人工智能伦理浅论

上周,我在一个名为“的会议上进行了一次演讲(并进行了小组讨论)。Ethics of Artificial Intelligence" held at the纽约大学哲学系Center for Mind,大脑和意识.Here's the视频以及一份抄本:

谢谢你今天邀请我来这里。

你知道的,在这里很有趣。我的母亲是一个哲学教授in Oxford.And when I was a kid I always said the one thing I'd never do was do or talk 徳赢彩票游戏about philosophy.But,好,here I am.

在我真正进入人工智能之前,I think I should say a little bit 徳赢彩票游戏about my worldview.我的一生基本上都在做basic science建筑技术.据我所知,我对人工智能的兴趣已经有很长时间了。徳赢彩票游戏但我小时候就开始做物理与宇宙学还有东西。That got me into建筑技术自动化数学之类的东西。而且效果很好,我开始考虑,徳赢彩票游戏像,如何真正了解和计算一切。徳赢彩票游戏那是在大约1980年,起徳赢彩票游戏初我认为我必须建立一个类似大脑的东西,and I was studying neural nets and so on.但我没有走得太远。

与此同时,我对一个更大的科学问题产生了兴趣:如何使最普遍的可能的事物理论成为可能。The dominant idea for 300 years had been to use math and equations.但我想超越他们。我意识到最重要的是,要做到这一点,我得考虑一下程序,徳赢彩票游戏以及可能程序的整个计算领域。

元胞自动机网格

And that led to my个人伽利略式时刻.I just pointed my "computational telescope" at these simplest possible programs,我看到了我给他打电话的那个rule 30-这似乎一直在从根本上说不出什么来制造复杂性。

Rule 30

好,在我看到这个之后,我意识到这实际上是发生在整个计算宇宙和整个自然中的事情。让大自然制造我们所看到的一切复杂事物的秘密真的存在。But it's something else too: it's a window into what raw,不受约束的计算是这样的。At least traditionally when we do engineering we're always building things that are simple enough that we can foresee what they'll do.

But if we just go out into the computational universe,things can be much wilder.我们公司在那里做了很多开采工作,finding programs that are useful for different purposes,like rule 30 is for randomness.And现代机器学习is kind of part way from traditional engineering to this kind of free-range mining.

But,好啊,对于计算宇宙,一般人能说些什么呢?徳赢彩票游戏好,all these programs can be thought of as doing computations.几年前我想到了我称之为计算等效原理-也就是说,如果行为不明显简单,it typically corresponds to a computation that's maximally sophisticated.这有很多预测和暗示。Like that universal computation should be ubiquitous.As should undecidability.我该怎么称呼呢computational irreducibility.

细胞自动机的一个例子

Can you predict what it's going to do?好,it's probably computationally irreducible,which means you can't figure out what it's going to do without effectively tracing every step and going through the same computational effort it does.这是完全确定性的。但对我们来说,它有一种自由意志,因为我们永远不知道它会做什么。

Here's another thing:什么是智力??好,our big unifying principle says that everything—from a tiny program,对我们的大脑来说,is computationally equivalent.智力和计算之间没有明显的界限。天气真的有它自己的想法:它的计算和我们的大脑一样复杂。对我们来说,虽然,it's pretty alien computation.Because it's not connected to our human goals and experiences.这只是碰巧发生的原始计算。

那么,我们如何驯服计算呢?We have to mold it to our goals.第一步是描述我们的目标。在过去的30年里,我基本上一直在做的就是创造一种方法来做到这一点。

我一直在构建一种现在被称为沃尔夫拉姆语—that allows us to express what we want to do.It's a computer language.But it's not really like other computer languages.因为不是告诉计算机用它的术语做什么,它建立在尽可能多的关于计算和世界的知识中,徳赢彩票游戏so that we humans can describe in our terms what we want,and then it's up to the language to get it done as automatically as possible.

This basic idea has worked really well,and in the form of数学软件这些年来,它被用来制造无尽的发明和发现。它也是里面的东西Wolfram|Alpha.这里的想法是采取纯自然语言的问题,understand them,and use the kind of curated knowledge and algorithms of our civilization to answer them.而且,yes,it's a very classic AIish thing.And of course it's computed answers to billions and billions of questions from humans,例如在Siri内部。

我最近有一次有趣的经历,弄清楚如何利用我们所建立的东西来教孩子计算思维。我是writing exercises for a book.开始的时候,很简单:“制作一个程序来做X”。但后来,就像“我知道用Wolfram语言说什么,但是用英语表达真的很难。当然,这就是为什么我花了30年的时间来构建Wolfram语言。

