This is part of a semi-monthly reading group on Eliezer Yudkowsky's ebook, Rationality: From AI to Zombies. For more information about the group, see the announcement post.

Welcome to the Rationality reading group. This week we discuss the sequence Fake Beliefs which introduces the concept of belief in belief and demonstrates the phenomenon in a number of contexts, most notably as it relates to religion. This sequence also foreshadows the mind-killing effects of tribalism and politics, introducing some of the language (e.g. Green vs. Blue) which will be used later.

This post summarizes each article of the sequence, linking to the original LessWrong posting where available, and offers a few relevant notes, thoughts, and ideas for further investigation. My own thoughts and questions for discussion are in the comments.

Reading: Sequence B: Fake Beliefs (p43-77)

B. Fake Beliefs

11. Making beliefs pay rent (in anticipated experiences). Beliefs networks which have no connection to anticipated experience we call “floating” beliefs. Floating beliefs provide no benefit as they do not constrain predictions in any way. Ask about a belief what you expect to see, if the belief is true. Or better yet what you expect not to see: what evidence would falsify the belief. Every belief should flow to a specific guess of anticipation, and should continue to pay rent in future anticipations. If a belief turns deadbeat, evict it. (p45-48)

12. A fable of science and politics. Cautions, though a narrative story, the dangers of that come from feeling attachment to beliefs. Introduces the Greens vs Blues, a fictional debate illustrating the biases which emerge from the tribalism of group politics. (p49-53)

13. Belief in belief. Through the story of someone who claims a dragon lives in their garage, a invisible, inaudible, impermeable dragon which defies all attempts at detection, we are introduced to the concept of belief in belief. The dragon claimant believes that there is a fire-breathing flying animal in his garage, but simultaneously expects to make no observations that would confirm that belief. The belief in belief turns into a form of mental jujutsu where mental models are transfigured in the face of experiment so as to predict whatever would be expected if the belief were not, in fact, true. (p54-58)

14. Bayesian judo. A humorous story illustrating the inconsistency of belief in belief, and the mental jujutsu required to maintain such beliefs. (p59-60)

15. Pretending to be wise. There's a difference between: (1) passing neutral judgment; (2) declining to invest marginal resources in investigating the sides of a debate; and (3) pretending that either of the above is a mark of deep wisdom, maturity, and a superior vantage point. Propounding neutrality is just as attackable as propounding any particular side. (p61-64)

16. Religion's claim to be non-disprovable. It is only a recent development in Western thought that religion is something which cannot be proven or disproven. Many examples are provided of falsifiable beliefs which were once the domain of religion. (p65-68)

17. Professing and cheering. Much of modern religion can be thought of as communal profession of belief – actions and words which signal your belief to others. (p69-71)

18. Belief as attire. It is very easy for a human being to genuinely, passionately, gut-level belong to a group. Identifying with a tribe is a very strong emotional force. And once you get people to identify with a tribe, the beliefs which are attire of that tribe will be spoken with the full passion of belonging to that tribe. (p72-73)

19. Applause lights. Sometimes statements are made in the form of proposals when themselves present no meaningful suggestion, e.g. “We need to balance the risks and opportunities of AI.” It's not so much a propositional statement, as the equivalent of the “Applause” light that tells a studio audience when to clap. Most applause lights can be detected by a simple reversal test: “We shouldn't balance the risks and opportunities of AI.” Since the reversal sounds abnormal, the unreversed statement is probably normal, implying it does not convey new information. (p74-77)

This has been a collection of notes on the assigned sequence for this week. The most important part of the reading group though is discussion, which is in the comments section. I pose some questions for you there, and I invite you to add your own. Please remember that this group contains a variety of levels of expertise: if a line of discussion seems too basic or too incomprehensible, look around for one that suits you better!

The next reading will cover Sequence C: Noticing Confusion (p79-114). The discussion will go live on Wednesday, 20 May 2015 at or around 6pm PDT (hopefully), right here on the discussion forum of LessWrong.

This is part of a semi-monthly reading group on Eliezer Yudkowsky's ebook, Rationality: From AI to Zombies. For more information about the group, see the announcement post.

Welcome to the Rationality reading group. This week we discuss the Preface by primary author Eliezer Yudkowsky, Introduction by editor & co-author Rob Bensinger, and the first sequence: Predictably Wrong. This sequence introduces the methods of rationality, including its two major applications: the search for truth and the art of winning. The desire to seek truth is motivated, and a few obstacles to seeking truth--systematic errors, or biases--are discussed in detail.

