Friday, November 1, 2013


This article appears in slightly different form in the Nov/Dec 2013 issue of IEEE Micro © 2013 IEEE.

This time I look at two books that provide insight into how to solve technical design problems.

Design Appreciation

Design for Hackers: /Reverse-Engineering Beauty by David Kadavy (Wiley, UK, 2011, 352pp ISBN 978-1-119-99895-2,, $39.95)

David Kadavy is a user interface designer working in Silicon Valley. Despite the general title, he has written a book about visual design. He sets the following goal for this book: 
If you want to learn to create great design yourself, if you want to gain design literacy, there simply is no way to do so with lists of rules. Instead, I want to provide you with a new set of eyes through which you can see the world anew.
This is the sort of goal a course in art appreciation or, changing eyes to ears, music appreciation might aim at. Knowing you (his audience), Kadavy uses the term "reverse engineering" to get your attention. Once he has your attention, however, he gives you a tour of subjects you didn't know you were interested in.

Kadavy begins with the Pantheon (not the Parthenon). Having been there, I can attest that it doesn't look particularly distinguished as you approach it from the streets of Rome. Once inside, most visitors are taken aback. If you have never seen the building, find an online reproduction of eighteenth century painter Giovanni Paolo Panini's The Interior of the Pantheon, and spend some time admiring it and taking in the geometric relationships. Proportions and the harmony of geometric shapes are design elements Kadavy wants your new eyes to see.

Built in 126 by the Emperor Hadrian, the Pantheon is meant as a monument to all of the Roman gods and designed to inspire awe in the mortals who enter. Its 142-foot spherical dome is the largest unreinforced concrete dome in the world. At the peak of the dome, a circular opening 30 feet in diameter provides natural lighting for the interior. As you enter, Kadavy says, "you're enveloped by another world, complete with its own sun." Kadavy explains how Hadrian's objectives combined with significant engineering challenges to produce the widely admired, frequently imitated design. For example, the square recesses in the dome not only create a striking pattern, but also help to reduce the weight of the dome to make it more supportable.

Kadavy tells how, when he was in Rome studying the origins of modern typography, he would go to the Pantheon just to watch people's reactions as they entered. This interest in how designs affect people is one of the reasons his book is so informative. Unfortunately, his interest in the origins of typography leads him to say more on the subject than his theme requires, so you might wish to skim that chapter. My first draft of my book Inside BASIC Games (Sybex, 1981) started with a lot of fascinating but barely relevant material, including an elaborate metaphor for computer programming that involved a player piano. My editor, the great Salley Oberlin, persuaded me to toss out the first 130 pages, and I've never been sorry. I think this book could have benefited from similar advice. Kadavy has excellent points to make about type, but he didn't have to go back to cave painting to lay the groundwork for them. Similarly, he could have gone directly to his points about color without rehashing the philosophy and physics of electromagnetic radiation or repeating information about perception that is handled more thoroughly elsewhere.

The publishers use the back cover to focus attention on Kadavy's discussions of why designers hate the Comic Sans font and why the Golden Ratio is no more useful in laying out designs than simpler ratios like 3:4. These are interesting topics, but marginal. The main topic of the book is visual design, and Kadavy's main point is that design occurs at all levels of a product. An elegant veneer added after the fact is not necessarily good design. It may just be lipstick on a pig. The pattern of square recesses in the dome of the Pantheon is good design, because it arises from the engineering needs and the objectives of the building. The Apple Aqua interface is good design because it exploits previously unavailable technology and responds to cultural forces and user expectations in ways that Kadavy explains.

Kadavy uses the metaphor of learning to dance. You need to learn the basic moves before you can tie them together. Kadavy tries to teach the individual moves (for example, using white space or exploiting type characteristics) by providing examples that vary one design move while holding others constant. This can make for tedious reading, but if you persist, you can learn the moves  -- and then start to dance.

One topic that lives up to the promise of the book's title is Kadavy's discussion of search engine optimization (SEO). He ties it to visual design by saying that design has always been about conveying information to the right audience. I would have accepted that premise and gone on, but he backs it up by digressing on what Aldus Manutius was doing in 1501 and what Jan Tschichold wrote in 1928.

Despite Kadavy's maddening propensity for cluttering his narrative with digressions -- something I'm sure he would never do in a visual design -- this is an interesting and informative book. I recommend it.