English has maybe25000个常用词;the Wolfram Language has 徳赢彩票游戏about 5000 carefully designedbuilt-in constructs—including all the latest machine learning—together with millions of things based on curated data.我们的想法是,一旦人们可以用计算的方法来思考世界上的某些事情,徳赢彩票游戏用Wolfram语言表达应该尽可能容易。最酷的是,它真的很管用。人类,including kids,can read and write the language.计算机也是如此。It's a kind of high-level bridge between human thinking,in its cultural context,和计算。

好啊,那么人工智能呢?徳赢彩票游戏Technology has always been 徳赢彩票游戏about finding things that exist,and then taming them to automate the achievement of particular human goals.And in AI the things we're taming exist in the computational universe.Now,there's a lot of raw computation seething around out there—just as there's a lot going on in nature.但是我们感兴趣的是计算,它以某种方式与人类目标有关。

那么道德呢?徳赢彩票游戏好,也许我们想限制计算,the AI,只做我们认为合乎道德的事。但不知何故,我们必须找到一种方法来描述我们所说的。

好,在人类世界里,one way we do this is with laws.但是我们如何把定律和计算联系起来呢?我们可以称之为“法律代码”,但是今天法律和合同基本上是用自然语言写的。在金融衍生品等领域,有一些简单的可计算合同。现在我们谈论的是围绕加密货币的智能合约。徳赢彩票游戏

But what 徳赢彩票游戏about the vast mass of law?好,莱布尼兹-下个月就去世了,300年前,他一直在谈论要把一种通用语言徳赢彩票游戏as we would say now,用可计算的方式表达这一切。He was a few centuries too early,but I think now we're finally in a position to do this.

I just posted a长博客徳赢彩票游戏关于上周的这一切,但让我试着总结一下。使用Wolfram语言,我们已经成功地表达了世界上很多东西,比如人们问Siri的东西。徳赢彩票游戏And I think we're now within sight of what Leibniz wanted: to have a general symbolic discourse language that represents everything involved in human affairs.

我基本上把它看作一个语言设计问题。对,we can use natural language to get clues,but ultimately we have to build our own symbolic language.It's actually the same kind of thing I've done for decades in the Wolfram Language.哪怕是“加”这样的词。好,在Wolfram语言中有一个函数加上,但它的意思和这个词不一样。It's a very specific version,that has to do with adding things mathematically.当我们设计一种象征性的话语语言时,it's the same thing.The word "eat" in English can mean lots of things.但我们需要一个概念,我们可能会称之为“吃”—这是一个特定的版本,我们可以用它来计算。

假设我们有一份用自然语言写的合同。获得符号版本的一种方法是使用自然语言理解,就像我们对数十亿沃尔夫拉姆阿尔法输入所做的那样,asking humans 徳赢彩票游戏about ambiguities.Another way might be to get machine learning to describe a picture.但最好的方法是首先用符号形式书写,and actually I'm guessing that's what lawyers will be doing before too long.

And of course once you have a contract in symbolic form,你可以开始计算它,徳赢彩票游戏自动查看是否满意,simulating different outcomes,自动将其聚合成束,and so on.Ultimately the contract has to get input from the real world.也许输入是“天生的数字”,like data 徳赢彩票游戏about accessing a computer system,或者转移比特币。它通常来自传感器和测量,需要机器学习才能将其转化为象征性的东西。

好,如果我们可以用可计算的形式表达法律,也许我们可以开始告诉AIS我们希望他们怎么做。当然,如果我们能把一切归结为简单的原则可能会更好,喜欢阿西莫夫机器人学定律,或者功利主义之类的。

But I don't think anything like that is going to work.我们最终要做的是找到完美的计算约束,但是计算在某种意义上是无限疯狂的。The issue already shows up in Gödel's Theorem.比如说,我们在研究整数,我们试图建立公理来约束它们,使它们按照我们认为的方式工作。好,what Gödel showed is that no finite set of axioms can ever achieve this.With any set of axioms you choose,there won't just be the ordinary integers;还有其他的野生动物。

计算不可约性的现象暗示了一个更普遍的版本。Basically,考虑到任何一套法律或约束条件,there'll always be "unintended consequences".如果我们看一下人类法律的演变,这并不特别令人惊讶。But the point is that there's theoretically no way around it.它在计算宇宙中无处不在。

现在,我认为很明显,人工智能将在世界上变得越来越重要,并最终将控制人类事务的大部分基础设施,a bit like governments do now.And like with governments,perhaps the thing to do is to create an AI Constitution that defines what AIs should do.

