know philosophical Trend uh you start with the philosophical problems and then you end with philosophers problems but at the time we shall we say uh the community of uh early stage career little guy me and other big big names we
一点英语都不会,说来挺不平衡的。当时我是本科一年级学生,我的英语水平差到几乎看不懂,所以我读那本书纯属偶然,它不是任何一门课的内容,你知道的,就是那种好奇心重、爱刨根问底的本科生,想多了解这个那个,刚好那本书出版了。嗯,它彻底革新了我的思维方式。当时有两本书,其中一本其实是一门伦理学导论课的指定读物,就是 G.E. Moore 的《伦理学》。突然之间,我发现了分析哲学这个世界,在那里你可以真正讨论问题,而不是讨论人。当然这话也不完全对,因为后来分析哲学基本上变成了关于罗素、维特根斯坦和弗雷格的研究,但这是任何一个……的宿命,你知道的哲学潮流,呃,你是从哲学问题出发的,最后落到的却是哲学家们的问题,但在当时,我们可以说,呃,那个圈子——刚起步的小人物,比如我,还有其他一些大名鼎鼎的人——我们处理的还是哲学
便签引用
4:05
were still dealing with philosophical problems and both from the ethic side EG mo and uh from the logic and philosophical logic uh Susan uh it it was fascinating how you could have remarkable clear and precise tools to tackle issues that you thought were interesting important um so that was the beginning uh I wrote a letter with a stand uh to Susan uh who was a warri and I started a little conversation with her then I got a scholarship from Rome University to go and work with her and Michael Dam oh yeah work in in in Oxford and what was a six months uh write your undergraded thesis abroad kind of project what Finds Me Now a y so I never left basically but it was fascinating but you ask actually sorry um the question how did I move from there to well the thing we need to introduce a third character uh who is not just uh e mo and Susan hack but uh poer kind of Po yeah I got lucky uh again uh poer was still alive I was retired uh I was riding some philosophical stuff on it work and and again as a gradate student would I
问题,既有伦理学这一边,比如摩尔(G.E. Moore),也有逻辑和哲学逻辑这一边,呃,苏珊(Susan Haack)。让人着迷的是,你竟然能拥有极其清晰、精确的工具,去处理那些你觉得有意思、重要的问题。呃,那就是起点。我贴了张邮票,写信给苏珊,呃,她当时在华威大学,我就这样跟她开始了一点点交流,后来我拿到罗马大学的奖学金,去跟她还有迈克尔·达米特一起做研究,噢对,是在牛津。本来只是个六个月、去国外写本科毕业论文那种项目,结果你看我现在还在这儿,所以基本上我就再没离开过。但那真的很迷人。不过你问的其实是——抱歉——你问的问题是我怎么从那里走到这一步的。这里我们需要引入第三个人物,不只是摩尔和苏珊·哈克,还有,呃,波普尔,算是波普尔吧。对,我运气不错,呃,那时候波普尔还在世,已经退休了。我当时在写一些关于他工作的哲学东西,然后又像一个研究生会做的那样,
便签引用
5:29
the usual letter with a stamp and said car unbelievable uh he actually replied so we had this very quick conversation I cherish that that letter I must have that particular letter in which he disagreed completed with me but that's another story but you know you're not you're a grad student and is not K po disagreeing with you you think you're really not part of the club yeah and so why I'm introducing him because um there's a particular uh not terribly welln remarkably important article by which is if I remember correctly long time epistemology without the knowing subject and in this article it speaks about how you can do not the knowledge epistemology know from what I would considered POA would have been horrified but an analytic s perspective in terms of um looking at just the knowledge itself I mean the the Dynamics of knowledge the distribution of knowledge what would you know the con would be called um the sort of epistemic logic approach to it so you start seeing the pieces and then a lot
照例贴上邮票写了封信,结果,天哪,难以置信,呃,他真的回信了。所以我们有过一段很短暂的交流,我一直很珍惜那封信,我肯定还留着那封信,信里他完全不同意我的看法,不过那是另一个故事了。但你知道,你当时只是个研究生,要是卡尔·波普尔都不来反驳你,你会觉得自己根本不算圈内人。对,所以我为什么要提他呢,因为,呃,他有一篇不算特别出名、但极其重要的文章,如果我没记错的话,标题很长,叫《没有认识主体的认识论》。在这篇文章里,他谈的是你可以不从知识出发做认识论——从我认为波普尔本人大概会吓一跳的角度说,是一种偏分析哲学的视角,也就是呃,只去看知识本身,我是说知识的动态、知识的分布,也就是今天我们会称之为,呃,认识逻辑那一路的进路。于是你开始看到一块块拼图,然后突然之间
便签引用
6:45
of sudden I realize actually this thing is not called knowledge because it's not owned by it's called information if you start talking about something out there that I may or may not have I may or may not send I may or may not sort of U shape that Stu there of which knowledge is made uh I thought this is really easy information I and I click with uh the last element in this conversation which is a personal par am I allowed yeah please please originally a long long long time ago but we talk about Lano in his teenager because we studed philosophy in high school my primary interest was philosophy religion uh Kor but you wouldn't yeah exactly so um but what I was interested in in the first year of my undergrad studies the reason why I wanted to study more about logic and logic and communication the so on I was interested in the um in how things that today I'm no longer sort of able to say without smiling um how God can send a message to humanity yeah the angel not the Angelos know the the messenger Etc right
我意识到,这东西其实不该叫知识,因为它不为谁所拥有,它该叫信息。如果你开始谈论某种在那儿的东西——我可能有也可能没有,我可能发送也可能不发送,我可能塑造也可能不塑造——那种构成知识的原料,呃,我想这不就是信息嘛。然后这就和这段故事里的最后一个要素接上了,那是一段很私人的部分——我可以讲吗?可以可以,请讲。很久很久以前,我们说的是卢西亚诺的少年时代,因为我们高中就学哲学,我最主要的兴趣是宗教哲学,呃,克尔凯郭尔。你大概想不到吧?对,就是这样。呃,但我在本科第一年真正感兴趣的、我之所以想更多地去学逻辑、逻辑与传播这些东西的原因,是我对——今天我已经没法一本正经地说出来、说了自己都要笑的那种问题——感兴趣:上帝怎么把讯息传给人类。对,天使,不对,是 angelos,你知道的,信使等等。
便签引用
8:03
and uh it's so I was interested in understanding from a Canon perspective Believe It or Not uh God's communication with Humanity not thinking God is a source humanity is the receiver and there is a channel how this channel could work if it works at all in order to sort of present a message um the the know the good message you know the the New Testament the uh it has the uh AOS has the same roote as yeah thean yeah so all that course classic studies to do for you H so put all this together and then now there's an interesting source and receiver that from philos religion Miracles then there's po purpose without KN subject then there's in Mo and anal philosophy there's the philology and so one day and that's that's the end of the story uh I was at at wolson College in oord uh I was a post doctor at that point and I was invited to give a talk in London open topic and I really wanted to do something about this stuff that we will be known afterwards as philosophy of information and I was with someone
呃,所以,信不信由你,我当时是想从一种正典的角度去理解上帝与人类的沟通,把上帝看作信源,人类是接收者,中间有一个信道,这个信道要怎么运作——如果它真能运作的话——才能传递一个讯息,呃,那个所谓的好消息,你知道的,新约。呃,福音(euangelion)跟天使(angelos)是同一个词根,对。所以那些古典学的功课都得做。呃,所以把这些放在一起,然后,这边有一个有意思的信源和接收者,来自宗教哲学、奇迹;那边有波普尔的《没有认识主体的认识论》;再有摩尔和分析哲学,还有语文学那一块。然后有一天——这就是故事的结尾了——呃,我在牛津的沃尔夫森学院,那时我是博士后,有人邀请我去伦敦做一个讲座,题目自定。我特别想讲一讲这些后来被称为信息哲学的东西。当时我和一个人在一起,那个人的名字我完全想不起来了,真是惭愧,这就是
便签引用
9:14
there someone I don't remember her name at all that shame on me that's the philosopher in typical like thly folk I must have bought that poor girl to death sitting on the on this uh little bench looking at the River wson College and I said you know what I could call this I could give a talk called what is the philosophy of information there is no section but I think we should have it and next thing I know they say oh yeah that's interesting why don't you come and give this talk what is it like what we can disc discover together it that's the beginning that's fantastic that's so cool my my audience will love that story too especially about kard or Kore we have a lot of uh kordians in the audience so that'll be fun it's it's really fascinating that you wanted to use Shannon information to talk about divine revelation I think that's that's really fascinating yeah and come yeah you've come a long way since then but um super soon later I'll go back to that question though that is my long long long journey at some point
典型的哲学家做派。我肯定把那个可怜的姑娘烦死了,我们就坐在那张小长椅上,看着沃尔夫森学院旁边的河,我说,你知道吗,我可以把这个叫做——我可以做一场讲座,题目就叫《什么是信息哲学》。虽然还没有这么个分支,但我觉得应该有。接着我就听说,噢,这挺有意思的,你干脆来讲这场吧,看看它到底是什么,看看我们能一起发现些什么。这就是开端。太棒了,这真是太酷了。我的听众也会很喜欢这个故事,尤其是关于克尔凯郭尔那段,我们观众里有不少克尔凯郭尔迷,所以那会很有意思。真的很迷人,你居然想用香农的信息论去谈神圣启示,我觉得这真的太有意思了。是啊,从那以后你走了很长一段路。不过,呃,等下我还会回到那个问题,那是我很长很长的一段旅程,到某个点上我会一路回到
便签引用
03失去信仰与灵性的不可知论
10:17
I would go back all the way to the question about God that is the question yeah um Dr FL uh maybe this is too personal but did have have you like um when you were talking or when you were thinking about that project were you were you a a believer in God yourself and and are are you not any longer and no longer H so when I I lost my faith roughly uh at that St during the first second year it is a four-year course and during the first second year I I realized that I had lost it okay it's not that the one day you wake up and say oh I don't have faith anymore but it it's like the foundations of a building sure it gets eroded and eroded and ered and one day Boom the whole team collapses but that was not the day when it really was under mind it was like a long long long process yeah I will say between being late teenager um uh by by the time I was 17 18 and I I had already almost lost entirely uh my fate having said that I didn't acquire another fate uh I never became an atheist okay I didn't want to exchange one belief for
关于上帝的那个问题,那才是问题所在。对。呃,弗洛里迪教授,也许这个问题太私人了,不过您——当您在谈论、或者说当您在构思那个课题的时候,您自己是一个信上帝的人吗?现在还是不是了呢?已经不是了。呃,我失去信仰大概就是在那个阶段,在第一、第二年——那是个四年制的课程——就在第一、第二年,我意识到自己已经失去它了。倒不是说某一天早上你醒来,说,噢,我没有信仰了,而是像一栋建筑的地基,当然,一点一点被侵蚀、被侵蚀、被侵蚀,然后有一天,砰,整栋楼塌了。但真正把它掏空的并不是那一天,那是一个很长很长的过程。我会说,大概是在十几岁后期,呃,到我十七八岁的时候,我的信仰几乎已经完全失去了。话虽如此,我并没有换上另一种信仰,呃,我从来没有变成无神论者。好吧,我不想拿一种信念去换
便签引用
11:32
another belief yeah um people know I mean uh you don't go around uh there's a famous phrase you know in philosophical science there uh absence of evidence is not evidence of sound absence right and you know if you're rigorous enough and serious enough you should acknowledge that and and yeah we may never know so I I describe myself today as a um spiritual agnostic okay that's as close as he gets yeah that's fascinating well yeah thanks thanks for that personal detail um that's that's really fascinating I I hate to do this to you because everyone does but you kind of since you invented the philosophy of information it comes with a territory can you help us out like what what is information there's a bunch of different ways to categorize it is there one Central core that that picks out you know what information is across different uh definitions well first of all let's say uh it's a mess that's the number one okay everybody should start from there because um I've seen people talking about information as if um using
另一种信念。对,呃,大家都知道,我是说,呃,你不能到处宣称——有句在哲学和科学里很有名的话,呃,没有证据并不等于有证据表明它不存在,对吧。而且你知道,如果你足够严谨、足够认真,你就应该承认这一点,而且,是啊,我们也许永远不会知道。所以我今天把自己描述为,呃,一个有灵性的不可知论者。好,也就到此为止了。对,这真有意思。嗯,是啊,谢谢您分享这么私人的细节,呃,这真的很迷人。我很不想这样问您,因为每个人都会问,但既然某种程度上是您发明了信息哲学,这也算是这份工作自带的,您能帮我们理清一下吗,信息到底是什么?给它分类的方式有很多种,是否存在一个核心的东西,能在不同的定义之间抓住信息之所以为信息的那一点?嗯,首先,我们得说,这是一团乱麻,这是第一点,好吧,每个人都该从这里出发。因为,呃,我见过有人谈论信息的时候,呃,大多数时候,尤其是在哲学里,用这个词的方式,就好像我们是那种
便签引用
04信息的三种含义
12:44
it most of the time especially in philosophy uh as if we were like the most obvious tool and therefore everybody understands me and and then that is where the confusion starts because of course there are many ways in which we understand what information uh is is um they have in common um normally uh something um and I'll tell you more in a moment let me start from uh since you mentioned the little book uh I don't feel guilty in telling people that it's all it's all in that little book yeah uh the very short into um so imagine you're talking about uh information in the world and it could be um say the um traces um left by an animal in the Forest uh it could the usual rings of a tree Cuts telling you the age of that tree so that is um shall we say the ontological conception of information yeah and then when we speak about that we might be even talking about no since there's there's been a Revival uh of LPS now uh the trace that is read uh by the PIN to be producing music I mean that is also a physical trace and you can point
最明显的工具,所以人人都能懂我的意思——可麻烦也正是从这儿开始的,因为我们理解“信息”的方式其实有很多种我们对信息的理解方式有很多种,它们通常有某些共同点,我待会儿会再多讲一些稍等一下——既然你提到了那本小书,我倒也不觉得不好意思告诉大家:其实全都在那本小书里了,就是那本《牛津通识读本》。设想你在谈论这个世界里的信息,它可以是,比方说,动物在森林里留下的痕迹,也可以是树桩上的年轮,它告诉你这棵树的年龄。这就是我们所说的信息的本体论概念。那么当我们谈论这个的时候我们甚至可能是在说——既然现在黑胶唱片又复兴了——被唱针读取的那道沟槽从而放出音乐,我是说,那也是一种物理痕迹,你可以指着它说:在某种意义上,这就是信息
便签引用
13:56
