与Holden Karnofsky的对话有关未来的慈善事业

||对话

Recently, Eliezer and I had an email conversation with Holden Karnofsky to discuss future-oriented philanthropy, including MIRI. The participants were:

然后,我们将电子邮件对话编辑为简化的对话,可用这里

另请参见Miri和Holden Karnofsky之间的四次对话:存在风险, onMIRI strategy, on透明研究分析金宝博娱乐,在流动效果

约翰·贝兹(John Ba金宝博娱乐ez)研究策略

||对话

约翰·贝兹(John Baez)肖像约翰·贝兹(John Baez)是U.C.数学教授河边。直到最近,他从事高级理论和量子重力。他的互联网专栏本周的发现可以追溯到1993年,有时被称为世界第一个博客。在2010年,关注气候变化和地球的未来,他转向了更实际的主题,并开始了方位角项目,是一项国际合作,为有兴趣拯救地球的科学家和工程师创造了一个焦点。他的研究金宝博娱乐现在重点是networks信息理论,这应该有助于我们了解主导生物学和生态学的复杂系统。金宝博官方

Luke Muehlhauser:在上一次采访中,我斯科特·亚伦森问他在试图在理论研究中取得进展时发现了“对金宝博娱乐象级研究策略”,我提供了一些例子。您对斯科特和我列出的研究策略有任何评论吗?金宝博娱乐您自己的哪种推荐策略会添加到列表中?


约翰·贝兹(John Baez): What do you mean by “object-level” research tactics? I’ve got dozens of tactics. Some of them are ways to solve problems. But equally important, or maybe more so, are tactics for coming up with problems to solve: problems that are interesting but still easy enough to solve. By “object-level”, do you mean the former?
阅读更多 ”

2013 in Review: Friendly AI Research

||MIRI Strategy

这是我个人和定性的第四部分self-review of MIRI in 2013,在其中我回顾了Miri的2013年友好AI(FAI)研究活动。金宝博娱乐1

Friendly AI research in 2013

  1. 在2013年初,我们decidedto shift our priorities from research plus public outreach to a more exclusive focus on technical FAI research. This resulted in roughly as much面向公众FAI research in 2013 as in all past years combined.
  2. 此外,我们的研讨会成功地确定了雇用候选人。我们预计将在2014年上半年雇用两名2013年研讨会参与者。
  3. 在2013年期间,我了解了许多有关如何创建FAI研究所和FAI研究领域的知识。金宝博娱乐尤其是…
  4. Miri需要吸引更多经验丰富的研讨会参与者。
  5. 广泛的社区可以进行金宝博娱乐许多FAI研究,而不必将其标记为FAI研究。但是,当研究人员本身将研究视为FAI研究时,取得了更大的进步。金宝博娱乐
  6. 沟通风格非常重要。

阅读更多 ”


  1. 自然而然地,“友好的AI研究”是一个辩论问题。金宝博娱乐对于本文的大多数文章,我将假设“友好的AI研究”的意思是“ Yudkowsky认为友好的AI研金宝博娱乐究”,而情报爆炸微观经济学除外,出于本文的原因。

MIRI’s February 2014 Newsletter

||消息letters

机器情报研究所金宝博娱乐

亲爱的朋友们,

请参阅下面的新电子书,新研究,新工作空缺以及Google的新AI伦理委员会的新闻。金宝博娱乐

金宝博娱乐研究更新

新闻更新

其他更新

与往常一样,如果您有任何疑问或评论,请随时告诉我们。

最好的,
Luke Muehlhauser
执行董事

新电子书:“比我们聪明”

||消息

Smarterthanuscover-tangled-200px我们很高兴发行一本由Miri委托并由牛津大学撰写的新电子书斯图尔特·阿姆斯特朗,并在Epub,Mobi,PDF以及亚马逊和Apple电子书商店提供。

当机器比人类聪明时会发生什么?忘记伐木终结者。人工智能(AI)的力量来自其智力,而不是体力和激光枪。人类引导未来不是因为我们是最强大或最快的,而是因为我们是最聪明的。当机器比人类聪明时,我们将把方向盘交给他们​​。这些强大的机器会有什么承诺和危险?斯图尔特·阿姆斯特朗(Stuart Armstrong)的新书以清晰和机智的方式引入了这些问题。

前往imickence.org/smarter-than-us/to grab a copy.

AndréPlatzer验证网络物理系统金宝博官方

||对话

André Platzer portraitAndré Platzer是计算机科学助理教授卡内基·梅隆大学。他开发了网络物理系统的逻辑基础,以表征其基本原理,并回答我们如何信任计算机控制物理过程的问题。金宝博官方他获得了硕士学位2004年,来自德国的卡尔斯鲁赫大学(TH)和博士学位。2008年,他从德国奥尔登堡大学获得计算机科学,之后他加入CMU担任助理教授。

Dr. Platzer received a number of awards for his research on logic of dynamical systems, cyber-physical systems, programming languages, and theorem proving, including an NSF CAREER Award and ACM Doctoral Dissertation Honoroable Mention Award. He was also named one of the Brilliant 10 Young Scientists by the Popular Science magazine and one of the AI’s 10 to Watch by the IEEE Intelligent Systems magazine.