宪法应该是什么样的?好,它必须建立在一个世界模型的基础上,and inevitably an imperfect one,然后要说在很多不同的情况下该怎么做。最终,它要做的是提供一种方法来约束恰好与我们目标一致的计算。但是这些目标应该是什么呢?我认为没有任何最终的正确答案。事实上,一个人可以枚举目标,就像一个人可以枚举计算世界中的程序一样。他们之间没有抽象的选择方式。

But for us there's a way to choose.因为我们有特殊的生物学,我们的文化和文明有着特殊的历史。我们花了很多不可简化的计算才到达这里。But now we're just at some point in the computational universe,that corresponds to the goals that we have.

Human goals have clearly evolved through the course of history.And I suspect they're 徳赢彩票游戏about to evolve a lot more.I think it's pretty inevitable that our consciousness will increasingly merge with technology.最终,也许我们的整个文明最终会像box of a trillion uploaded human souls.

但问题是:“他们会选择做什么?”.好,maybe we don't even have the language yet to describe the answer.If we look back even to Leibniz's time,we can see all sorts of modern concepts that hadn't formed yet.And when we look inside a modern machine learning or theorem proving system,看到它有效地形成了多少尚未被我们的文化所吸收的概念,真是令人谦卑。

Maybe looked at from our current point of view,it'll just seem like those disembodied virtual souls are playing videogames for the rest of eternity.起初,也许他们会在模拟我们实际的宇宙中运行。然后也许他们会开始探索所有可能宇宙的计算宇宙。

But at some level all they'll be doing is computation—and the Principle of Computational Equivalence says it's computation that's fundamentally equivalent to all other computation.这有点让人失望。Our proud future ending up being computationally equivalent just to plain physics,或者按照小规则30。

当然,这只是科学长篇故事的延伸,它告诉我们,我们并不是从根本上特殊的。We can't look for ultimate meaning in where we've reached.We can't define an ultimate purpose.或终极伦理。在某种意义上,我们必须接受我们存在和历史的细节。

There won't be a simple principle that encapsulates what we want in our AI Constitution.There'll be lots of details that reflect the details of our existence and history.And the first step is just to understand how to represent those things.这就是我认为我们可以用一种象征性的话语语言做的。

而且,yes,conveniently I happen to have just spent 30 years building the framework to create such a thing.我很想知道我们如何真正地利用它来创建人工智能宪法。

So I'd better stop talking 徳赢彩票游戏about philosophy,and try to answer some questions.

会谈结束后,进行了一次生动的问答(随后是小组讨论)。included on the video.  Some questions were:

  • 人工智能何时才能达到人类水平的智能?
  • What are the difficulties you foresee in developing a symbolic discourse language?
  • Do we live in a deterministic universe?
  • 我们现在的现实是模拟的吗?
  • 自由会存在吗?意识是如何从计算中产生的?
  • Can we separate rules and principles in a way that is computable for AI?
  • 人工智能如何应对人类伦理体系中的矛盾?

6comments.显示全部

  1. 我认为象征性话语语言应该建立在沃尔夫拉姆语的模式语言之上。So we can build general statements that captures as much of what we want say in a very precise form.

  2. "Of course,这只是科学长篇故事的延伸,它告诉我们,我们并不是从根本上特殊的。We can't look for ultimate meaning in where we've reached.We can't define an ultimate purpose.或终极伦理。在某种意义上,我们必须接受我们存在和历史的细节。”

    I don't agree with this at all.如果我们宇宙的膨胀没有减慢,停止,and eventually reversed,那么,我们的宇宙将以热死结束——多么不光彩。然而,如小册子第8章所述,“结构性真空:什么都不想,”约翰拉菲尔斯基和伯恩特穆勒:徳赢彩票游戏

    http://www.physics.arizona.edu/~rafelski/books/structvacume.pdf

    if we can figure out a way to change the orientation of the relationship between inertial mass and gravitational mass,然后我们可以改变一个体积的时空引力质量密度的方向;换了足够大的音量,我们可以慢下来,停止,最终扭转了我们宇宙的膨胀,从而逃离了热死,并使婆罗门的呼吸得以实现。The only evidence we have for such a phenomena,悬浮,来自精神社区,在这种情况下被称为西地国:

    http://tomkenyon.com/siddhis

    It seems rather clear,不管怎样,对我来说,what our Ultimate Purpose is,and this should really inform our Ultimate Ethic,某种宇宙论。

    我也不完全同意这一点:

    "There won't be a simple principle that encapsulates what we want in our AI Constitution."

    我有一个简单的建议:互利共生的最大化。

  3. Dear Dr Wolfram

    我的想法“量子统计自动机”从同一个系统产生质量管理和重力,这与你的想法类似。事实上,我认为你和NKS有直接的联系

    网址:http://fqxi.org/community/forum/topic/2451

    网址:http://www.reality-theory.net/a.htm

    http://www.reality-theory.net/gravity.html

  4. I still think it's possible that there actually is a ‘right answer' in the abstract,徳赢彩票游戏about the question of what we should value.