to it like that is information say in one sense what it's a that pattern so a lot of people often say well look what you what we have here is information because there's a pattern that and now we move to the a different concept they're all related okay yeah uh that pattern may have meaning it's not just a random pattern it's not like um say um clouds in the sky sure that pattern for example the the patter or on an LP the uh grave pattern read by the the pin or um my fingertip fingerprint sorry uh left on on a glass Etc yeah they have meaning they might have consequence Etc when we talk about uh meaningful patterns uh then we start talking about meaningful data and that is already entering into the realm of uh semantic information so you're no longer just talking about oh oh there's a particular structure or pattern out there um but you're talking about information about something so it's no longer information as as as something information about something that is the usual meaning that we tend
在这一层意义上,它就是那个“模式”。所以很多人常说:你看,我们这儿有的就是信息,因为这里有一个模式好,现在我们转到另一个概念上去了——它们都是相关的。那个模式可能是有意义的,它不只是一个随机的图案,不像天上的云那样。当然,那个模式,比如说,唱片上的那道沟纹,被唱针读取的纹路,或者我指尖的指纹——抱歉——留在玻璃杯上的痕迹等等,它们是有意义的它们可能有某种后果等等。当我们谈论有意义的模式时,我们就开始谈论有意义的数据了,而这已经进入语义信息的领域了。所以你不再只是在说“外面有某种特定的结构或模式”,而是在说关于某个东西的信息它不再是“作为某物的信息”,而是“关于某物的信息”,这也是我们平常谈到信息时
便签引用
15:13
to endorse if you're talking information but oh I want to have some information or information Des or I didn't get any information uh so our information these days is everything now you can't take decisions without information that's what we mean so content about something else so there's a reference and there's content content doesn't help if you don't describe that becomes data plus meaning plus syntax yeah so the data have a particular structure and they have a particular if you move all the way to uh adding also uh a true value that is false or or or not then you have one kind of semantic information which is the typical factual semantic information that you find in Wikipedia frame table sort of uh information so there's a Third Kind which doesn't have a truth value which is no which has a meaning um has a structure semantic and syntax but doesn't have any truth value and is uh all the Third Kind information that we uh describe as information for something Imagine music files uh uh I send you a
倾向于采用的含义:“哦,我想要一些信息”,或者“信息台”,或者“我没收到任何信息”。所以如今信息就是一切,你没有信息就没法做决定,这就是我们的意思。所以是关于别的什么东西的内容,于是有指称,也有内容;如果你不加描述,内容本身是没用的那样一来它就成了数据加意义加句法。所以数据有特定的结构,也有特定的含义。如果你再往前走一步,还加上真值,也就是真或假,那你就得到了一种语义信息,也就是典型的事实性语义信息,就是你在维基百科信息框里找到的那类信息。然后还有第三种,它没有真值,它有意义有结构、有语义也有句法,但没有任何真值,这就是我们所说的第三类信息我们把它描述为“为了某事的信息”。想象一下音乐文件:我给你发一个文件,在某种意义上我给了你一些信息
便签引用
16:30
file I send you some information in one sense a certain amount of data which have particular uh say meaning particular structure my bad being zipped or not Etc but then you open it and does he play music well that is information for something in order to play music right is it a recipe for a cake okay that is information for something is an algorithm you get a gist but it's not necessarily vertical or not you wouldn't say is this is this a true true or false right right so uh people we use true meaning genuine adep but not philosophy oh is it is it really your your your true Grandma re is that right genuine one yes if we are careful so information as something in the world information about something semantic factual and information for something which is more the algorithm kind once you have these three there's a that's a big step forward in terms of understanding what we're talking yeah I I love that yeah and and some other so information in the book it's information as reality information for information
也就是一定量的数据,它们有特定的含义、特定的结构,压缩过或没压缩等等;但接着你把它打开它就能放音乐——那就是“为了某事的信息”,为了放出音乐。对吧?或者是一个蛋糕的食谱,好那也是“为了某事的信息”,是一种算法。你大致明白了吧?但它不一定有真值性,你不会说“这个是真的还是假的”。对,对。所以我们说“true”是“真正的、地道的”那个意思,不是哲学意义上的:“这真的是你你真正的奶奶吗?”——对吧?是那种“货真价实”的意思。是的,如果我们仔细区分的话:作为世界中某物的信息关于某物的信息,也就是语义的、事实性的信息;以及为了某事的信息,也就是更接近算法的那一类。一旦你有了这三种区分,在理解我们在谈什么这件事上,就是一大步了。对,我很喜欢这个划分。还有别的一些,比如书里说的:作为实在的信息、为了……的信息、关于实在的信息,或者说面向世界的信息
便签引用
05年轮、界面与比特:信息在哪里
17:35
about reality um or for for the world I I love this distinction it's been really helpful um for me in thinking about um get getting clear because a lot of people will conflate these uh especially on the more popular level and then say like well everything just is information but they mean information about reality it's that that type of information everything is that um I wonder if all information I don't know another word to say except for like intentionality the information about reality has a particular intentionality it's this information is about whatever thing is out there in the world maybe it's like representational when I think of information as reality and I think about either fingerprints or rings on a tree is there in an intentionality to that type of information like the Rings in the in the tree is that about the age of the tree or anything no not really there's a there's a correlation that we can exploit okay so we can correlate uh some uh something in the world say a has a particular feature
我很喜欢这个区分,它对我理清思路真的很有帮助,因为很多人会把这几种混为一谈,尤其是在比较大众的层面上,然后就说“万物皆信息”但他们其实指的是关于实在的信息,是那一类信息,说一切都是那个。我在想,是不是所有信息——我想不出别的词,只能说“意向性”——“关于实在的信息”具有某种特定的意向性,就是说这个信息是关于世界上某个东西的,也许有点像表征。而当我想到“作为实在的信息”,想到指纹或者树的年轮时这类信息里有意向性吗?比如树里的年轮,它是“关于”树的年龄的吗或者关于什么别的?不,不太是。这里存在的是一种我们可以加以利用的相关性。也就是说,我们可以把世界上的某个东西关联起来:比如说 A 具有某个特征 F,而这个特征与另一个东西 B 具有另一个特征
便签引用
18:40
F and uh uh which is correlated to something else B having another feature G say and so if you um that particular the tree cut has not that feature those Rings which is correlated to the Cent Tre having a particular age but that correlation um is what we exploit in terms of then acquiring the information about the tree it would be a little bit funny but let me put it this way uh to say oh the tree meant to send us a message imagine you have some unknown ancient Lang that you discover on an island un Island tomorrow and a dead civilization but we find signs that have clearly a structure that is the first Hallmark of information yeah thing repeats it has a regularity things come with a syntax you see the syntax you actually see the the the patterns you you start thinking these patterns are not natural they have been put there by someone you iner that there must be a meaning a semantics they're sending us a message or they are sending a message they're broadcasting a message you need to crack that particular s of uh code
G 相关联。所以说,那个树桩具有那些年轮这一特征,而这一特征与那棵树具有某个年龄相关联。但正是这种相关性,是我们用来获取关于这棵树的信息的东西。这么说会有点滑稽——让我这么讲吧——要是说“哦,这棵树是想给我们发一条消息”。想象一下,明天你在一座岛上发现了某种未知的古代语言,一个已经消亡的文明,我们找到了一些明显有结构的符号——结构是信息的第一个标志。对,东西会重复,有规律性,带着句法。你看到了句法,你确实看到了那些模式你开始想:这些模式不是自然形成的,是有人放在那儿的。你推断这背后一定有意义、有语义,他们在给我们发消息,或者说他们在发出消息、在广播某条消息,你需要破解那套密码。对吧?我们得小心,有时候我们会太过比喻化,说“哦
便签引用
19:54
right uh we need to be careful sometimes we be too metaphorical um and say oh the the this the Rings in the tree is a is a code for metaphors let's be careful uh the tree doesn't say anything doesn't want to tell us anything um this is also useful when it when it comes to AI uh no matter what you get it is still patterns AI didn't mean to say anything didn't did have a plan did have an intention uh it just that no with all those correlations um on this side you can tell the age of the tree uh but don't confuse know the the tree and its intentionality we what we do with that pattern that's that's really helpful so um in in diving down on that point a little bit deeper so um I posted about having a conversation with you and I just mentioned um environmental information kind of to Jin up uh interest in the in the conversation in a philosopher of science uh was was uh I would say she was triggered she she was likew you dendrochronologists uh have to do so much work to derive the information out
树的年轮是一种密码”——那是比喻,我们得小心:树什么也没说,它并不想告诉我们任何事。这一点在谈到人工智能时也很有用:不管你得到什么,那仍然只是模式AI 并不打算说什么,它没有计划,也没有意图。只是说,凭着所有那些相关性,你可以推断出树的年龄;但别搞混了:树本身和所谓的意向性,跟我们拿那个模式做了什么,是两回事。这真的很有启发。那么就这一点再往深挖一点:我之前发帖说要和你做一次对话,我顺口提了一下“环境信息”,算是给这场对话预热一下热度,结果一位科学哲学家——我该说她被戳中了。她说:树木年轮学家得做大量工作,才能从树的年轮里提取出
便签引用
21:05
of the tree rings and so tree rings are not identical to information and in my head I'm thinking you know the the whole I go in for John surel's uh Observer relativity argument so I I like that I think there needs to be an observer to be to be able to interpret the information like you said there's no the tree doesn't have its own intentionality but it seems like the whole idea of dendr chronology of finding the age of a tree from free rings presupposes that there is information in there to be correlated at least right uh you you could put it the other way around that it goes around until we find something that is s that the correlation helps us to do it I mean it might have done not science fiction it might turn out that no those those Rings don't tell anything about the age I mean there is something of course about those all the identity of the individual I mean um with the fingerprints um the Tre had that Str structure because by leing in a particular environment Etc accumulates right those
信息,所以年轮并不等同于信息。而我心里想的是,你知道的,那整套我是接受约翰·塞尔的“观察者相对性”论证的,所以我喜欢这个说法,我认为得有一个观察者才能就是能够解读那些信息,就像你说的,树本身并没有它自己的意向性但整个树轮年代学的思路——从年轮推断树的年龄——似乎预设了那里面确实有可供关联的信息,对吧,呃,你也可以反过来说,就是一直试下去,直到我们找到某种东西,让这种关联能帮我们做到这件事,我是说,这也不算科幻,事情也可能是另一种结果:那些年轮根本说明不了年龄。当然,那里头确实有些东西关乎这个个体的身份,我是说,呃,就像指纹一样,呃,树之所以有那种结构,是因为它长在某个特定的环境里等等,慢慢积累,对吧,那些特定的特征缓缓沉积下来,变成它那一圈圈特定的年轮,所以年龄和
便签引用
22:02
particular feature that are slowly sedimenting into its sort of uh particular rings so there is a direct correlation between age and time if you like of growth rings no growing and us being able to as well reverse engineer the Rings not back to the age right so there not magic there's nothing strange but we could imagine for example it's a little bit like reading in archaeology reading uh a particular site and the strata of the the archaeological strata and say no this belongs to say this particular age of humanity and then I know it's it's a like many thousand years old um Set uh and this was like 10 10,000 years then is 6,000 years well why well because those are the the artifacts those are are the particular say geological um events that enable us to date so there is something there that helps us to acquire the information about the it is tricky and that's why you know we find so much confusion because people because we use uh um information as reality uh almost as if the information
生长年轮的时间之间存在直接的关联,对,它的生长,和我们能够反过来去逆向推导这些年轮、回推到年龄,对吧,所以这里没有魔法,没有什么怪事,但我们可以设想,比如说,这有点像考古学里的解读,解读某个遗址,解读那些地层,考古地层然后说,这属于人类的某个特定时期,然后我就知道,它有好几千年那么久,呃,比如说这个是一万一万年,那个是六千年。那为什么呢?因为那些器物、那些特定的比如地质事件,让我们能够断代,所以那里确实有某种东西帮我们获取关于它的信息。这事很微妙,这也正是为什么,你知道,我们看到那么多混乱,因为人们,因为我们把信息当作实在,呃,几乎好像信息就在那儿,只等着被发现,是的,好像你是在从那里
便签引用
23:13
were there and just needed to be discovered yeah as if you were digging out the information from there would confuse this with um imagine a different example uh a radio sending a signal from uh imagine we receive a signal a radio signal from another galaxy has been traveling I don't know for million of years and their signalization is there I mean that could there no science fiction I that could happen turn way more than other things we cannot travel faster than light but we could receive one day that would be astonishing I me it would be a change in human history but imagine that we get that radio signal um that reconstruction is completely different that is not reading so ring trees or finding my finger prints on uh on a glass someone somewhere broadcast that and so there's a then we're back to to the real thing know to the source the message and the receiver that is much more Shannon but this the tree is not sending anything we are able to interpret that particular pattern in such a way that we
把信息挖出来。人们会把这个跟另一种情况搞混,呃,设想一个不同的例子,一台无线电发出信号,来自呃,设想我们收到一个信号,一个来自另一个星系的无线电信号,已经传播了,我不知道,几百万年,而他们的信号编码就在那里,我是说,这也不是科幻,这是可能发生的,比起别的事情反而更有可能,我们没法超光速旅行,但我们某天可能收到这样的信号,那会非常震撼,我是说那将是人类历史的一次改变。但设想我们收到了那个无线电信号,呃,那种重建是完全不同的,那不是解读树的年轮,也不是在玻璃杯上找到我的指纹,而是某个地方有人发出了广播,所以就有了于是我们又回到真正的那套东西了,回到信源、信息和接收者,那更接近香农的模型。但树并没有在发送任何东西,是我们有能力以某种方式解读那个特定的模式,让我们可以利用这个模式
便签引用
24:23