Design Recipes

101 Design Ingredients to Solve Big Tech Problems by Eewei Chen (Pragmatic Bookshelf, Dallas TX, 2013, 290pp, ISBN 978-1-93778-532-1,, $36.00)

The book jacket describes Eewei Chen as a digital tech strategist, former team leader at Microsoft, creative director at Conchango and at ThoughtWorks Europe, and more. Elsewhere ( he describes himself as a technology futurist, agilest, and lean design philosopher who has worked in the new media creative industry since 1993. He is apparently a very busy person, which, along with his lean design philosophy, may explain why he wrote such a concise handbook. He provides 101 ingredients, each on a two page spread that follows a precise format. He follows these with ten recipes. Their format is also constrained. Each is four or five pages.

One of the key reasons this book communicates effectively is its illustrator, Robert André. For each design ingredient and recipe, he provides a one-page illustration that conveys the basic idea, enabling Chen to describe it in a few words, accompanied by links to other material. Of the 266 pages in the main body of the book, 111 are devoted to André's illustrations.

Each ingredient has a captioned illustration on the left page and four elements on the right page: a quote, a sentence labeled The Problem, a section labeled The Solution, and a section of footnotes consisting almost entirely of hyperlinks. The solution section consists of a short paragraph followed by three bullets. Each bullet contains a pithy title followed by a brief explanation. Most have a footnote.

Each recipe has an illustration, then lists six ingredients taken from the list of 101. Each ingredient is accompanied by an icon derived from its illustration. The six recipe steps correspond exactly to the six ingredients. Each recipe ends with a section titled Tips on How to Apply This Recipe. Each ingredient is a principle (for example, Lead by Example), not necessarily tied to a specific company. Each recipe is associated with a company. For example, the recipe for effective leadership focuses on Jeff Bezos's leadership of Amazon. One of that recipe's ingredients is Lead by Example. The recipe for world domination consists of Chen's idea of the six ingredients of Google's success, though he seems to be undecided about the sixth ingredient.

Many people have noted that a rigid format (for example, a sonnet, a limerick, or a haiku) leaves plenty of room for creativity. However, the structure Chen has chosen makes it hard to tie things together. For example, ingredient 60 is Check the Data. On the left page is an illustration of a teacup with a teabag steeping in it. The caption says 
"People may be taking a break, not stumbling across a barrier." 
The right page begins with a quote from Albert Einstein: 
"Not everything that can be counted counts, and not everything that counts can be counted." 
The problem is stated as
 "Companies don't monitor customer usage closely enough to see what's really going on." 
The solution is stated as 
"Analyze and interpret data as part of the design and build process. Shed light on uncertainties, especially if you aren't sure why they really exist." 
The three bullets that elaborate on this solution are as follows:

  • Have assumptions to test. There is no point to looking at data if you do not know what you are looking for to start with. List your biggest assumptions and measure success by seeing if they have led to improvements to key performance indicators or success metrics. The footnote leads to a page on the Advanced Performance Institute website that defines and discusses key performance indicators.

  • Monitor changes over time. Don't just take one random look. Continue to monitor performance each time you make an improvement, and track changes with groups of users. This is known as batch and cohort analysis. The footnote leads to a blog post by Ash Maurya about using actionable metrics in a lean startup.

  • Don't make things up. There is no such thing as random data--there is only data you have not interpreted yet. Get to the bottom of any unusual behavior, and don't go for the obvious, unfounded answer just because it's easier to accept. You'd be succumbing to the false-consensus effect. The footnote leads to a blog post entitled "Why We All Stink as Intuitive Psychologists: The False Consensus Effect."
Chen uses this ingredient in "A Recipe for Lean Startup in Large Organizations," in which he describes techniques used by his collective, HaaYaa. He says in the recipe instruction associated with the Check the Data ingredient:

Identify data across key experiences that indicate success and failure. Analyze the data; gain enough insight and validated learning to make improvements. To keep track of the results of these improvements, I set up customer-experience teams to work with data-analytics teams on monitoring usage stats to see each improvement's effect on groups of cohorts. I also recommend running multivariate tests on variations of a design concept to see which one is most successful.
 Chen obviously knows a lot. This book organizes a great deal of information into a concise presentation. If you like material explained in a leisurely way, this is not the best book for you. But this is a great example of a minimalist style. It might take a while to dig out everything Chen hints at, and along the way you may take some interesting side trips.

Wednesday, May 1, 2013

Unconscious Meaning

This article appears in slightly different form in the May/Jun 2013 issue of IEEE Micro © 2013 IEEE.

This time I look at a short work that contains a large number of surprising ideas.