Highlights来自以下采访的普拉策的想法:

  • Keymaera是用于网络物理系统的正式验证工具。金宝博官方它看到了许多现实世界的应用程序,例如在欧洲火车控制系统金宝博官方
  • Keymaera的功能来自Platzer的“差异动态逻辑”,该逻辑基于早期工作(例如)动态逻辑
  • Contra Levitt, Platzer isn’t convinced that autonomous robots will need to use decision theory, though they may.
  • Platzer contrasts formal verification (“did I build the system right?”) with formal validation (“did I build the right system?”). One way to make progress on the latter is to develop what Rushby called “deep assumption tracking.”

阅读更多 ”

Gerwin Klein关于形式方法

||对话

Gerwin Klein肖像格温·克莱因(Gerwin Klein)是高级首席研究员金宝博娱乐NICTA, Australia’s National Centre of Excellence for ICT Research, and Conjoint Associate Professor at theUniversity of New South Wales在澳大利亚悉尼。He is leading NICTA’s Formal Methods research discipline and was the leader of the L4.verified project that created the first machine-checked proof of functional correctness of a general-purpose microkernel in 2009. He joined NICTA in 2003 after receiving his PhD from Technische Universität München, Germany, where he formally proved type-safety of the Java Bytecode Verifier in the theorem prover Isabelle/HOL.

His research interests are in formal verification, programming languages, and low-level systems. Gerwin has won a number of awards together with his team, among them the 2011 MIT TR-10 award for the top ten emerging technologies world-wide, NICTA’s Richard E. Newton impact award for the kernel verification work, the best paper award from SOSP’09 for the same, and an award for the best PhD thesis in Germany in 2003 for his work on bytecode verification. When he is not proving theorems and working on trustworthy software, he enjoys travelling and dabbling in martial arts and photography. Together with Tobias Nipkow he has just published an online draft of the text book具体语义使用Isabelletheorem prover.

Highlightsof Klein’s thoughts, from the interview below:

  • Verifying code not designed for verification is very difficult and costly. Such “post-mortem verification” has other disadvantages as well.
  • 程序设计师可以使用抽象,模块化和明确的体系结构选择,以帮助使复杂的系统与人类透明。金宝博官方
  • There are two ways probability can play a role in verification: (1) direct probabilistic reasoning, in the logic or in a setting where the program itself is probabilistic, or (2) standard non-probabilistic reasoning paired with subjectively uncertain reasoning about what a guarantee means for the overall probability that the system will work as intended.
  • 真正的自主系统似乎还有很长的路要走金宝博官方,但是如果您可以制造出真正出乎意料的行为的系统,那么您就无法使其安全。

Luke Muehlhauser:在您即将发表的论文中”全面的正式验证操作系统微粒,”您和您的合着者描述了用于确保其验证的内核设计(SEL4)。您还讨论了保持正式证明当前的过程,“随着系统的要求,设计和实施在将近十年的时间里,它[更改]。”金宝博官方

How “micro” is the seL4 microkernel? That is, what standard functionality does it provide, and what functionality (that is common in larger OS kernels) does it不是提供?


格温·克莱因(Gerwin Klein): It is pretty micro: it is a true micro-kernel in the L4 tradition. This means it is very small compared to traditional monolithic OS kernels: Linux has on the order of a few million lines of code, seL4 has on the order of 10,000 lines of C. seL4 provides only the core mechanisms that you need to build an OS. These include threads and scheduling, virtual memory, interrupts and asynchronous signals, as well as synchronous message passing. The maybe most important feature of seL4 is its strong access control mechanism that can be used together with the rest to build isolated components with controlled communication between them. These components could be small isolated device drivers or an entire Linux guest OS, the developer can choose the granularity.

理念是:如果一种机制严格需要实现特权硬件访问,则在内核中,但是如果可以使用其他机制作为用户级任务实现它,则它在内核之外。令人惊讶的功能可能出现在外面:例如,设备驱动程序,图形,文件和文件系统,网络堆栈,甚至磁盘分页。金宝博官方您(希望)得到的是,棘手的复杂性集中在内核中,用户级的操作系统组件变得更加模块化。

阅读更多 ”

与雅各布·斯坦哈特(Jacob Steinhardt)谈论美里策略

||对话,MIRI Strategy

2014年1月21日,Miri会见了Jacob Steinhardt,讨论Miri战略。参与者是:

我们记录并转录了大部分对话,然后为清晰,简洁和保护某些内容的隐私进行了编辑和解释。由此产生的编辑成绩单已完全可用这里(6页)。

Topics discussed include research communication style, inferential distance, FAI research directions, and Knightian uncertainty.