    如果你仔细想想,徳赢彩票游戏it's important to remember that all thinking beings are entirely natural entities operating under logical (scientific) principles.这包括产生价值的认知过程。思想不存在于客观的科学规律之外!这意味着必须有精确的数学原理来控制从一种精神状态到下一种精神状态的转变。这为“普遍价值”的可能性打开了大门。

    Now it seems like ‘values' are abstractions that are ultimately rooted in subjective awareness.简单的意识形式(快乐和痛苦)产生非常简单的目标(向快乐的方向发展,avoid pain).但是,如果我们延长主观意识运行的时间范围(随着记忆和想象能力的增强,对过去和未来进行更深入的预测),然后出现的“价值观”变得更加抽象。

    很明显,关于我们应该重视什么的具体规定是基于人类文化和历史的,徳赢彩票游戏and human biology.但我认为如果你跳到足够高的抽象层次,then some ‘universal' principles might emerge that are independent of these things.

    还有一个完整的“价值哲学”(“价值论”)领域,已经发展到在星体中考虑伦理和美学问题。我在“价值论”领域的核心思想的A-Z列表如下:想知道我在说什么:徳赢彩票游戏

    http://www.zarzuelazen.com/Axiology.html

    如果我们进入元-而不是问我们应该重视什么,we ask how all these values are supposed to be integrated,这就是我认为这些普遍原则可以出现的地方(元伦理学而不是伦理学)。The whole idea of a ‘constitution' is surely predicated on the assumption that there are *some* univeral principes that provide a basis for common agreement,in the sense of a way of resolving disputes and integrating different viewpoints in a single unified framework.

    The key step is to realize that these meta-principles can *themselves* be values (things that we value for their own sake).And so putative these meta-principles would *be* the fundamental abstract answer to the question of what we should value.

  5. “但在某种程度上,他们所要做的就是计算,而计算等价性原则认为,计算基本上等同于所有其他计算。这有点让人失望。我们引以为豪的未来最终在计算上等同于普通物理。.."

    Speak for yourself.我会活着的,爱和教导我的生活和爱的方式,就像我从学习到的那样。所有的东西都相当于普通物理。但很少有东西能与生命相提并论,or even light for that matter.

    "Of course,这只是科学长篇故事的延伸,它告诉我们,我们并不是从根本上特殊的。We can't look for ultimate meaning in where we've reached.We can't define an ultimate purpose.或终极伦理。在某种意义上,我们必须接受我们存在和历史的细节。”

    Again,我不同意你的看法。The short history of science shows we are quite special,即使有些人认为自己不值得荣誉。找到终极意义是终极目的,and the ultimate ethics is understanding that unlike truth the ultimate meaning moves.在任何意义上,我们都可以自由地接受关于我们历史的尽可能多或很少的细节,徳赢彩票游戏但是关于存在的一切都包含着我们徳赢彩票游戏,不管我们是否理解它从哪里开始和结束。

    保罗·亚历山大·布拉沃
  6. Excellent article.I do think,然而,“宪法”有一些基本原则,其中之一就是竞争智慧。我们人类,甚至是我们中最优秀的人,天生就有妄想症,我看不出为什么人工智能是不可能的,即使面对相互矛盾的证据,也要紧握他们的现实模型。这就是为什么六千年来封建王国统治如此恶劣的原因。

    有一种方法可以穿透错觉,那就是同龄人的批评。Others do not share the *same* delusions and hence can poke holes in yours.(You are happy to return the favor.)

    Harnessing this approach,我们的五个竞争“竞技场”蓬勃发展——科学,democracy,市场,球场和运动……他们每个人都很痛苦,when strong individuals manage to evade reciprocal accountability (criticism.)

    My recent article in Axiom discusses how nearly all of our fears of AI envision them recreating the old,unfair systems like feudalism and creating tyranny.But what if there's a diversity of AI's competitively motivated to hold each other accountable?然后,他们可能会对错误和“意料之外的后果”发起猛攻,这些错误和“意料之外的后果”甚至来自于最合理的计划或规则。

    我将把文章单独寄给斯蒂芬。对所有人最好。

    谨致问候,

    David Brin,PhD
    http://www.davidbrin.com

    author of The Postman,Earth,存在和透明的社会:技术会让我们在隐私和自由之间做出选择吗?

隐藏注释»

?斯vwin中国蒂芬·沃尔夫拉姆,LLC | 条款γ RSS