A User's Guide to Thought and Meaning by Ray Jackendoff (Oxford, New York, 2012, 274pp ISBN 978-0-19-969320-7,, $29.95)

According to Steven Pinker's blurb on the dust jacket, 
"Ray Jackendoff is a monumental scholar in linguistics who, more than any other scholar alive today, has shown how language can serve as a window into human nature. Combining theoretical depth with a love of revealing detail, Jackendoff illuminates human reason and consciousness in startling and insightful ways." 
Jackendoff is the author of many books on linguistics and cognition, but in this one he presents an overview of some key concepts to a broad audience. He says that as a traditional scholarly treatise it would be a thousand pages long -- if he ever finished it. The usual downside of presenting lots of ideas in a short space is that a book can become, like McLuhan's Understanding Media, too dense for ordinary humans to grasp. But down-to-earth examples, simple diagrams, and a few cartoons -- which provide the revealing detail Pinker refers to -- make this book a pleasure to read.

Cognitive Perspective

The heart of Jackendoff's argument is that thought and meaning are almost completely unconscious; we are aware of pronunciations, sentences, visual surfaces, and a small set of inklings that arise from unconscious processes. The inklings, called character tags, give us the feeling, for example, that a certain sound or visual surface is meaningful, significant, good, taboo, based on sensory input, and so forth. If you say "thit," I'm aware that you said something meaningless, but the mental processes that produce that awareness are as unavailable as those that tell me when to breathe. It's hard to explain what happens in your head when I say, "Osculating means doing this."

Jackendoff presents dozens of small examples that refute many widely held ideas and lead him to conclude that meaning is unconscious. Reading them is enlightening and delightful. They often contrast the ordinary perspective -- the one we're all born with -- which is natural, but can lead to paradoxes (there's no such thing as sunsets), with the cognitive perspective, which always asks, "How does the brain do that?" For example, "John slept until the bell rang" entails a single sleep, while "John jumped until the bell rang" entails potentially many jumps. Nothing in the actual sentences conveys that difference. We can use words after the fact to explain the difference, but it is immediately apparent without that step. Whether the jumping is a one-time event or a repeated activity is an aspect of the unconscious meanings that the sentences don't express.

This book is about software in the sense that Jackendoff is concerned with how the brain provides the experiences of language, thought, and meaning that we are all familiar with. From the ordinary perspective we have no trouble understanding that "the bear chased the lion" and "the lion was chased by the bear" mean the same thing. From the cognitive perspective, we know that this understanding arises from brain activity.  But just as we don't look at digital signals to figure out how a computer executes an algorithm written in Java, Jackendoff doesn't try to explain this phenomenon by looking at neural and chemical activity. He focuses on data structures, information flow, and the states of character tags.

Many people have tried to explain consciousness (for example, see Micro Review, Mar/Apr 1992, where I review Daniel Dennett's Consciousness Explained). Jackendoff reviews some of the more popular theories and challenges them to explain the observed phenomena. He believes that whatever consciousness is, it enables us to perform certain language-based kinds of reasoning, but it gives us limited access to the most important and powerful brain activities that support the way we attach meaning to events in the world. Sentences like "the bear chased the lion" and "the lion was chased by the bear" are different handles for closely linked entries in unconscious data structures. Those structures contain information about lions, bears, chasing, and English grammar. They provide the means by which we can recognize "the wargon chased the olifump" as likely to be an instance of the same sort of activity, even though we don't recognize the words "wargon" or "olifump." The structures also define the conceptual relationships that show types and characteristics: bears are animals, mammals, predators, intelligent creatures, four-legged, and so forth. They exhibit mother/child relationships similar to but different from those of lions or humans. If we see a bear, we know it's a bear even if we have never seen that particular bear before.

Jackendoff, unlike his mentor Noam Chomsky, thinks communication is why language developed, with rational thinking as a side benefit. Rational thinking is important, but it isn't what most of us think it is. As Lewis Carroll pointed out in What the Tortoise Said to Achilles, every syllogism relies on a hidden syllogism in an infinite regress. When we say, "All humans are mortal; Socrates is human; therefore, Socrates is mortal," we have a hidden syllogism that lets us reason that this is of the form "All A are B; C is an A; therefore C is B," and that if we can line up humans, mortal, and Socrates with A, B, and C, then the statement about Socrates is true.  Ultimately, we rely on an unconsciously generated character tag to tell us whether the reasoning is correct.