can soort of Leverage the pattern to reconstruct back towards the age of the tree or for example the presence of say someone on a CRI scene because of the to some extent uh the fingerprints were there and then of course shell Holmes shows that no he wasn't there was someone else someone put the glass there the guy Etc ETA well Dr flurty um I wonder if the situation changes uh if uh if I leave my fingerprint on a glass with the intention of leaving my unique fingerprint so maybe maybe I'm kidnapp napped but I I I press it on there so that someone can find this and say oh Parker's been here does that change the um information from like environmental to uh I don't know like like analog or to does it have an intentionality now because I've intended to communicate through my unique fingerprint yeah so um what we have changed uh and I think you find that also in the the book uh we have change if you like the interface through which we're looking at that piece of information uh in computer science
反推回树的年龄,或者比如说,反推某人曾出现在犯罪现场,因为在某种程度上,呃,指纹确实在那儿。当然,福尔摩斯会告诉你,不,他并不在场,是另有其人,是有人把那只杯子放在那儿的那家伙,等等等等。好的,弗洛里迪教授,呃,我想知道情况会不会变,呃,如果,呃,如果我是有意把指纹留在杯子上,故意留下我独一无二的指纹,比如说,也许我被绑架了,但我把手按在上面,好让有人能发现它,然后说,哦,帕克来过这里。这会不会把这个信息从环境性的,变成,呃,我不知道,变成类似模拟的,或者说,它现在是不是有了意向性,因为我意图通过我独特的指纹去传达什么?是的,呃,我们改变的是——我想你在书里也能看到这一点——呃,我们改变的,如果你愿意这么说的话,是我们看待那条信息所透过的界面,呃,在计算机科学里
便签引用
25:27
called level of attraction but we don't have to be too fancy I mean we can call it interface so if you're looking for example at a with an interface of the police looking for someone who might have been kidnapped uh or say uh I I'm lost in a forest and I start leaving know uh pieces of my clothes here here here here yeah imagine that scenario versus uh I'm in a forest and I'm losing bits of my stuff now my my let me let me do like more more um intelligent my backpack uh I have a backpack there's a hole I'm losing pieces in a forest I'm in a forest I'm lost I'm trying to be found so I'm intentionally leaving pieces in the forest from an interface that doesn't have why am I asking this question for doesn't have a purpose the two phenomena are identical essentially all I know is that someone is in the park and is leaving and there that remnants bits of the backpack left is he meant to be found is he lost that just have a hole in the backpack and losing stuff as he goes that is intentional that is what we
叫做抽象层级,不过我们不必这么文绉绉,我们叫它界面就好。所以如果你是从警方的界面去看,警方在找一个可能被绑架的人,呃,或者说,呃,我在森林里迷路了,我开始沿路留下,呃,我衣服的碎片,这儿一片那儿一片,是的,设想这个场景对比另一种,呃,我在森林里,东西一点点掉出来。让我,让我说得更,更聪明一点:我的背包,呃,我有一个背包,上面破了个洞,我在森林里一路掉东西。第一种是我在森林里迷路了,我想被人找到,所以我有意地在森林里留下东西。而从一个不具备——我为什么要问这个问题呢——不具备目的的界面来看,这两种现象本质上是一模一样的。我所知道的全部就是,有个人在这片林子里,一路留下背包掉出来的残片。他是想被找到吗?他是迷路了?还是他背包破了个洞,边走边掉东西?那是有意的吗?这就是我们在解读时所做的事,呃,这本身是好事,呃,但我们得有能力
便签引用
26:39
do in terms of interpretation um which is good uh but we need to be able to distinguish this from what do we have in front of us now sometimes and I hope I'm I'm not confusing things too much now in philosophy science we speak of how data underdetermine the theory yeah in other words you can have many many data points and you can reconstruct sometimes those data points in more than one way often well that's the same story about the me losing bits of stuff in in in the park uh am I try to be found or or or just have a hole in my backpack yes so of course if the police thinks that I'm lost Etc everybody will interpret that as an attempt to be found but it might be just an accident I just got lucky and without me knowing uh I left the trace and people find me fantastic that that's helpful so um so I understand I think I I think I understand information if you if you want to analyze it in terms of uh data and meaning because then we can go in on Data and we can say well data are made of uh bits which is just like one unit
把它和我们眼前实际有的东西区分开。有时候——我希望我没把事情讲得太乱——在科学哲学里,我们会说数据对理论是欠决定的,对吧,换句话说,你可以有很多很多数据点而这些数据点有时可以有不止一种重建方式,而且常常如此。这跟我在林子里一路掉东西是同一个故事,呃,我是想被找到,还是只是背包破了个洞?是的,所以当然,如果警方认为我迷路了等等,所有人都会把那解读成一次求救的尝试,但那也可能纯属意外。我只是走运,在我自己不知情的情况下,呃,我留下了痕迹,人们找到了我,太好了,这很有帮助。所以呃,所以我理解,我想我,我想我能理解信息——如果你愿意把它分析为数据加意义的话,因为那样我们就可以深入谈数据,我们可以说,数据是由比特构成的,比特就像一个单位
便签引用
27:46
and I understand that in the binary uh conversation I'm not sure I understand it in a physical um information as reality understanding what what would a what would a bit be in the physical world yeah no well well first of all the the bit normally let again unfortunate um difficult word to to use um can either be uh part of the code we use in order to no code some information so imagine like the telegraph no line doline do line so is the world is is the Hamlet made of do no of course not but we can transform the whole Hamlet into and this on the other side of the Atlantic you know good old days so the bit is that is a code or very rarely is something you find in the world I mean the world doesn't tend to be binary the the way the way we experience it uh we normally experience it as a continer analog as in there's always something in between something else um uh there was um this famous uh phrase nature doesn't make jumps uh H well the fact that it doesn't make jumps means you know if you translate into 21st
我在二进制的语境里能理解这一点。但我不太确定我能在物理层面理解它,呃,把信息当作实在来理解——在物理世界里,一个比特会是什么?是的,呃,好,好,首先比特这个词,通常,唉,又是一个很不巧的、很难用的词,呃,它要么是我们所使用的编码的一部分,用来给某些信息编码。所以设想电报,对吧,滴、划、滴、划、滴、划。那么世界是不是,哈姆雷特是不是由滴和划构成的?当然不是。但我们可以把整部《哈姆雷特》转换成这个,然后传到大西洋另一边,你知道,那些美好的旧时光。所以比特要么是一种编码,要么极少数情况下是你在世界中找到的东西——我是说,世界并不倾向于是二进制的,我们体验它的方式,呃,我们通常体验到的是连续的、模拟的,也就是说两个东西之间总还有别的东西。呃,有一句很有名的话:自然不作跳跃。呃,嗯,它不作跳跃这件事意味着,你知道,翻译成21世纪的说法就是:自然不是二进制的,呃,它不是从0跳到1,或者从
便签引用
29:06
century nature is not binary um it doesn't move from zero to one or from yes no highle so uh current electricity in the circuit hole or no hole on a CB ET so if we are not talking about the code but we talking about the sort of information if you like in reality out there as a reality that normally is analog uh imagine a map okay so you have a map that map is is continuous uh it's own line lines and say the the underground map so that you can travel from one point to the next um the extraordinary Revolution that we have undergone uh roughly you if you want to have a hero we can become volunteering but no it comes earlier by but even live Hobs yeah is that we found a way of encoding every kind of information in the world waves including so continum into discrete uh elements that's your code and we have machines that can handle the code automatically in fact today we have machines that not only can handle the code automatically but they can do so by learning from the app NS not AI there Evolution has been
是跳到否、从高电平到低电平,呃,电路里有没有电流、电路板上有没有打孔等等。所以如果我们谈的不是编码,而是谈那种,如果你愿意这么说的话,外部实在中的信息,作为一种实在,那它通常是模拟的。呃,设想一张地图,好,你有一张地图,那张地图是连续的,呃,它有自己的线条,比如说地铁线路图,让你能从一个点走到下一个点。呃,我们所经历的那场非凡革命,呃大致上——如果你想找个英雄,我们可以说是布尔,不过其实更早,甚至莱布尼茨——是我们找到了一种办法,把世界上各种各样的信息,包括波,也就是连续统,编码成离散的元素,那就是编码;而我们有了机器可以自动处理这些编码。事实上今天我们有的机器不仅能自动处理编码还能通过从大量样本中学习来做到这一点,这就是AI。这个演进过程非常了不起,因为到了那一步,呃
便签引用
06AI:能动性与智能的解耦
30:28
extraordinary because at that point uh in a few steps which are historically amazing you have from the analog continuous world to the digital binary world to a machine that can handle that uh efficiently more and more and more to the extent that that machine through statistics can also find the fundamental patterns in the original analog imagine English all the English spoken and start digging out from the deepest so well of structures in that language patterns that can be reused for us to do useful work for example I don't know translate this into that well that is no a journey that has been almost uh a miracle now if you pass me there so dangerous analogy um to me that is really extraordinary uh Journey uh from so say from the first computer to uh CH GPT yeah thinking about uh oh we're creating some kind of form of intelligence is it's almost like not getting how big the transformation is it's it's almost like saying like you are beling the extraordinary human Ingenuity that went into all these
只用了几步——从历史尺度看这实在惊人——你就从模拟的、连续的世界,走到了数字的、二进制的世界,再走到一台能够高效处理它的机器,而且越来越高效,以至于那台机器还能通过统计的方式,反过来在原本的模拟世界中找出根本性的模式。设想英语,所有说出来的英语,然后开始从最深处挖掘出那种语言中的结构、那些可以被复用来替我们干活的模式,比如说,我不知道,把这句话翻译成那句话。这,这是一段几乎堪称奇迹的旅程——如果你允许我用这么一个危险的类比的话。呃,对我来说,那真是一段非凡的,呃,旅程,呃,从,比如说,从第一台计算机到ChatGPT。是啊,而想着说,哦,我们在创造某种形式的智能,这几乎就是没搞懂这场转变有多大这几乎就等于是在,你是在贬低投入到所有这些步骤中的那种非凡的人类智慧,反而以为,你知道,只要
便签引用
31:56
steps and instead you think that know by clicking something that you not quite show these black boxes all of a sudden boom intelligence arises it's not the way it works that's amaz I've never thought of it that way I I was listening to you earlier and you talked about you now now everything clicks for me where you're saying um the thing about AI this from one of your lectures is that we no longer have to be we no longer have to use intelligence to park a car for uh you know uh self-driving cars or to play chess no longer and that is a really big deal it's it's you've outed intelligence and that's huge that shows how intelligent we are that we're able to do that but when you say no that thing has to be intelligent because it's doing what we do now you've actually belittled the achievement of the programmers and such that uh if you think in terms of um extraordinary remarkable powerful benign or not intelligence you belong to uh to the Past you're still a modern in a bad sense of modern uh there a good sense
点一下什么东西,那些你并不太看得明白的黑箱,突然之间,砰,智能就冒出来了。事情不是这么运作的。这太惊人了,我从来没这么想过。我刚才在听你讲,你谈到,现在,现在一切都对上了。你说的——关于AI,这是你某次讲座里的说法——就是我们不再需要我们不再需要动用智能去停一辆车,呃,你知道,靠自动驾驶汽车,或者去下国际象棋,不再需要了。这是件非常了不起的事这,你把智能从中剥离出去了,这太重要了,这恰恰显示了我们有多聪明,我们竟然能做到这一点。但当你说,不对,那个东西一定是有智能的,因为它在做我们做的事——这时你其实是在贬低程序员之类的人所取得的成就。是的,呃,如果你还用那种非凡的、了不起的、强大的、良性或非良性的智能的框架去思考,呃,你就属于过去了,你还是个现代人——是坏的那种意义上的现代,呃,也有好的那种意义
便签引用
33:00
but in a bad sense of modern thinker you're still you still stuck with Frankenstein okay the reproduction of ourselves one way or another they may or may not take over may beny may may not but the the Frankenstein uh the Golem me um that is not quite getting the way more me interesting Revolution which is extraordinary and know you one way of putting it as you said is to for the first time in human history we have been able to decouple successful agency and intelligence yeah that never happened before I mean we had never managed to do something successfully in view of a goal improving at zero intelligence so that's what I normally I tell students here said look imagine parking a car at zero intelligence don't do that with no with your parents C they will not be pleased oh imagine cooking at zero intelligence uh or doing the dishes at Z intelligence playing chance at Zer intelligence but warehousing at zero intelligence okay uh just or uh collecting strawberries I'm I'm actually referring to actual industrial
但在坏的那种意义上的现代思想者,你还,你还困在弗兰肯斯坦里,好吧,就是以某种方式复制我们自己,它们可能接管,也可能不接管,可能良善,也可能不良善。但弗兰肯斯坦、呃、魔像那一套,呃,那是没有真正抓住那场更有意思得多的革命,那场革命才是非凡的。而,你知道,用你刚才的说法之一,就是人类历史上第一次,我们能够把成功的能动性和智能解耦。是的,这在以前从未发生过。我是说,我们从来没能在零智能的情况下,朝着一个目标成功地做成某件事,并且越做越好。所以这就是我通常跟这里的学生说的:你看,设想在零智能的情况下停一辆车——别拿你父母的车试,他们不会高兴的。哦,设想在零智能的情况下做饭,呃,或者在零智能的情况下洗碗、在零智能的情况下下棋,或者在零智能的情况下管理仓库,好吧,呃,或者,呃,采草莓——我说的其实是真实的工业应用——在零智能的情况下造一辆
便签引用
34:12
applications building a car at zero intelligence impossible the mess the disaster the cost would be all over the place today we have as we speak not like oh one day no no for years now we have had machine means they can do that successfully and improving because of course they learn from their output uh to get uh their output as an input so that their performance gets better and better up to a point where we are not completely satisfied I mean the parking is perfect the no the strawberries are collected in the most efficient and effective way Etc yeah that divorce between agency and intelligence there is historical and allow me for this extra look into the future that is where the problems are the challenges the opportunities but also the problems now when we talk about the ethics of AI uh the real Point here is in that sort of gap between agency and intelligence that we are generating and increasing in that Gap you find all the difficulties the Privacy the the bias the copyright and intellectual property