This post summarizes each article of the sequence, linking to the original LessWrong posting where available, and offers a few relevant notes, thoughts, and ideas for further investigation. My own thoughts and questions for discussion are in the comments.

Reading: Preface, Biases: An Introduction, and Sequence A: Predictably Wrong (pi-xxxv and p1-42)

Introduction

Preface. Introduction to the ebook compilation by Eliezer Yudkowsky. Retrospectively identifies mistakes of the text as originally presented. Some have been corrected in the ebook, others stand as-is. Most notably the book focuses too much on belief, and too little on practical actions, especially with respect to our everyday lives. Establishes that the goal of the project is to teach rationality, those ways of thinking which are common among practicing scientists and the foundation of the Enlightenment, yet not systematically organized or taught in schools (yet).

Biases: An Introduction. Editor & co-author Rob Bensinger motivates the subject of rationality by explaining the dangers of systematic errors caused by *cognitive biases*, which the arts of rationality are intended to de-bias. Rationality is not about Spock-like stoicism -- it is about simply "doing the best you can with what you've got." The System 1 / System 2 dual process dichotomy is explained: if our errors are systematic and predictable, then we can instil behaviors and habits to correct them. A number of exemplar biases are presented. However a warning: it is difficult to recognize biases in your own thinking even after learning of them, and knowing about a bias may grant unjustified overconfidence that you yourself do not fall pray to such mistakes in your thinking. To develop as a rationalist actual experience is required, not just learned expertise / knowledge. Ends with an introduction of the editor and an overview of the organization of the book.

A. Predictably Wrong

1. What do I mean by "rationality"? Rationality is a systematic means of forming true beliefs and making winning decisions. Probability theory is the set of laws underlying rational belief, "epistemic rationality": it describes how to process evidence and observations to revise ("update") one's beliefs. Decision theory is the set of laws underlying rational action, "instrumental rationality", independent of what one's goals and available options are. (p7-11)

2. Feeling rational. Becoming more rational can diminish feelings or intensify them. If one cares about the state of the world, it is expected that he or she should have an emotional response to the acquisition of truth. "That which can be destroyed by the truth should be," but also "that which the truth nourishes should thrive." The commonly perceived dichotomy between emotions and "rationality" [sic] is more often about fast perceptual judgements (System 1, emotional) vs slow deliberative judgements (System 2, "rational" [sic]). But both systems can serve the goal of truth, or defeat it, depending on how they are used. (p12-14)

3. Why truth? and... Why seek the truth? Curiosity: to satisfy an emotional need to know. Pragmatism: to accomplish some specific real-world goal. Morality: to be virtuous, or fulfill a duty to truth. Curiosity motivates a search for the most intriguing truths, pragmatism the most useful, and morality the most important. But be wary of the moral justification: "To make rationality into a moral duty is to give it all the dreadful degrees of freedom of an arbitrary tribal custom. People arrive at the wrong answer, and then indignantly protest that they acted with propriety, rather than learning from their mistake." (p15-18)

4. ...what's a bias, again? A bias is an obstacle to truth, specifically those obstacles which are produced by our own thinking processes. We describe biases as failure modes which systematically prevent typical human beings from determining truth or selecting actions that would have best achieved their goals. Biases are distinguished from mistakes which originate from false beliefs or brain injury. Do better seek truth and achieve our goals we must identify our biases and do what we can to correct for or eliminate them. (p19-22)

5. Availability. The availability heuristic is judging the frequency or probability of an event by the ease with which examples of the event come to mind. If you think you've heard about murders twice as much as suicides then you might suppose that murder is twice as common as suicide, when in fact the opposite is true. Use of the availability heuristic gives rise to the absurdity bias: events that have never happened are not recalled, and hence deemed to have no probability of occurring. In general, memory is not always a good guide to probabilities in the past, let alone to the future. (p23-25)

6. Burdensome details. The conjunction fallacy is when humans rate the probability of two events has higher than the probability of either event alone: adding detail can make a scenario sound more plausible, even though the event as described necessarily becomes less probable. Possible fixes include training yourself to notice the addition of details and discount appropriately, thinking about other reasons why the central idea could be true other than the added detail, or training oneself to hold a preference for simpler explanations -- to feel every added detail as a burden. (p26-29)

7. Planning fallacy. The planning fallacy is the mistaken belief that human beings are capable of making accurate plans. The source of the error is that we tend to imagine how things will turn out if everything goes according to plan, and do not appropriately account for possible troubles or difficulties along the way. The typically adequate solution is to compare the new project to broadly similar previous projects undertaken in the past, and ask how long those took to complete. (p30-33)