In Thinking Fast and Slow (Farrar, Straus and Giroux, 2011), Daniel Kahneman popularizes System 1, the fast, intuitive mode of thought and System 2, the slow, rational mode of thought. Jackendoff says these correspond to his ideas of unconscious and conscious thought and that they are not separate. System 2 rides on top of System 1 and uses its capabilities. If I encounter a bear with its cubs in the woods, System 1 tells me to head immediately in another direction, while System 2 helps me reason about what the bear might do. System 2 is thought that is linked to a cognitive correlate of consciousness, namely, a data structure that corresponds to our subjective experience of hearing language in our heads and provides a handle for the unconscious thought.

Images, smell, taste, touch, and even the sense of where we are in the world (proprioception)  provide additional handles to unconscious meanings and structures. Blind children, led along two walls of a room to an opposite corner, have no trouble returning along the diagonal to their starting point. This shows that we have unconscious spatial maps that are distinct from visual images.

Jackendoff speculates on the structures that support unconscious thinking. In addition to meanings linked into conceptual structures and spatial maps, every entity that we deal with in the long or short term has a reference file, which holds everything we know about it. The reference file for the mama bear lets us keep her in mind while we ponder other facts that may be helpful. Rational (conscious) thinking enables us to create reference files for thoughts, so we can manipulate and explore them without losing track of them.

Character tags -- of which Jackendoff postulates fewer than a dozen -- play an important role in his model. They explain how we perceive the world as "out there" and how we distinguish actual perception from mental images or dreams. He postulates a character tag that gives us a sense of whether the visual surface in our head arises from our minds or from sensory input. This character tag is often ineffective during dreaming, and it sometimes gives schizophrenics the wrong message. Another character tag provides a sense of whether we are in control of ongoing events. This forms the basis of our sense of free will.


Jackendoff's model of unconscious structures and character tags does more than simply explain the relationships between language, thought, and meaning. It provides a coherent explanation of how we understand and experience the world. For example, from the ordinary perspective, we ask, "What is truth?" Jackendoff answers by showing that question to be hard to answer. He shows pictures of men of varying degrees of baldness, looks at the one in the middle and asks, "Is Ed bald?" He prefers the cognitive perspective, which asks, "What does the brain do when it judges a statement to be true?"  This leads to the conclusion that judging truth largely happens unconsciously, and that the well known phenomena of confirmation bias and denial play a part.

The fact that meaning is unconscious and that System 1 thinking is fast and pretty accurate does not mean that rational thinking is useless. In fact, rational thinking is a large part of what distinguishes humans from other animals. Language, which characterizes System 2 thinking, enables us to give thoughts reference files of their own, describe and manipulate the information provided by character tags, and engage in hypothetical reasoning. We can transform the results of intuitive reasoning into explicit steps and challenge each one.  The famous example is the bat and ball that cost a total of $1.10, with the bat costing a dollar more than the ball. Intuitive reasoning immediately says that the ball costs ten cents, but rational thinking, plodding along methodically, arrives at the correct answer.

All of this has implications for teaching, learning, or becoming a virtuoso of art or science. Jackendoff, an accomplished concert clarinetist, illustrates this by telling how his chamber group spent 15 minutes deciding how to play the first six seconds of the Brahms Clarinet Quintet. From this he extrapolates to how to integrate rational and intuitive thinking in music, theater, sports, art, writing, and every other skilled activity -- even reasoning itself. The goal is always "flow" -- that state in which it all goes so well we surprise ourselves.

Jackendoff tries to show that the arts matter as much as science. I believe that he is correct but it's a hard argument to make. Rational thinking, because it is based on language, conceals aspects of thought that language cannot express. Science resonates with the rational while the arts resonate with the intuitive. Science looks for ever broader generalizations that minimize the surface of appearances. Artistic appreciation seeks intricate and subtle details and patterns. It reveals the character of the surface -- not what is said, but how. Because System 2 rides on System 1, you can't have rationality without the underlying intuition, and the better you train the intuition, the better the rationality will be. As I said, it's a hard argument to make.

The book is a deceptively easy read. I got through it once and realized that I had missed many key points. I had to start over and take careful notes. The book is densely packed with insights and ideas, which are well worth the effort of grasping them.

Monday, April 1, 2013

Ethics of Big Data

This article appears in slightly different form in the Mar/Apr 2013 issue of IEEE Micro © 2013 IEEE.

This time I look at a book that anyone who has ever done anything online should read.