车,不可能,那会一团糟,是场灾难,成本会失控。而今天,就在我们说话的当下,不是说哦,将来某一天,不不,已经好多年了,我们已经有机器能够成功地做到这些,并且越做越好,因为当然,它们会从自己的输出中学习,呃,把自己的输出当作输入,所以它们的表现变得越来越好,好到某种程度——虽然我们还不完全满意——我是说,停车是完美的,草莓也被以最高效、最有效的方式采下来了,等等。是的,能动性与智能的这场离婚,是具有历史意义的。请允许我再多往未来看一眼:问题、挑战、机遇——但也包括问题——都在那里。现在当我们谈AI伦理时,呃,真正的关键就在能动性与智能之间的那道我们正在制造并不断拉大的裂缝里。在那道裂缝里,你能找到所有的难题:隐私、偏见、版权和知识产权
便签引用
35:22
um the control uh over uh resources the exploitation um keep going not the fake uh deep fake ex it's all there in the fact that we can actually as for for the past few years we've been able to do things remarkably successfully more and more at zero intelligence that is the thing that we should be understanding much better and design much better yeah that's so fascinating I've I've been thinking of um I'm me in a class right now on the the philosophy of artificial intelligence and and I've been thinking about trying to come up with some kind of Distinction because I know that the the AI folks over on the computer science side of things they don't like the word artificial intelligence because they think it's genuine intelligence I think it's perfect it's a perfect word it is artificial it's it's a representation of the intentions of the programmer but I wonder I wonder if you would put um like booby traps in the same uh category as like a self-driving car like you know think back to maybe
呃,对资源的控制,呃,剥削,呃,还可以继续数下去,还有伪造,呃,深度伪造等等。这一切都源于这样一个事实:我们其实在过去这些年里,已经能够在零智能的情况下越来越成功、越来越出色地做成事情这才是我们应该更好地去理解、更好地去设计的东西。是的,这太引人入胜了。我一直在想,呃,我现在正在上一门关于人工智能哲学的课,我一直在想能不能提出某种区分,因为我知道,计算机科学那边搞AI的人他们不喜欢“人工智能”这个词,因为他们认为那是真正的智能。我觉得这个词好极了,是个完美的词——它就是人工的,它是程序员意图的一种呈现。但我想问,我想问你会不会把,呃,比如陷阱机关,跟自动驾驶汽车归到同一类,比如说,你想想,也许是
便签引用
07第四种能动者:河流、狗、人之外
36:25
the Bronze Age booby trap someone steps on this uh something and a dart shoots into their chest uh is that is that still agency without intelligence in the same way just in more rudimentary form or no no so uh this is also another topic where we need a few distinctions but um the kind of artificial intelligence that we have uh and uh and we have had it for a little while but this one is no the sort of statistical kind as opposed to the not deductive kind that we had in the past it's way more power um is the ability to learn that is the kind of agency talking about that's right the B trap doesn't learn it's the same way in which an earthquake is a form of agency but not the kind that we're talking about because the an earthquake doesn't learn it's not that today say oh no I'm really an evil force I want to do more damage next time I do a better job like I wait not for everybody to sleep and then I go of course not but if you think of um imagine uh forms of agency and this is for anyone who's listening this is
青铜时代的陷阱:有人踩到某个东西,一支飞镖就射进他的胸口。呃,这算不算,这算不算同样意义上的、没有智能的能动性,只是形式更原始?不不,呃,这也是另一个需要做几点区分的话题。不过,呃,我们如今拥有的这种人工智能,呃,其实我们已经拥有一阵子了,但这一种是,呃,统计式的那种,而不是我们过去那种演绎式的,它要强大得多,呃,关键在于学习的能力,那才是我们在讲的那种能动性。没错,陷阱机关不会学习,这就跟地震是一种能动性、但不是我们在讲的那一种,是同一个道理,因为地震不会学习,它不会说今天,哦不,我可是个邪恶的力量,我下次要造成更大的破坏,我要干得更漂亮,比如我要等所有人都睡着了再动手。当然不会。但如果你设想,呃,设想各种形式的能动性——这里要提醒所有正在听的人,接下来这套说法
便签引用
37:26
completely unorthodox uh don't try this at home um and no I've been criticized in fact one of the reviewers of the anties of AI uh the last before up when I make this distinct them said that this was complete Bonkers so luckily luckily I had enough replies to say that they hadn't quite understood what I was talking about but here we go so an ala imagine agency coming in degrees okay something very very poor no no like low degree High degree the highest degree uh is probably the one that is omnipotent imagine God that can do anything learn anything know anything but it TR God for a moment uh for no and on this planet what we have is um interaction with the world if you make a difference in the world then you are already an agent MH sometimes we talk about atmospheric agents for their reason yeah rain earthquake a river a river that not goes from the mountain all the way to the to the sea is an agent uh for good or bad because because it has one condition the minimal condition makes a difference in
完全不合正统,呃,别在家里模仿。呃,而且,我确实挨过批评,事实上《AI伦理学》的一位审稿人呃,就是在我做这个区分的那一部分之前,说这完全是胡扯。所以幸好,幸好我有足够的回应可以说明,他们并没有完全理解我在讲什么。好,那我们开始。所以,呃设想能动性是有程度之分的,好吧,有非常非常弱的,不不,比如低程度、高程度,而最高的程度,呃,大概就是全能的那种,设想上帝,能做任何事、学任何事、知道任何事。不过我们先把上帝放一边,呃,对,那么在这个星球上,我们拥有的是,呃,与世界的互动:如果你在世界中造成了差别,那你就已经是一个行动者了。嗯有时候我们会说“大气营力”,就是这个道理,对吧,雨、地震、河流——一条从山上一路流到海里的河流,就是一个行动者,无论好坏,因为,因为它满足了一个条件,那个最低条件:造成差别
便签引用
38:41
the world change so intera then you have good old days uh something in between uh that are your average animal now say a dog a dog can learn it can also have intentions you know it means to survive for example um he will stay away from uh danger um he will not like another bigger dog uh coming so there's there's both learning and let's call it intentionality okay and of course the highest degree of that is the human being where there is all this and the ability to overcome all this not only all this but also say the conscious ability to control all this and say you know what I want to commit suicide the dog will not uh or you know what I don't have to be afraid of something um or you know I can keep my hand uh my finger for 30 seconds on the flame of a candle how many stupid jokes have we done when we we kids that because I want to yeah ad do so that kind of conscious control that comes as an extra form of agency MH now these were the three now the river the dog and me and that's life on on planet Earth since we came out of
世界的变化如此互动,然后是过去那些老好时光,呃,介于两者之间的,就是一般的动物,比如说狗,狗是能学习的,它也可以有意图,你知道,比如说是为了活下去,呃,它会躲开危险,呃,它不会喜欢另一只更大的狗,呃,靠过来,所以这里既有学习,也有我们姑且称之为意向性的东西,好当然,这方面程度最高的就是人类,人身上有这一切,还有超越这一切的能力不只是这一切,还有,比方说,有意识地控制这一切的能力,可以说,你知道吗,我想去自杀。狗不会这么干,呃;或者说,你知道吗,我不必害怕某样东西,呃;再或者,我可以把手,呃,把我的手指在蜡烛的火焰上放上 30 秒,我们小时候干过多少这种蠢事,就因为我想对,就是要这么干。所以那种有意识的控制,是一种额外的能动性形式。嗯,刚才说的是这三种河流、狗,还有我,这就是自从我们走出洞穴以来地球上的生命图景。而今天出现了第四个角色
便签引用
39:59
the caves today there is a fourth player something that is not like the river only because he's interactive but he can learn it's not like the dog because it doesn't have intention it's not like I'm going to say oh no uh I don't feel like uh eating this today uh and then oh I bark because I want to go out uh for a walk okay so um it doesn't have that intentionality and certainly doesn't have the conscious self-control planning uh in two years I want to go and know explore the Amazonian R exam so there is a Fourth Kind of agency here which we have encountered in the past for which we don't have conceptual tools up to it what happens that some people will think of this just as another form of agency interact it's a tool it's no more than say a drill okay it's not more than a car the river kind of agency it does what it does our controls bad idea at least my old car no the new one learns but the old one didn't didn't learn it said I said oh no it's a new form it's almost like a dog is growing
一种东西,它不像河流,只因为它是交互的,但它能学习;它也不像狗,因为它没有意图,它不会说,哦不,呃,我今天不想吃这个,呃,然后,哦我叫两声是因为我想出去,呃,去散个步。好,所以,呃,它没有那种意向性,当然也没有那种有意识的自我控制和规划——呃,两年后我想去探索亚马逊雨林。所以这里存在第四种能动性,是我们过去从未遇到过的,我们还没有足以把握它的概念工具。结果就是,有些人会把它仅仅当作另一种能动性,交互式的,它是个工具,不过就像一把电钻罢了,好吧,不过就像一辆车,河流那种能动性,它做它该做的事,我们来控制——这想法不好;至少我那辆老车是这样,新车会学习,可老车不会,不会学习。也有人说,哦不,这是一种新形式,它几乎像一条正在长大的狗,它有那个……它就像一个五
便签引用
41:13
it has the the it's it's like a five year old child Etc the technology the science we have should be good enough to tell us that that at least admit that this is not the case today I think it's bongus anyway but no sure we're talking about my iPhone okay uh or we're talking about chat GPT it's certainly not like us so we need to admit that we have a new player uh that the game has become a little bit more complicated uh we build it we control it uh or at least we have the responsibility of controlling it we can turn it on and off um it can do things uh that are amazing you can learn from its own performance it's not like a re it's not like a dog it's not like a human being what we need is a bit of work from philosophers understanding this new form of agency that is the opening that I would find exciting uh any uh researcher listening to this I mean you have an amazing Topic in in your hands this is the 21st century if you try to reduce this form of agency to oh but really is some kind of
岁小孩等等。我们拥有的技术和科学,应该足以告诉我们,至少让我们承认,事实并非如此。反正我觉得今天这些说法都是胡扯,但当然啦,我们说的是我的 iPhone,好吧,呃,或者我们说的是 ChatGPT,它肯定不像我们。所以我们需要承认,我们有了一个新玩家,呃,这个局面变得有点更复杂了,呃,是我们造出它、我们控制它,呃,或者至少我们有控制它的责任,我们可以把它打开、关掉,嗯,它能做一些,呃,很惊人的事,它能从自己的表现中学习,它不像河流它不像狗,也不像人。我们需要的是哲学家做点工作,去理解这种新的能动性形式,那才是我觉得真正激动人心的突破口,呃,任何听到这段话的研究者,我是说,你们手上有个了不起的题目这就是 21 世纪。如果你非要把这种能动性还原成,哦但其实它就是某种智能,不是的,不是这样,你只是
便签引用
42:18
intelligent not just not by you just stuck with a no I don't want to admit that something entirely new has really happened I have an old model my hermeneutics is still modern and all I have are these three categories so it has to be one two or three since it's not a river and it's not like a it must be all either like a dog or like a human being yeah open up think in terms of Welcome to a four kind um and it's defitely reasonable I mean we engineer this uh problem solving task uh uh caring um artifacts this this this um this even the question of agency so so tough to think about and like it it uh puts an exclamation point on your idea that we need to have better concepts for this this this new thing because I'm thinking maybe there's some literature on group agency and so is is chat gbt like a group agent because I'm interacting with I don't know 3.5 but there's also four what's the connection between those two and also if if I'm using it at the same time you are is this does it make sense to even call it
卡在那儿说,不,我不想承认真的发生了某种全新的事情。我用的是旧模型,我的诠释学还停留在现代我手上只有这三个范畴,所以它必须是一、二或三。既然它不是河流,也不像……那它就必须要么像狗,要么像人。是啊,打开思路,按第四种来想吧。嗯,这确实很有道理,我是说,我们是工程化地造出这些,呃会解决问题的、呃,呃,会照料人的,嗯,人工物,这个这个这个,嗯,就连能动性这个问题本身都太难想清楚了,而且它,呃给你那个想法加上了一个惊叹号:我们需要更好的概念来把握这个这个这个新东西,因为我在想,也许有一些关于群体能动性的文献,那 ChatGPT 算不算一个群体行动者?因为我在跟,我也不知道,3.5 互动,但还有 4,这两者之间是什么关系?还有,如果我在用它的同时你也在用,这……把它叫作一个行动者到底还说得通吗?你明白我的意思吧,就是说,呃,我们不想说它是
便签引用
08智能无定义:群岛与国界
43:31
like an agent you know what I mean like uh we we don't want to say it's intelligent but I think the AI folks say well intelligence is a functional concept and it's just you know solving solving puzzles yeah so yeah so about intelligence at some point if it BEC just a matter of vocabulary I'm happy with any vocabulary I mean okay you could call it the genus as far as I'm concern I don't care um the question is what that um the consequences of that um choice of vocabulary because um for example if then you start getting worried that he might take over the universe and do well that's silly okay but if I call say this pen intelligent because you know the more I use it the less ink I can see here that's very intelligent yeah okay fine so as of tomorrow not this the intelligent pen right yeah it doesn't matter yeah and a lot of that is is advertisement is hype is selling the product so let's let's be a bit more philosophical about it you want to be careful about the words because words have consequences they have
智能的,但我想搞 AI 的人会说,智能是个功能性概念,无非就是,你知道,解决解决难题。对,是啊,所以关于智能,说到某个程度,如果这只是个用词问题,那我用什么词都行,我是说,好吧,就我而言你叫它天才都可以,我不在乎,嗯,问题在于那个,嗯,那种用词选择带来的后果,因为,嗯,比方说,如果你接着开始担心它可能会接管整个宇宙,那,这就很傻了,好吧。但如果我说这支笔是智能的,因为你看,我用得越多这里能看到的墨水就越少,那可真智能啊,是吧,好,行,那从明天起它就不叫这个了,叫智能笔,对吧,是啊,这无所谓。是啊,而这里面很多都是广告、是炒作、是在卖产品,所以我们对此还是哲学一点吧。你得对用词小心,因为词是有后果的,它们带着各种含义。而智能,虽然我们并没有一个定义,而且顺便
便签引用
44:36
implications and intelligence although we don't have a definition and by the way for anyone curious just just look around and see how many disciplines have defined in how many ways yeah it's a joke yeah so um so the real question becomes oh if you don't have a a definition for intelligence uh what are we talking about well we do have some criteria and you want to know whether these criteria are here being uh deployed or not for example um stop playing a game because something else is more important all of a sudden you're playing chess with a computer uh computer is winning or you're winning it doesn't matter but at some point you actually stop the the fire alarm goes on go go go the computer will keep playing he will win the game and burn with the house yeah uh so intelligence for example is also a matter of flexibility there is social intelligence there's musical intelligence there's that intuitive intelligence that lets you sort of postpone some news to a friend because today is not the right