8. Illusion of transparency: why no one understands you. The illusion of transparency is our bias to assume that others will understand the intent behind our attempts to communicate. The source of the error is that we do not sufficiently consider alternative frames of mind or personal histories, which might lead the recipient to alternative interpretations. Be not too quick to blame those who misunderstand your perfectly clear sentences, spoken or written. Chances are, your words are more ambiguous than you think. (p34-36)

9. Expecting short inferential distances. Human beings are generally capable of processing only one piece of new information at at time. Worse, someone who says something with no obvious support is a liar or an idiot, and if you say something blatantly obvious and the other person doesn't see it, they're the idiot. This is our bias towards explanations of short inferential distance. A clear argument has to lay out an inferential pathway, starting from what the audience already knows or accepts. If at any point you make a statement without obvious justification in arguments you've previously supported, the audience just thinks you're crazy. (p37-39)

10. The lens that sees its own flaws. We humans have the ability to introspect our own thinking processes, a seemingly unique skill among life on Earth. As consequence, a human brain is able to understand its own flaws--its systematic errors, its biases--and apply second-order corrections to them. (p40-42)

It is at this point that I would generally like to present an opposing viewpoint. However I must say that this first introductory sequence is not very controversial! Educational, yes, but not controversial. If anyone can provide a link or citation to one or more decent non-strawman arguments which oppose any of the ideas of this introduction and first sequence, please do so in the comments. I certainly encourage awarding karma to anyone that can do a reasonable job steel-manning an opposing viewpoint.

This has been a collection of notes on the assigned sequence for this week. The most important part of the reading group though is discussion, which is in the comments section. I pose some questions for you there, and I invite you to add your own. Please remember that this group contains a variety of levels of expertise: if a line of discussion seems too basic or too incomprehensible, look around for one that suits you better!

The next reading will cover Sequence B: Fake Beliefs (p43-77). The discussion will go live on Wednesday, 6 May 2015 at or around 6pm PDT, right here on the discussion forum of LessWrong.

Update: When I posted this announcement I remarkably failed to make the connection that the April 15th is tax day here in the US, and as a prime example of the planning fallacy (a topic of the first sequence!), I failed to anticipate just how complicated my taxes would be this year. The first post of the reading group is basically done but a little rushed, and I want to take an extra day to get it right. Expect it to post on the next day, the 16th

On Thursday, 16 April 2015, just under a month out from this posting, I will hold the first session of an online reading group for the ebook Rationality: From AI to Zombies, a compilation of the LessWrong sequences by our own Eliezer Yudkowsky. I would like to model this on the very successful Superintelligence reading group led by KatjaGrace. This is an advanced warning, so that you can have a chance to get the ebook, make a donation to MIRI, and read the first sequence.

The point of this online reading group is to join with others to ask questions, discuss ideas, and probe the arguments more deeply. It is intended to add to the experience of reading the sequences in their new format or for the first time. It is intended to supplement discussion that has already occurred the original postings and the sequence reruns.

The reading group will 'meet' on a semi-monthly post on the LessWrong discussion forum. For each 'meeting' we will read one sequence from the the Rationality book, which contains a total of 26 lettered sequences. A few of the sequences are unusually long, and these might be split into two sessions. If so, advance warning will be given.

In each posting I will briefly summarize the salient points of the essays comprising the sequence, link to the original articles and discussion when possible, attempt to find, link to, and quote one or more related materials or opposing viewpoints from outside the text, and present a half-dozen or so question prompts to get the conversation rolling. Discussion will take place in the comments. Others are encouraged to provide their own question prompts or unprompted commentary as well.

We welcome both newcomers and veterans on the topic. If you've never read the sequences, this is a great opportunity to do so. If you are an old timer from the Overcoming Bias days then this is a chance to share your wisdom and perhaps revisit the material with fresh eyes. All levels of time commitment are welcome.

If this sounds like something you want to participate in, then please grab a copy of the book and get started reading the preface, introduction, and the 10 essays / 42 pages which comprise Part A: Predictably Wrong. The first virtual meeting (forum post) covering this material will go live before 6pm Thursday PDT (1am Friday UTC), 16 April 2015. Successive meetings will start no later than 6pm PDT on the first and third Wednesdays of a month.

Following this schedule it is expected that it will take just over a year to complete the entire book. If you prefer flexibility, come by any time! And if you are coming upon this post from the future, please feel free leave your opinions as well. The discussion period never closes.

Topic for the first week is the preface by Eliezer Yudkowsky, the introduction by Rob Bensinger, and Part A: Predictably Wrong, a sequence covering rationality, the search for truth, and a handful of biases.