说一句,任何好奇的人只要四处看看,就会发现有多少学科用多少种方式定义过它,是啊,简直是个笑话。是啊,嗯,所以真正的问题变成了:哦,如果你没有智能的定义,呃,那我们到底在谈什么?其实我们确实有一些标准,你想知道的是这些标准在这里有没有,呃,被用上。比如说,嗯,停下不玩游戏了,因为别的事情更重要。你正跟电脑下国际象棋,呃,电脑在赢,或者你在赢,这都不重要,但到某个时刻你真的会停下来——火警响了,快走快走快走,而电脑会继续下,它会赢下这盘棋,然后和房子一起烧掉。是啊,呃,所以智能比方说也是一个灵活性的问题有社会智能,有音乐智能,还有那种直觉性的智能,让你懂得把某个消息先压一压、晚点再告诉朋友因为今天不是合适的日子。呃,所以通常人们问我,哦那么智能到底是什么,智能
便签引用
45:45
day uh so normally when people ask me oh well so what is intelligence intelligenc is all the times no is the counterpart of all the forms of stupidity that we can have to day if it was stupid no to leave the passport at home when you're flying it would have been intelligent to take the passport with you flying it would be stupid to leave the the finger in the door when you close the door of your car that was stupid yes so it would have been intelligent not to leave the finger that it would be stupid to think that two plus2 is 22 then he's intelligent to think that t plus is four if it was stupid to say that thing to that friend today really parket don't you know Laura I said or John is said terrible day you could have that was stupid it would have been intelligent not to you you read the newspaper you can't summarize it really that's stupid like intelligence not to read the newspaper but then you know the variety of things uh that we call intelligent is almost boundless because it's another way of
在所有情况下都是……不对,它是我们日常各种愚蠢的对立面。如果说坐飞机出门把护照落在家里是愚蠢的,那么带上护照去坐飞机就是智能的;那关车门的时候把手指夹在门里就是愚蠢的,对吧,那么不把手指留在那儿就是智能的;如果认为二加二等于二十二是愚蠢的,那么认为二加二等于四就是智能的如果今天对那个朋友说那句话是愚蠢的,真的,天哪,你不知道吗,我说了劳拉怎样怎样,或者说了约翰怎样怎样那天他过得很糟,你本可以不说的——那就是愚蠢,不说才是智能的;你读了报纸却没法把它总结出来,真的,那就很蠢,那还不如不读报纸来得聪明。但你知道,我们称之为智能的那些事情,呃,几乎是无穷无尽的,因为这只是换个说法在说“做人”,呃,说拥有一段成功的心智生活,而这不是
便签引用
46:53
saying Being Human uh having a a a successful mental life and that is not something you can pinpoint like a definition like water equal H2O because that is science and it's good science but that's when we have IR regimented vocabularies that can handle not concept that are clear and precise in this case we have an archipelago of islands uh you can call the acal males but nor iseland is the mald M and that would be silly it's the whole arip yeah and people say oh okay well kind of getting it um let me give you another analogy it's like people looking for the uh the border between Italy and France is it in France or is it in Italy it's a border yeah oh but it has to be somewhere no it's what makes one different from the other uh is that line that is neither here nor there because if you pass it you are in one place and you so you need to be a little bit more so uh in should we say intelligent flexible uh mentally and then uh the whole question is this machine intelligent or not becomes either trivial or mis formulated trivial
你能像下定义那样精确指出来的东西,比如水就是 H2O,因为那是科学,而且是好科学但那是在我们拥有规整化的词汇、能处理清晰而精确的概念的时候。而在眼下这种情况里我们面对的是一片群岛,呃,你可以把它们统称为群岛,但没有哪一座岛就是马尔代夫,那样想就太傻了这是整片'群岛',对,人们会说'哦,好吧,大概明白了',嗯,我再给你打个比方,就像是有人在找意大利和法国之间的那条国界——它到底是在法国境内还是在意大利境内?它是一条边界啊,对'但它总得在某个地方吧',不,它正是让这一边区别于那一边的东西,就是那条既不在这边也不在那边的线因为你一旦越过它,你就到了另一边,所以你得稍微更……我们该说更聪明、更灵活一点,在思维上然后'这台机器到底有没有智能'这个问题,要么变得毫无意义,要么就是问错了。说它毫无意义,是因为这取决于你所说的
便签引用
48:08
because like depends on what you mean by intelligence if for example he runs calculations well my pocket calculator is intelligent it can do maths like no one else I know amazing yeah uh or uh is Mis formulated because we don't have uh the archipel doesn't apply at all it's comparing uh oranges and apples yeah I find most sort the discussion about the philosophy AI of that kind like either trivial or pointless that's fascinating I I think that your your three main ways of talking about information are super helpful uh I I I haven't seen them used a lot in AI conversations but maybe that's it would Tamp down a lot of the hype as well because you're saying like no you you're you're mixing categories you're you're saying you're talking about information for reality in these algorithms as if it were information as reality or you know or something else you're you're you're mixing categories exactly I wonder that that's very what put no a mixing those categories generates confusion to know end and
'智能'是什么意思。如果比如说指的是会做运算,那我的口袋计算器就是有智能的,它算数比我认识的任何人都强,太厉害了,对吧。或者说这个问题问错了,因为'群岛'那套在这里根本不适用,这是在拿橘子和苹果作比较。对,我觉得大多数这类关于 AI 哲学的讨论,要么琐碎无聊,要么没有意义。这太有意思了。我觉得你区分信息的那三种主要方式非常有帮助。我在 AI 的讨论里很少看到有人用它们,但也许这正好能压下去很多炒作,因为你等于在说:不对,你是在混淆范畴,你说的是在谈算法里'关于实在的信息',却把它当成了'作为实在的信息',或者别的什么东西你是在混淆范畴。没错。我在想,这非常……混淆这些范畴会带来无穷无尽的混乱,还有
便签引用
09大脑编码与延展心智
49:11
worries about something that honestly it's like zombies yeah I'm not worried about zombies I'm not worried about AI in the sense of zombies I'm worried about humans using AI a lot right yeah and it's and it's just straight up equivocation it's this is a helpful tool for showing that we're we're equivocating on the term uh intelligence maybe or or AI I I wonder there's H I want to get into some of the goofy stuff the more out there type stuff um I think you're I think you can actually help me a lot with thinking about the extended mind uh thesis of of chmer and Clark so you you talked about Carl Popper's paper um epistemology without the knowing subject and how you thought actually that's not knowledge that's information I wonder um are our thoughts in in my mind right now whatever the mind is are they more analog or digital are they more like a record or more like a CD and then I have some following questions after that so uh of course the brain is an analog engine not least because um everything
对某些东西的担忧,说实话那就像担心僵尸一样。对,我不担心僵尸,我也不担心那种意义上的AI,我担心的是人怎么用 AI。对,很多时候这就是彻头彻尾的偷换概念,而这个工具很有用正好能说明我们在'智能'或者'AI'这个词上偷换了概念。我在想……我想聊一些更离谱的东西更天马行空的那一类。我觉得你其实能帮我很多,去想清楚查尔默斯和克拉克的延展心智论题。你之前谈到卡尔·波普尔那篇《没有认识主体的认识论》以及你为什么认为那其实不是知识、而是信息。我想问,此刻在我心里的那些想法——不管心智是什么——它们更像是模拟的还是数字的?更像唱片还是更像 CD?之后我还有几个后续问题。嗯,当然,大脑是一台模拟的机器,尤其因为说到底,一切都取决于它能得到
便签引用
50:14
works in terms of uh how much blood he gets or not the end of the day and uh there's a lot uh that can be um repres presented uh and was represented a lot in the past as uh digital in terms of neurons they fire they don't fire yes however uh now we disc for example that there are waves uh of firing uh that determine also uh our neurological uh life processes um so in a way it's a little bit like asking whether music is digital or analog these are two codes and uh uh you can encode music as analog and digital um yeah every time you do that you may or may not uh uh lose something you may be lost less or uh it might be lossy you know that you may actually keep keep translating and at the end it's almost unrecognizable uh you must have seen this this experiment run actually by uh my wife is a neuroscientist she's she's a professor of Neuroscience here she was chair of Neuroscience in oxland we came together to Yale and she's now the director of the NeuroScience Center here pres science and a friend of ours A
多少血液供应。而且有很多东西可以被表征,过去也常常被表征为数字的,就是神经元要么放电、要么不放电。不过现在我们发现,比如说存在成波的放电活动,它同样决定着我们神经层面的生命过程。所以某种意义上,这有点像在问音乐是数字的还是模拟的。这是两套编码,你可以把音乐编码成模拟的也可以编码成数字的。对,而每次这么做,你可能会、也可能不会丢掉些什么,可能丢得少一点,也可能是有损的。你知道,你可以一遍遍地转译下去,到最后几乎认不出原样了。你一定看过这个实验——其实是我太太做的,她是神经科学家,是这里的神经科学教授,她曾经是牛津的神经科学系主任,后来我们一起来了耶鲁,她现在是这里神经科学中心的主任……还有我们的一位朋友,尤其是她的朋友,也算我的半个朋友,在伯克利大学
便签引用
51:36
friend of hers in particular friend of mine by proxy the University of Berkley um they run zik M and they tried to reconstruct what people were listening to yeah by looking and then in turn out there was the war being floyed I don't know if you have uh actually heard listened to it they had kind of they had to tell you that it's being fly [Music] like okay it it's barely recognizable but it's impressive okay and then people say oh my goodness AI reads the mind yes right but how how confused can well first of all this is a bit of a trick um we've been able to do that for a while it's just that we do this much much better now uh the second thing is waves out an analog uh brain not collects those waves you register the waves of the brain and is this basically translation from one code to another code to another code to another code okay of course it's very loss you lose a lot enormously because from what was the actual sound to what we get through uh the vibration of eardrums in the brain and and then what
他们做了那个实验,试图重建人们当时正在听的东西,靠读取大脑,结果发现那是平克·弗洛伊德的曲子。我不知道你有没有真的听过那段还原,他们其实得先告诉你那是平克·弗洛伊德。[音乐] 然后你会觉得,好吧,几乎听不出来,但确实很了不起。接着人们就说:天哪,AI 会读心了对吧。但这误会得有多深啊。首先,这多少有点障眼法,这种事我们已经能做很久了只是现在做得好太多了。第二点是,脑波是模拟的,大脑并不是在'采集'这些波你记录下大脑的波形,而这基本上就是从一套编码翻译到另一套编码,再到另一套再到另一套。当然,这个过程损耗极大,你丢掉的东西多得惊人,因为从最初真实的声音,到经由耳膜振动进入大脑的东西,再到大脑就这些
便签引用
52:53
the brain is encoding in terms of this vibrations then what comes out you know that is being floyed because you know what it was what came in right right you see what I mean it's not that you you could interpret that if you didn't know that that was what came in sure any it's translation and and if you know how much Neuroscience is done these days um that should not be surprising but back to you question therefore um are we talking about an analog or Digital World um I think that as a I find it a misplaced um uh question what we need what we have um is um signals that come to the world which we can encode as digital or analog but we can switch all the time now if you think that this is too fancy uh remember that uh light is light photons depending on the frame they look like points or a wave sure and asking whether light is a wave no continuous or made of points one one Wonder bullets uh means not having understood the what you're talking about yeah because it it is one or the other of both depending on the uh essentially
振动所编码的内容,最后到输出的东西——你之所以知道那是平克·弗洛伊德,是因为你本来就知道输入是什么,对吧明白我的意思吗?如果你不知道输入是什么,你根本没法把它解读出来。当然,这全都是翻译。而且如果你了解如今神经科学做到了什么程度,这一点其实并不意外。回到你的问题:那么,我们说的是一个模拟的世界还是数字的世界?我觉得这是个提错了地方的问题。我们所拥有的,是来自世界的信号,我们可以把它编码成数字的或模拟的,而且我们随时可以来回切换。如果你觉得这说法太玄,那想想光:光是光子,取决于参照框架,它看起来像是点,或者像是波。对。而问'光到底是波、是连续的,还是由点、由一颗颗小子弹构成的',这就意味着你并没有真正搞懂自己在说什么对,因为它是这个还是那个、或者两者都是,取决于你在自己和光之间放了什么样的界面。那么,说回和
便签引用
54:15
uh interface that you put between you and the light well so the so the connection to the external mind thesis is is um look I I have a a notebook and this these are all my thoughts about your work and they are a guide for me to uh you know it's like my my active memory or something like that and they they seem like they're they're representations of my thoughts but I'm wondering you know are are are notes in a notebook do they count as analog or digital and I would think they would be digital even though it sounds weird because we say it's an analog notebook but it's not it really is a question of interpretation okay it it unless you ask what that question is forh um there is no way of answering it yes or no independently well so so Chalmers and uh Clark would say um perhaps you know maybe our mind is extended out into the notebook because our our thoughts are there and they use you know an example of uh an Alzheimer's patient who uses their notebook to guide themselves throughout the world and they
延展心智论题的联系:你看,我有一个笔记本,里面全是我对你工作的想法,它们对我来说是一份指引,就像是我的活动记忆之类的东西,它们看上去像是我思想的表征。但我想问,笔记本上的笔记算是模拟的还是数字的?我会觉得它们是数字的,哪怕这听起来很怪,因为我们平时会说那是'模拟的'笔记本但其实不是。这真的取决于怎么解释。除非你先问清楚这个问题是为什么而问的,否则根本没办法孤立地回答是或不是。好,那么查尔默斯和克拉克会说,也许我们的心智延展到了那个笔记本里,因为我们的思想就在那里。他们举了个例子,一位阿尔茨海默病患者靠笔记本来指引自己在世界中行动,他们说,在非常实在的意义上,这就是……我也不确定他们说的是哪一种
便签引用
10心智是副词:关系本体论
55:26
say well in a in a very real sense this is uh and I don't know how with the sense they talk about they don't say whether it's metaphorical or analogical or or univocal but they want you to think that this is like the conclusion is that like this is your mind isn't just in your skull it extends into your wherever your thoughts go yeah know I think I think is a is a the way I read it is a is a nice metaphor um anything more than that and I would be very critical um first of all because um uh I never seen I never touch I never sort of kick a mind I mean mind if you start think your mind almost like a liquid in your brain that PS out and makes everything wet around you sure that's a bad idea yeah uh we have a mental life um uh and for people who think that is a bit strange imagine uh someone say look the world uh is made of objects and actions and properties of objects so there are cars cars can be readed or and can go can run that's great what is mental there the mental with the analogy is
意义,他们没交代这是隐喻的、类比的,还是单义的,但他们希望你这样想——结论就是:你的心智不只在你的颅骨里,它延展到你的思想所到之处。嗯,我觉得,我读下来的感受是,这是个不错的隐喻。要是再多说一步我就会非常批判了。首先是因为,我从没见过、没摸过、也没踢过一个心智。我的意思是,如果你开始把心智想象成脑子里的某种液体,会渗出来把所有东西都弄湿在你周围,你确定那是个坏主意吗,是的,呃,我们有心智生活,嗯,呃,而对那些觉得这有点奇怪的人来说设想一下,呃,有人说,你看,世界呃是由物体、动作以及物体的属性构成的,所以有汽车汽车可以被开,或者可以走、可以跑,很好,那心智在这里是什么呢,按这个类比,心智其实是“怎么样”,所以设想一下,心智基本上是
便签引用
56:39
how so imagine basically the mental is an adverb we don't we don't have a mind we live mindfully if you have an pass me this this this oversimplification but if you have a a adverbial idea of mind and as in my case you think that a second order onology of their kind uh is really what matters that things are not they're not things and then property of things and then interactions between things but actually there are relations that constitute things then uh there is mindfulness that gets or or or leaving mindfully uh they get incarnated if you like in you in me or the new baby or someone else um then the picture is really almost upside down yeah uh and I find that philosophically more convincing because it doesn't uh create a ontology of something like a mind out of thin air yeah uh it's very dangerous like imagine with a different analogy say which I hope is a bit more in um I think it's the same story I'm telling democracy I live in a democracy well actually you live in a house this democracy thing oh oh oh you live in a
一个副词,我们并没有一个心,我们是有心智地活着,请容许我这个这个过度简化,但如果你对心持一种副词式的理解,而且像我这样,你认为那一类的二阶本体论呃才是真正重要的,也就是说事物并不是它们不是先有事物、再有事物的属性、再有事物之间的互动,实际上是关系关系构成了事物,那么呃就有了心智性,或者说有心智地活着,呃它们被具身化,如果你愿意这么讲,体现在你身上、我身上、或者刚出生的婴儿、或者别人身上,嗯那么整幅图景几乎是倒过来的,是的,呃我觉得那在哲学上更有说服力,因为它不会呃凭空造出一个像“心”这样的东西的本体论是的,呃那非常危险,比如换一个类比想想,我希望这个更……嗯,我觉得是同一个故事,我在说民主,我生活在一个民主之中,可其实你住在一栋房子里,这个民主的东西,哦哦哦,你是生活在一个
便签引用
58:06
place where actions and interactions are democratic oh you live democratically now I understand or someone runs runs quickly okay where is the quickly uh what please not point to if you start looking at a thing based ontology yeah when in fact you're talking about uh ways and relational ways of not existing the confusion leads to some kind of really funny things like oh my mind is out there right what I understand the metaphor um but it would be much more interesting to say that in a in a life live mindfully of course I am the node at the center of this network but the mindfulness know is not just this node is all the relations that I have with an Neer yeah so final analogy so in case anyone think like this guy makes no sense whatsoever um it's like thinking in terms of um roundabouts and um or Crossroads if you prefer roundabouts is a very English way uh Crossroads and Roads and ask people what comes first the crossroad or the road there like oh there the roads uh Crossroads is where the
行动和互动都是民主的地方,哦你是民主地生活着,现在我懂了,或者有人跑得,跑得很快,好,那个“很快”在哪儿呢,呃请指给我看,如果你开始用一种以事物为基础的本体论去看,是的,而事实上你谈的是呃各种方式,各种关系性的存在方式,这种混淆会导出一些很滑稽的东西,比如“哦,我的心在外面”对吧,我理解这个隐喻,嗯但更有意思的说法是,在一种生活里,有心智地生活当然,我是这张网络中心的那个节点,但心智性并不只是这个节点,而是所有的关系我与周围一切所建立的所有关系,是的,那么最后一个类比,以防有人觉得这家伙完全在胡说,嗯这就像用嗯环岛,嗯,或者十字路口,如果你更喜欢的话,环岛是很英国的说法,呃十字路口和道路,然后问人们先有十字路口还是先有路,他们会说,哦是路,呃十字路口就是路恰好交汇的地方,那我的心是什么
便签引用
59:31
roads happen to be has what is my mind my mind is a crossroad is roundabout is where everything that interacts in a certain way uh my past memories my way of speaking L English my uh my life my experiences my feelings my current state of the B all these relations all these ways of being come together mhm and it's very concrete I mean don't cross the roundabout without stopping checking and going it's totally concrete but thinking in terms of first roundabouts first the mind and then a mindful uh sort of existence is to put the crossroads first and then build the roads to connect them yeah it makes no sense if we have a relational antology and I know that this is maybe going down a rabbit Trail but so we we don't have to if you if you don't want to but uh I have the substance uh ontologist folks in my head that they're they're saying you know a relation is a relation between things so if you have a relation without things it's like what are we even talking about and it's like the Chicken and the Egg I
我的心是一个十字路口,是一个环岛,是所有以某种方式互动的东西交汇之处,呃我过去的记忆、我说英语的方式、我的呃我的生活、我的经历、我的感受、我身体当下的状态,所有这些关系,所有这些存在方式汇聚到一起,嗯,而且它非常具体,我是说,别不停车、不观察就直接冲过环岛,要停、要看、再走,这完全是具体的,但如果先想环岛、先有心,然后才有一种有心智的呃某种存在,那就是把十字路口摆在前面,然后再修路把它们连起来,是的,如果我们采取一种关系本体论,这就完全讲不通……关系本体论,我知道这也许是在跑题,所以我们不必聊,如果你,如果你不想的话,不过呃我脑子里有那些实体呃本体论者,他们在说,你知道的,一个关系是事物之间的关系,所以如果你有一个没有事物的关系,那我们到底在谈什么这就像鸡生蛋蛋生鸡,我同意,所以这里我们需要两步,好,首先有一种做法,就是
便签引用
11信号、本体论与形而上学的边界
1:00:38
agreed so here we need two steps okay well first of all there is a way which not the two step there's a way of doing um what's called structural realism uh in phal science um and you can do that in an informational way that tries to solve both both the chicken and the egg yeah but that is a bit visualiz and for the folks know listening to this not just looking informational structural realism and as that's the RO you can take we we yeah we did a whole episode on that for folks interested there go yeah but let's keep it uh uh simple and straightforward on an everyday not not philosoph science but everyday experience so you can really have two kinds here at least two kinds way of speaking about reality one is I have access like I look I check maybe with tools maybe with science maybe with Theory but I have the means to tell you what the world is like MH okay maybe improvable but there is that possibility is open the other one that says look forget about what it's out there what is out there is you can see where I'm going
不是那两步,有一种做法是嗯所谓的结构实在论,呃在科学哲学里,嗯你可以用一种信息论的方式来做,它试图同时解决鸡和蛋两边的问题,是的,但那有点专门,而对在听这段的朋友们来说,不只是看,去搜一下“信息结构实在论”,那就是你可以走的路我们,我们,是的,我们专门做过一整期讲那个,感兴趣的朋友可以去看,是的,但我们还是保持呃呃简单直接一点,用日常的,不是不是科学哲学,而是日常经验,所以这里你其实可以有两种,至少两种谈论实在的方式,一种是我有通道,比如我看、我查,可能借助工具、可能借助科学、可能借助理论,但我有办法告诉你世界是什么样的,嗯,好,也许可以改进,但那个可能性是开放的;另一种说法是,你看别管外面是什么,外面的那个东西——你能看出我要往哪儿走——它在向我们发送信号,我们得到的
便签引用
1:01:47
it's sending signals to us what we get are the signal the signals is what build our own onology uhhuh my ontology which is know this talking to you this pen in my head is the result the outcome of this particular receiver processing those particular signals so that this is the reality in which know I am living and it's completely real I mean the signals are real the everything is concrete I mean I'm not going like oh there's no real real forget about that's we leave that to childish games and games played by philosophers talking about the real life the life when you go to the do the grocery come back what kind of reality are we talking about the metaphysical one reality in itself for the cence the nomenon or the outcome of my interactions with the world through this particular interface epistemically speaking that is my body blah blah blah my culture and so on my language and the outcome of that call that ontology once you have the distinction I'm getting to the chicken and egg uh no this is this is really
就是信号,而信号构建了我们自己的本体论,嗯哼,我的本体论,也就是这个,跟你说话,我手里这支笔……在我脑中,这是这个特定的接收器处理那些特定信号的结果、产物,所以这就是我所生活其中的实在,而且它完全真实,我是说信号是真的,一切都是具体的,我不是要说,哦根本没有什么真正的真实,算了吧,那种我们留给幼稚的游戏,留给哲学家玩的游戏,我们谈的是真实的生活,你去买菜、再回来的那种生活,我们说的是哪种实在呢,是形而上学的那种,实在自身、物自体、本体,还是我通过这个特定界面与世界互动的结果从认识论上说,那个界面就是我的身体,等等等等,我的文化之类,我的语言,而那个结果,把它叫做本体论,一旦你有了这个区分——我这就说到鸡和蛋——呃不,这个这个真的很
便签引用
1:02:52
helpful the chicken and egg problem is for the metaphysician uhhuh the metaphysician who wants to tell you what the world is in itself independently of my or a human or a particular interface or anything uh as if he had not a God's eye perspective on the world classic know it has to tell you what the foundational bits are and what comes first and what comes later so roundabouts things and their relations between things notes and Rel and they don't know I mean they they just they've been running in circles because of course chicken and eggs what well and and because because um if if we do take a Content picture it's like well you're trying to escape your own mind right if you're if you're going for Trans IDE so you're also cting your own language your own categories from think in terms of trying to do metaphysics try to do onology meaning okay how do we come up what's the best model that we can get of this epistemic ontology that we have on us side and here the beauty of modeling something is that you can decide yeah
有帮助,鸡生蛋的问题是形而上学家的问题,嗯哼,是那种想告诉你世界自身是什么样子的形而上学家,独立于我的、或人类的、或任何特定界面,呃仿佛他拥有一种上帝之眼的世界视角,经典做法,他得告诉你最基础的部件是什么,什么先来、什么后来所以是环岛、事物和事物之间的关系,节点和关系,而他们不知道,我是说他们只是一直在原地打转因为当然是鸡和蛋嘛,而且而且因为因为嗯,如果如果我们真采取一种康德式的图景,那就像是你在试图逃出自己的心,对吧,如果你如果你走先验唯心论那条路,那你也在……你自己的语言、你自己的范畴,从试图做形而上学、试图做本体论的角度想,意思是,好,我们怎么得出我们能得到的、关于我们这一侧这个认识论本体论的最佳模型,而这里,建模的妙处在于你可以自己决定,是的,你可以说,我们就决定这样开始建模,因为这样更合理,更
便签引用
1:04:01
you can say let's decide that we start modeling because it make more sense it's more fruitful it's more explanatory let's decide that we start with lines not with points yeah and then you start on the Whiteboard saying okay line there a relation and relation links A and B I've just done it oh you just got out of the trick well it's just saying how do I want to model the relationship between egg and chicken I'm going to start chronologically make more sense start with with the egg and then the chicken yeah or but in real life I'm not doing metaphysic that's your that's your problem not so that's good the steps are epistemic onology not a metaphysical approach and on the epistemic ontology realize that we have to model the epistemic onology philosophically in order to understand it and because in modeling you're actually designing what you need yeah you can start wherever the heck you want yeah well I wonder can you so some people want to take that epistemic uh ontology and then run it run it back into metaphysics and say um
有成果、更有解释力,我们就决定从线开始,而不是从点开始,是的,然后你在白板上开始写,好那里有条线,一个关系,这个关系连接 A 和 B,我刚就做到了,哦你刚从这个套里跳出来了,嗯这只是在说,我想怎么给蛋和鸡的关系建模,我打算按时间顺序开始,这样更合理,先从蛋开始,然后是鸡,是的,或者,但在现实生活里我不是在做形而上学,那是你的,那是你的问题,不是我的,所以这挺好,那两步是:认识论本体论,而不是形而上学的路子;然后在认识论本体论上认识到我们必须在哲学上给这个认识论本体论建模,才能理解它,而因为在建模时你其实是在设计你所需要的东西,是的,你想从哪儿开始就从哪儿开始,是的,嗯我想问,你能不能,有些人想把那个认识论呃本体论拿过来,然后把它反推回形而上学,说嗯,那信息就是一种中立一元论
便签引用
1:05:09
well information is a neutral monism then right because look at look at all the work we can do epistemically with information and and the signals that we receive so it looks like let's just say that the numeral realm just is information right is that what's going on in what do make of that so you can uh there is there is um Freedom so to speak so you can um or to use find terminology your ontological commitment can go much much further and say look I have an epistemic ontology it works very well in fact my epistemic onology works so well that I can put robots on Mars so that's a good solid not no way of understanding the signal that that the ultimate nominal source is sending to me interpreting them at the best way such an extent successful know and therefore this instrumental successfulness in my Humanity so in the way uh the human uh uh epistemic onology Works totally fine do I have to say next step and that's the way the world is well if you like but that is an Act of Faith there's nothing that forces you to
对吧,因为你看,看我们在认识论上,用信息、用我们接收到的信号能做多少事,所以看起来,那我们干脆说本体界就是信息,对吧,是这样吗,你怎么看这个,所以你可以呃有,有嗯所谓的自由,所以你可以嗯,或者用蒯因的术语,你的本体论承诺可以走得更远更远,说,你看,我有一个认识论本体论,它运作得非常好,事实上我的认识论本体论好到我能把机器人送上火星,所以那是一种很扎实的方式,去理解那个终极的本体源头发送给我的信号,以最好的方式解释它们,成功到这种程度,因此这种工具性的成功,在我的人性范围之内,所以就呃人类的呃呃认识论本体论运作得完全没问题,那我是否必须再走下一步,说世界就是这样,嗯,你要愿意也行,但那是一种信仰之举,没有任何东西迫使你
便签引用
1:06:19
say that okay I mean you you're welcome um but in a way you know as any good scientist scholar philosopher what we should do is to stay within what is guaranteed by our understanding yeah if we want to say of and above what we have reasons to believe that oh there's so much more so essentially is is a friction free commitment like yes or not I mean it's it's it's anyone game um yeah and we don't have to uh we we are free from uh so normally a commitment is required that's why uh in this case it would be just uh 100% volunteering an ontological commitment to something that is not required yeah that me can I do it yes anytime right do I have to do it is Imagine The Following thing like is the radio send signals send music I'm on here I receive the music I interpret the the signals as music I also want to believe that the signals represent the intrinsic nature of the radio you are free to do that do they I it's unlikely okay the radio sends signals um now someone say no no no look the radio is s is sending a specific
那么说,好吧,我是说,你,请随意,嗯但某种意义上,你知道,作为任何一个好的科学家、学者、哲学家,我们应该做的是停留在我们的理解所能保证的范围之内,是的,如果我们想说超出我们有理由相信的东西,说哦,还有那么多更多的,所以本质上这是一种无摩擦的承诺,要或不要,我是说这,这,这谁都可以玩,嗯是的,而我们不必呃,我们可以免于呃,通常承诺是被要求的,这就是为什么呃在这种情况下,那只会是呃百分之百自愿的,对某个并不被要求的东西作出本体论承诺,是的,那我能做吗,能随时可以,对吧,但我必须做吗,设想下面这件事,就像收音机发出信号、发出音乐,我在这头,我接收到音乐,我把这些信号解释为音乐,我还想相信这些信号代表了收音机的内在本性,你可以自由这么做,它们真代表吗,我看不太可能,好,收音机发出信号,嗯现在有人说,不不不,你看收音机发的是一种特定的信号,是关于它自身本性的信号,嗯那是个有意思的想法,也许吧,但我们怎么知道呢,是的
便签引用
12模拟假设与「哲学家的问题」
1:07:42
kind of signals signals about its own nature well that's an interesting thought so maybe but how do we know yeah you right and and and that is where my Canon sort of uh breaks start going like red like like no this it and and there's nothing wrong met it's like um uh storytelling is anyone's game um enjoy go like yeah have fun like but then I reten and it's much better yeah well so so final final question for you and you've been so generous with your with your time here this is this is for the audience my my audience loves simulation hypothesis and I can see a connection here where someone might say well look if we're okay so we go in for for transcendental idealism of a certain stripe and we're receiving these signals and uh we we we can't say what the thing in itself is does that mean that we're living you know in a simulation even if even even if not a uh a computer simulation you know is it is the world that we see the laban's velt or the uh the world for me whatever terminology we use ontology the world of ontology is
你说得对,而且而且那就是我那套康德式的刹车呃开始亮红灯的地方,像像,不,这,而且而且这也没什么错,形而上学,这就像嗯呃讲故事,谁都能玩,嗯玩得开心,去吧,是的,玩得开心,不过那样的话我还不如去读小说,那要好得多,是的,嗯那么那么最后一个问题问你,你在时间上已经非常大方了,这个这个是给观众的,我的,我的观众很喜欢模拟假设,我能看到这里有个联系,有人可能会说,嗯你看如果我们,好,我们接受某种类型的先验唯心论,我们接收着这些信号而呃我们我们我们说不出物自体是什么,那这是否意味着我们生活在,你知道,一个模拟之中,即使即使不是呃计算机模拟,你知道,我们所看到的这个世界,生活世界,或者呃对我而言的世界,不管我们用什么术语用本体论,本体论的世界,那是一个被模拟的世界吗,也许是被我们自己的心所模拟,或者说,那正是
便签引用
1:08:50
that a simulated World simulated by our own minds maybe or that that's what gives a bad name to philosophy unfortunately this kind of uh discussions um um and that's what gets philosopy departments closed down because people think oh that's what you're doing okay we don't need you interesting uh but no let me take it seriously for a moment um it's like U it's like saying um we live in a world full of um gos H you can't you can't touch them you can't see them in fact you can't even prove that they are there but they're there yes or no right what kind of question is that the premise undermines any possibility of providing a ration answer to the price do we or do we not live in a simulation generated by the future the premise undermines any possibility of a reasonable answer request H so it's just a it's just a trick it's one more trick the philosopher enjoy but they should not play or at least not publicly because when Society sees us wasting time after this silly things then the Judgment no
让哲学声名不佳的东西,很遗憾,就是这类呃讨论,嗯嗯,也正是这种东西让哲学系被关掉因为人们会想,哦你们干的就是这个,好吧,那我们不需要你们,有意思,呃不过不,让我认真对待它一会儿嗯这就像,呃这就像在说嗯我们生活在一个满是嗯鬼魂的世界里,嗯,你摸不到它们,你看不到它们,事实上你甚至无法证明它们在那儿,但它们就在那儿,是还是不是,对吧,这算什么问题啊,这个前提摧毁了任何给出理性回答的可能性,我们究竟是不是生活在一个由未来生成的模拟之中,这个前提摧毁了任何合理回答的可能性,嗯,所以这只是个,只是个把戏,又是哲学家爱玩的一个把戏但他们不该玩,或者至少不该公开地玩,因为当社会看到我们把时间浪费在这些蠢事上,那样的评判,是我们自找的,好吧,如果哲学就是这个,那人们会说,嗯
便签引用
1:10:12
we call for it okay if that is what philosophy is then people will ask well not with my taxpayer money no enjoy it Saturday and Sunday yeah need something else so let me let me put it in a different way um there are a lot of questions that we ask there um I call them Absolute questions that provide by being POS no constraint whatsoever in terms of what could be the space or a reasonable answer H if you have removed by asking that question any chance of providing an answer that makes any sense well why are you asking the first question in the first place so what what is so sometimes and that's where we took the wrong step and with all the St experiments simulation and there's a confusion at the source MH and it's the malicious demon confusion from Dad onwards at least but also Plato when it says the the ring that makes you invisible so all all these things I mean we're talking about play and B so were they just ridiculous people even play even no what they were doing were thought experiments as means to an end right
别花我的纳税钱,不行,周六周日自己玩去,是的,得干点别的,那么让我,让我换个方式说,嗯有很多我们提出的问题,嗯我把它们叫做绝对问题,这些问题一被提出,就完全没有提供任何约束,关于合理回答的空间可能是什么,嗯,如果你通过提出那个问题,已经排除了给出任何有意义回答的机会那你一开始为什么要问这个问题呢,所以什么,什么是,所以有时候,而那就是我们走错一步的地方,还有所有那些思想实验、模拟之类的,源头上就有一个混淆,嗯那就是恶魔的混淆,至少从笛卡尔以来,不过柏拉图也一样,当他讲那枚能让你隐身的戒指,所以所有所有这些东西,我是说我们谈的是柏拉图和笛卡尔,那他们只是荒唐的人吗,连柏拉图也也不,他们所做的是思想实验,是达到目的的手段,对吧,它们本身不是目的,当然,他们
便签引用
1:11:29
they were not ends themselves sure they were not asking uh is it possible no to uh be constantly know deceived by a demon do I live or do I not live in a world in which a demon is creating all this no Matrix simulation it's always the same story a virtu that wasn't asking that stupid question because he was a genius he was asking how can I possibly test at the very end of no the road in the most dramatic sense like road test this car what are the most extreme weather conditions for this car called human knowledge under which he will crack and the mostes are imagine a demon can he even pass that test he can at least no K know cogo Etc then I can drive that car even when a demon is around I hope people are getting the analogy here yeah it's a great but that was you know but think was an engineer you shouldn't forget yeah so when when he says look let me test this artifact that is human knowledge putting so much pressure until he cracks does he crack even when I go total blast no Malo no something Remains the C bingo I can
并不是在问呃,是否可能呃一直被一个恶魔欺骗,我是不是生活在一个恶魔在制造这一切的世界里,什么《黑客帝国》、模拟,总是同一个套路,笛卡尔并不是在问那个愚蠢的问题,因为他是个天才,他问的是,我怎样才能在最极端的意义上去测试就像给这辆车做路测,最戏剧化的意义上,对这辆叫做人类知识的车来说最极端的天气条件是什么,在什么条件下它会崩掉,而最极端的就是设想一个恶魔,它连那个测试都能通过吗,它能至少,你知道,“我思”等等,那么即使有恶魔在旁边我也能开那辆车,我希望大家听懂了这个类比,是的,很棒,但那,你知道,别忘了他是个工程师,是的,所以当当他说,你看,让我来测试这个叫做人类知识的人造物,施加巨大的压力直到它崩裂,它会崩吗,即使我全力开火,什么恶魔不,还有东西留下来,那个“我思”,宾果,我可以从那里重建一切,砰,这是经典,我们今天还读它,正是因为
便签引用
1:12:44
rebuild everything from there boom no it's a class we still read it because of that but not because he was wondering do I or do I not leave in a world where malicious making me think that in front of the five light see that is totally useless because if that is not the not the tool you're using to test the theory but it's actually the end of the game yeah then you're playing the wrong game yeah that game is a waste of time and that's what we started with that is not a pH a philosophical question that is a Philosophers question and no one can at all about philosophers questions everybody cares about philosophical questions and there distinction I hope no if you there's one message for this keep this in mind am I asking a philosophical question or am I asking philosophers questions extended mind virtual reality simulation AI taking over the world that we live no trace and we'll be in the chapter of how philosophy can be so embarrassing no thank you okay there is no book like that that I want to see
这一点,而不是因为他在琢磨我是不是生活在一个恶魔让我以为眼前有火光的世界里,你看,那完全没有用,因为如果那不是,不是你用来检验理论的工具,而是实际上成了这场游戏的目的,是的,那你就玩错了游戏,是的,那个游戏是浪费时间,而那正是我们开头讲的,那不是一个哲学问题,那是一个哲学家的问题,而没有人会在意哲学家的问题每个人都在意哲学问题,这里有个区分,我希望,如果这次只有一条信息值得记住的话记住这一点:我问的是一个哲学问题,还是一个哲学家的问题,延展心智、虚拟实在模拟、AI 接管我们所生活的世界,这些都不会留下任何痕迹,只会被写进“哲学能有多丢人”那一章里,不必了,谢谢,好吧,没有那样的书,我倒真想看到那样一本书摆在后面的书架上,那就是我一直
便签引用
1:13:59
back in back there that's have I being uh abrasive enough yeah it's so good doc Dr FL this this has been fantastic thank you so much for for all your time and and letting me you know go through all these even philosopher questions this has been great um folks that's gonna have to do it for now this has been Parker peny and as always all glory to God
嗯,够犀利了,对,太棒了,博士,弗博士,这次真的太精彩了,非常感谢您抽出这么多时间,也谢谢您让我把这些哲学问题都问了一遍,这次聊得特别好。嗯,各位,今天就先到这里,我是帕克·佩尼,一如既往,一切荣耀归于神。
便签引用
视频总结 · 一句话概括与核心要点

一句话概括

弗洛里迪梳理了"信息"的三重含义(作为实在、关于实在、为了实在),并据此主张当前 AI 的真正革命不是造出了智能,而是史上首次把"成功的能动性"与"智能"脱钩,从而催生了一种河流、狗、人之外的第四类能动者。

核心要点

  • 信息的哲学起点是三条线索的汇合。 弗洛里迪本科时读了苏珊·哈克《逻辑哲学》意大利译本和摩尔《伦理学》,被"讨论问题而非讨论哲学家"的分析传统吸引;之后写信给尚在世的波普尔并获回信,其《无认知主体的认识论》一文让他意识到"不属于任何人却可被拥有、传送、塑造的东西"应叫信息而非知识。第三条线索是他少年时的神学兴趣:用信源—信道—接收者模型思考"神如何向人类传递福音"。三者合流后,他在牛津沃尔夫森学院以博士后身份提出题为"什么是信息哲学"的讲座,该领域由此得名。
  • 信息至少分三种,混用是绝大多数困惑的来源。 一是"作为实在的信息":树的年轮、脚印、黑胶唱片的沟槽,本质是世界中的模式;二是"关于实在的信息":有意义的数据 + 语法 + 真值,即维基百科式的事实性语义信息;三是"为了实在的信息":音乐文件、菜谱、算法,有意义和结构但无真假可言。他强调"真"在日常语境中常指"地道、名副其实",与哲学上的真值不是一回事。
  • 年轮不"想"告诉你任何事,信息来自我们对相关性的利用。 树龄与年轮之间只是可被开采的相关性(A 有特征 F 与 B 有特征 G 相关),树本身无意图。类比:发现无人岛上有规律重复的符号时,我们先看到句法结构,再推断背后有语义和发信者。但把年轮说成"密码"是危险的隐喻。他直接把这条原则推到 AI:无论模型输出什么,都只是模式,AI 没有打算说任何话,也没有计划和意图。
  • 同一物理痕迹在不同"接口"下意义不同,这是数据对理论的欠决定。 主持人问:若我有意留下指纹以便被找到,信息性质是否改变?弗洛里迪用计算机科学中的"抽象层次"回答:森林里的碎布可能是求救记号,也可能只是背包破了洞,两种现象在物理上完全相同,区别在于诠释接口。这与科学哲学中"数据欠决定理论"是同一问题。
  • 世界本身是模拟的,"比特"是编码而非自然物。 自然不做跳跃,因此不是二进制的;比特几乎只存在于我们发明的代码中(如电报的点划)。人类史上真正的革命是从莱布尼茨、霍布斯经布尔到图灵,发现可以把一切连续信息编码为离散代码,再造出能自动处理该代码、进而从中统计性地挖出深层模式(如全体英语语料中的结构)的机器。从第一台计算机到 ChatGPT 的这条路是人类才智的奇迹,说"智能自己涌现了"反而贬低了工程师的成就。
  • AI 的真正突破是"零智能的成功能动性",问题也全部藏在这个缺口里。 历史上首次,停车、下棋、仓储、采草莓、造车这些任务能在零智能下成功完成并持续改进(输出回馈为输入)。他明确说:AI 伦理中的隐私、偏见、版权、资源控制、剥削、深度伪造等所有难题,都源于能动性与智能之间这道正在扩大的鸿沟,而不是"机器会不会觉醒"。他不担心 AI 像僵尸,担心的是人类如何使用 AI。
  • 能动性应分级,AI 是史无前例的第四类。 第一类是河流、地震:仅与世界互动、造成改变;第二类是狗:能学习且有意向(趋利避害);第三类是人:在此之上还有自觉的自我控制,能违抗本能(自杀、把手指按在烛火上三十秒)。AI 能互动、能学习,但无意向、无自觉规划,既不是钻头也不是"五岁小孩"。他称这是尚无概念工具可用的新对象,是 21 世纪哲学最值得研究的空白;坚持把它归入前三类的人仍困在"弗兰肯斯坦/魔像"式的现代思维里。陷阱不会学习,因此和地震同类,不算这种能动性。
  • "智能"无定义但有判准,"机器是否智能"这一问要么琐碎要么提错了。 判准之一是灵活性:火警响时人会弃棋逃生,电脑会赢完棋跟房子一起烧掉;还有社交、音乐、直觉等多种智能。他把智能定义为一切愚蠢行为的对立面(忘带护照、认为 2+2=22、在朋友糟糕的一天说错话),因此它是一片群岛而非一座岛,像意法边界一样"既不在此也不在彼"。若智能就是会算数,那计算器早已智能;词汇随意选,但要警惕它带来的后果,如"AI 会接管宇宙"式的恐慌,多数只是广告与炒作。
  • 心灵是副词,不是名词;延展心灵论只是好隐喻。 脑是模拟引擎(靠血流和神经放电波),问"思想是模拟还是数字"如同问光是波还是粒子,答案取决于接口。伯克利团队从脑波重建受试者听的平克·弗洛伊德,只是在多重有损编码之间翻译,且必须先知道输入是什么才认得出,绝非"AI 读心"。他主张关系性本体论:正如我们不"拥有"民主而是"民主地生活",我们不拥有心灵而是"有心地活着";心灵像环岛,是道路交汇处,先有环岛再修路是本末倒置。查默斯与克拉克的延展心灵论把心灵想成会渗出来把周围弄湿的液体,超出隐喻就站不住。
  • 鸡与蛋、纽米纳、模拟假说都是"哲学家的问题",不是"哲学问题"。 他区分"认知本体论"(我们作为特定接收者处理世界信号得到的实在,完全真实,能把机器人送上火星)与形而上学的"世界本身"。鸡蛋难题只困扰想以上帝视角谈本体的形而上学者;在建模时你可以自由选择从线(关系)而非点(事物)开始。把信息本体论进一步说成中立一元论或"本体界就是信息",是自愿的信仰行为,如同相信收音机的信号揭示了收音机的内在本性,允许但毫无必要。模拟假说则是"绝对问题":其前提本身取消了任何合理回答的空间,好比问"世界上有摸不到看不见也无法证明的鬼,有没有"。

结论与值得注意的细节

  • 弗洛里迪为笛卡尔辩护:恶魔论证是极端条件下对"人类知识"这辆车的压力测试,目的是找到崩溃点之外仍存留的东西(我思),再从那里重建,而非真的追问"我是否活在幻觉中"。他说笛卡尔本质上是工程师;柏拉图的隐身戒指同理,思想实验是手段而非目的。
  • 他警告哲学系被裁撤正是因为社会看到哲学家公开沉迷延展心灵、虚拟现实、模拟假说、"AI 统治世界"这类问题,直言"周末自己玩可以,别花纳税人的钱"。
  • 个人信仰方面:他在本科一二年级间逐渐失去信仰,但拒绝换成无神论,理由是"证据缺席不是缺席的证据",自称"灵性的不可知论者"。
  • 他现任教于耶鲁,妻子是神经科学家,曾任牛津神经科学系主任,现为耶鲁神经科学中心主任,提到的脑波重建实验来自她的伯克利同事。
  • 关于"人工智能"这一名称,主持人认为它恰恰准确:是"人工的",是程序员意图的再现。弗洛里迪表示叫什么都行,只要清楚词汇的后果。
  • 他提到《信息:极简导论》一书,并称本次谈话的核心区分都在书中;也提到《AI 伦理学》一书的一位审稿人曾认为四类能动者的划分"完全荒唐"。
核心句型 · 9
1. It's not that … but it's like …
“It's not that the one day you wake up and say oh I don't have faith anymore but it's like the foundations of a building”
先否定一种直觉解释,再用比喻给出真实图景。适合纠正误解、讲述渐变过程。仿写:It's not that he suddenly quit, but it's like a slow leak.
2. absence of evidence is not evidence of absence
“Absence of evidence is not evidence of sound absence”
倒装对称的固定格言,用于拒绝从「没证据」跳到「不存在」。可套用同构:correlation is not causation。
3. no longer X as something, but X about something
“It's no longer information as something information about something”
用介词变化区分概念层次(as / about / for),是学术表达里最经济的区分手段。仿写:not data as objects, but data about objects.
4. for the first time in human history we have been able to decouple A and B
“For the first time in human history we have been able to decouple successful agency and intelligence”
「史上第一次」+ decouple,强调某种历史性拆分。用于提出核心论点时加重分量,A 和 B 须为常被绑定的两概念。
5. imagine doing X at zero intelligence
“Imagine parking a car at zero intelligence”
祈使句 imagine + 动名词 + at zero …,用极端量化制造反直觉画面,一连串排比后再收束。仿写:imagine translating at zero understanding.
6. it would be stupid to …; it would have been intelligent to …
“It would be stupid to leave the finger in the door … so it would have been intelligent not to leave the finger”
用对立的虚拟语气成对定义概念,靠否定面勾勒正面。注意 would be 与 would have been 的时态呼应。
7. the question becomes either trivial or misformulated
“The whole question is this machine intelligent or not becomes either trivial or mis formulated”
either … or 给问题本身分类,而不是回答问题,是哲学讨论中「拆解提问」的标准句式。后面各接一个 because 说明。
8. the premise undermines any possibility of …
“The premise undermines any possibility of providing a ration answer”
指出前提自我拆台,是驳斥「不可证伪」问题的利器。搭配 any possibility of + 动名词。
9. am I asking a philosophical question or a philosophers' question?
“Am I asking a philosophical question or am I asking philosophers questions”
同一词根的形容词与所有格并置制造对照(philosophical / philosophers'),一句话完成价值判断。可仿:a scientific problem or a scientists' problem.
词汇精讲 · 115 · 按出现顺序
pretentious /prɪˈtenʃəs/ adj. 0:52
装腔作势的,矫饰的
swine /swaɪn/ n. 0:52
猪;此处自嘲「粗人、土包子」
seminary /ˈsemɪneri/ n. 1:52
神学院
set Book phr. 2:53
(课程的)指定教材,英式说法
tackle /ˈtækəl/ v. 4:05
着手处理(难题)
cherish /ˈtʃerɪʃ/ v. 5:29
珍视,珍藏
epistemology /ɪˌpɪstəˈmɑːlədʒi/ n. 5:29
认识论
epistemic logic phr. 5:29
认识逻辑,研究「知道」「相信」等算子的模态逻辑分支
philology /fɪˈlɑːlədʒi/ n. 8:03
语文学,古典文献学
eroded /ɪˈroʊdɪd/ v. 10:17
侵蚀,逐渐削弱(erode 的过去分词)
rigorous /ˈrɪɡərəs/ adj. 11:32
严谨的,严格的
agnostic /æɡˈnɑːstɪk/ n. 11:32
不可知论者
comes with a territory phr. 11:32
标准说法 come with the territory:是这份工作/身份必然附带的
ontological /ˌɑːntəˈlɑːdʒɪkəl/ adj. 12:44
本体论的
Revival /rɪˈvaɪvəl/ n. 12:44
复兴,回潮
semantic /sɪˈmæntɪk/ adj. 13:56
语义的
endorse /ɪnˈdɔːrs/ v. 15:13
认可,采用(某种用法)
syntax /ˈsɪntæks/ n. 15:13
句法,符号的组合规则
truth value phr. 15:13
真值(逻辑学:真或假)
gist /dʒɪst/ n. 16:30
要点,大意;get the gist 大致明白
conflate /kənˈfleɪt/ v. 17:35
混为一谈
intentionality /ɪnˌtenʃəˈnæləti/ n. 17:35
意向性(心灵哲学:心理状态「关于」某物的指向性)
representational /ˌreprɪzenˈteɪʃənəl/ adj. 17:35
表征性的
correlation /ˌkɔːrəˈleɪʃən/ n. 17:35
相关性
exploit /ɪkˈsplɔɪt/ v. 18:40
利用(此处无贬义)
Hallmark /ˈhɔːlmɑːrk/ n. 18:40
标志,典型特征
regularity /ˌreɡjəˈlærəti/ n. 18:40
规律性
crack /kræk/ v. 18:40
破解(密码);第 60 段作「崩溃、垮掉」
metaphorical /ˌmetəˈfɔːrɪkəl/ adj. 19:54
隐喻的,比喻性的
dendrochronologists /ˌdendroʊkrəˈnɑːlədʒɪsts/ n. 19:54
树木年代学家(通过年轮断代)
triggered /ˈtrɪɡərd/ adj. 19:54
(网络用语)被戳到痛处、情绪被激起
presupposes /ˌpriːsəˈpoʊzɪz/ v. 21:05
预设,以……为前提
Observer relativity phr. 21:05
观察者相对性(Searle:某些特征依赖观察者才成立)
sedimenting /ˈsedɪmentɪŋ/ v. 22:02
沉积,沉淀
reverse engineer phr. 22:02
逆向推导,反向工程
strata /ˈstreɪtə/ n. 22:02
地层(stratum 的复数)
artifacts /ˈɑːrtɪfækts/ n. 22:02
人工制品,文物
astonishing /əˈstɑːnɪʃɪŋ/ adj. 23:13
令人震惊的
Leverage /ˈlevərɪdʒ/ v. 24:23
借助,利用(某种优势)
remnants /ˈremnənts/ n. 25:27
残余物,碎片
underdetermine /ˌʌndərdɪˈtɜːrmɪn/ v. 26:39
欠决定(科学哲学:证据不足以唯一确定理论)
binary /ˈbaɪneri/ adj. 27:46
二进制的,二值的
discrete /dɪˈskriːt/ adj. 29:06
离散的(区别于连续)
encoding /ɪnˈkoʊdɪŋ/ v. 29:06
编码
Ingenuity /ˌɪndʒəˈnuːəti/ n. 30:28
巧思,创造才能
belittled /bɪˈlɪtld/ v. 31:56
贬低,轻视
benign /bɪˈnaɪn/ adj. 31:56
良性的,无害的
decouple /diːˈkʌpəl/ v. 33:00
解耦,使分离
agency /ˈeɪdʒənsi/ n. 33:00
能动性,行动能力(哲学术语)
Golem /ˈɡoʊləm/ n. 33:00
魔像,犹太传说中被赋予生命的泥人
in view of a goal phr. 33:00
以某目标为着眼点
warehousing /ˈwerhaʊzɪŋ/ n. 33:00
仓储管理
intellectual property phr. 34:12
知识产权
exploitation /ˌeksplɔɪˈteɪʃən/ n. 35:22
剥削
booby traps /ˈbuːbi træps/ n. 35:22
陷阱机关,饵雷
Jin up phr. 35:22
应为 gin up:煽起、炒热(兴趣、热度)
rudimentary /ˌruːdɪˈmentəri/ adj. 36:25
原始的,初级的
deductive /dɪˈdʌktɪv/ adj. 36:25
演绎的
unorthodox /ʌnˈɔːrθədɑːks/ adj. 37:26
非正统的
Bonkers /ˈbɑːŋkərz/ adj. 37:26
(口语)疯了的,荒唐的
omnipotent /ɑːmˈnɪpətənt/ adj. 37:26
全能的
hermeneutics /ˌhɜːrməˈnuːtɪks/ n. 42:18
诠释学;此处指理解世界的解释框架
hype /haɪp/ n. 43:31
炒作,夸大宣传
take over phr. 43:31
接管,控制
criteria /kraɪˈtɪriə/ n. 44:36
标准,判据(criterion 的复数)
deployed /dɪˈplɔɪd/ v. 44:36
运用,投入使用
counterpart /ˈkaʊntərpɑːrt/ n. 45:45
对应物,对立面
boundless /ˈbaʊndləs/ adj. 45:45
无限的
pinpoint /ˈpɪnpɔɪnt/ v. 46:53
精确指出
regimented /ˈredʒɪmentɪd/ adj. 46:53
严格规整的(蒯因用语:规整化的语言)
archipelago /ˌɑːrkəˈpeləɡoʊ/ n. 46:53
群岛
trivial /ˈtrɪviəl/ adj. 46:53
琐碎的,无足轻重的
oranges and apples phr. 48:08
标准说法 apples and oranges:无法相比的两类东西
Tamp down phr. 48:08
压下去,抑制
equivocation /ɪˌkwɪvəˈkeɪʃən/ n. 49:11
偷换概念,一词多义导致的谬误
goofy /ˈɡuːfi/ adj. 49:11
傻气的,古怪的
out there phr. 49:11
(口语)离谱的,天马行空的
lossy /ˈlɔːsi/ adj. 50:14
有损的(压缩、转换中丢失信息)
by proxy phr. 51:36
间接地,通过他人
barely recognizable phr. 51:36
几乎认不出来
misplaced /mɪsˈpleɪst/ adj. 52:53
(问题、担忧)放错地方的,不恰当的
photons /ˈfoʊtɑːnz/ n. 52:53
光子
univocal /juːˈnɪvəkəl/ adj. 55:26
单义的,字面同一含义的
analogical /ˌænəˈlɑːdʒɪkəl/ adj. 55:26
类比的
oversimplification /ˌoʊvərˌsɪmplɪfɪˈkeɪʃən/ n. 56:39
过度简化
incarnated /ɪnˈkɑːrneɪtɪd/ v. 56:39
化身,具身化
out of thin air phr. 56:39
凭空
roundabouts /ˈraʊndəbaʊts/ n. 58:06
环岛(英式说法)
rabbit Trail phr. 59:31
跑题的岔道(类似 rabbit hole)
substance /ˈsʌbstəns/ n. 59:31
实体(哲学术语)
structural realism phr. 1:00:38
结构实在论(科学哲学立场)
epistemically /ˌepɪˈstemɪkli/ adv. 1:01:47
从认识论上说
God's eye perspective phr. 1:02:52
上帝之眼视角,超越任何立场的全知视角
running in circles phr. 1:02:52
原地打转,徒劳无功
fruitful /ˈfruːtfəl/ adj. 1:04:01
富有成果的
explanatory /ɪkˈsplænətɔːri/ adj. 1:04:01
有解释力的
chronologically /ˌkrɑːnəˈlɑːdʒɪkli/ adv. 1:04:01
按时间顺序地
neutral monism phr. 1:05:09
中立一元论(心与物同源于中性基质)
ontological commitment phr. 1:05:09
本体论承诺(蒯因:理论承认哪些东西存在)
instrumental /ˌɪnstrəˈmentəl/ adj. 1:05:09
工具性的
an Act of Faith phr. 1:05:09
信仰之举,没有证据的相信
friction free phr. 1:06:19
无摩擦的,毫无代价的
intrinsic nature phr. 1:06:19
内在本性
anyone's game phr. 1:07:42
谁都能玩、胜负未定的事
simulation hypothesis phr. 1:07:42
模拟假设(我们可能生活在计算机模拟中)
stripe /straɪp/ n. 1:07:42
类型,派别;of a certain stripe 某一类型的
premise /ˈpremɪs/ n. 1:08:50
前提
undermines /ˌʌndərˈmaɪnz/ v. 1:08:50
破坏,削弱
gives a bad name to phr. 1:08:50
败坏……的名声
taxpayer money phr. 1:10:12
纳税人的钱
malicious demon phr. 1:10:12
恶魔(笛卡尔怀疑论思想实验)
means to an end phr. 1:10:12
达到目的的手段
deceived /dɪˈsiːvd/ v. 1:11:29
被欺骗
road test phr. 1:11:29
路测;比喻在实际条件下检验
abrasive /əˈbreɪsɪv/ adj. 1:13:59
尖刻的,不留情面的
精读便签
下载便签 手机:长按图片也可保存
← 上一期 · NO.144College Lecture Series - Neil Postman - "The Surrender of Culture to Technology" 下一期 · NO.146 →Hegel & Marx - Bryan Magee & Peter Singer (1987)
苏菲周报 · THE WEEKLY 每周一封,
追问一个大问题。
苏菲拉底的每周来信,写这一周在追问的问题和看到的回应。
苏菲拉底
ASK THE BIG QUESTIONS · THINK DEEPLY · SEE THE WORLD DIFFERENTLY
苏菲拉底微信公众号二维码 微信公众号
© 2026 苏菲拉底 · 内容仅供学习 [email protected]