Ethics of Big Data by Kord Davis with Doug Patterson (O'Reilly, Sebastopol CA, 2012, 82pp, ISBN 978-1-4493-1179-7, $19.99, )

Kord Davis played with technology from an early age, and he loved learning its underlying principles and mechanisms. He chose to study philosophy because it gave him the best of both worlds: rigorous analysis and uncovering the way things work. Now, with a degree in philosophy from Reed College, he advises high-tech companies about how to align their business practices with their values and principles. Doug Patterson teaches business ethics. His discussions with Davis led to many of the ideas in this book.

For many years society has struggled with the implications of gathering and analyzing personal data. The vastly increased speed and quantity of such activities have created a qualitatively new situation, generally known as big data. Some of the hardest questions businesses face today arise out of big data. For example, with enough information about your environment, someone can know a lot about you without knowing your name or Social Security number. And by combining such troves of information from disparate sources, an organization can build an intrusive dossier about you without actually violating the well-intentioned privacy policies of the organizations that originally collected the data.

Davis defines the forcing function of big data as the push, whether we like it or not, 
"to consider serious ethical issues including whether certain uses of big data violate fundamental civil, social, political, and legal rights." 
Companies were once thrilled at the prospect of knowing which color car is purchased most often in Texas in the summertime. Now they can know how much toothpaste your family bought from them last year.  Does it matter if Target knows you are pregnant before your husband does?  What if your boss does?  Considering such issues falls outside the usual business discussions, even at the strategic level. Davis presents a framework for such consideration. Because each business situation is different, he includes both abstract and specific elements.

Davis begins with the basic concepts:  identity, privacy, ownership, and reputation. These concepts are central to many ethical discussions, so he wants to establish their definitions and scopes. Identity concerns the relationship between our online and offline lives. Privacy issues boil down to who should control access to data. Ownership concerns not just who owns collected data but which rights can be transferred and what obligations collecting or receiving such data entails. Reputation is about what you can trust. Big data provides both sources of error and checks and balances. 

In addition to these abstract elements, each business has its own values and principles. Big data is ethically neutral, that is, the ethical questions arise from aligning organizational and societal values with organizational actions. Davis presents a system for carrying out this alignment. He believes that by gaining competence in this area, organizations can help to shape public opinion, not just react to it.

Many people are concerned with their right to privacy. Davis feels that using the term "privacy right," because of its connotation of absoluteness, can prejudge some of the ethical issues that arise as organizations try to align their policies with their values. He prefers the term "privacy interest," which covers the gamut from no interest to absolute rights. That is, a right is the strongest kind of interest.

Similarly, Davis avoids the term "personally identifying information," because the technological limitations that define that term are constantly changing. What was once not considered personally identifying can, with new technology, be easily associated with a person. Davis uses the term "personal data" to refer to any data generated in the course of a person's activities. Digital transactions in business or social areas capture related information distinct from the data of the transaction, and by this definition, all of it is personal data.

Davis contrasts the digital and non-digital situations. If you show someone a picture at a party, the event leaves little residue. Post the same picture on your Facebook page, and the permanent record includes plenty of ancillary personal information. Deciding which personal information is ancillary and which is not has an ethical impact. Davis advocates explicitly and transparently evaluating that impact. Doing so starts with articulating organizational values. Values can change over time, and trying to align values with actions can lead to reconsidering those values. Thus, Davis defines an ethical decision point as a  cycle, iterated indefinitely, consisting of the following activities: inquiry, analysis, articulation, and action.

Inquiry means discovering and discussing the organization's core values. Sometimes organizations have one set of values in their founding documents or public relations literature but show by their actions that they have a different set of unstated values. Inquiry aims at discovering the organization's actual values. An example of a value is "We value transparency in our use of big data."

Analysis means reviewing the organization's actual or proposed data-handling practices and determining how well they align with the identified values. For example, analysis might ask whether deducing that a woman is pregnant and placing ads for nursery furniture in front of her aligns well with a stated value of telling customers how the organization uses their personal information.

Articulation means writing the results of analysis, that is, stating explicitly where values and actions align and where they don't.

Action means producing and executing plans to close alignment gaps and to prevent new ones from developing. For example, you might decide to place a button labeled "Why am I seeing this ad?" next to the ad for nursery furniture.

Deciding which organizational activities require this ethical decision point process is tricky. One method, which Davis recognizes as not entirely satisfactory, is to look for the "creepy" factor. Like the Supreme Court's definition of pornography, this determination involves the phrase "I know it when I see it."  For example, it strikes many people as creepy when a retail firm determines whether a woman is pregnant based on her online behavior, so the organization should use the ethical decision point process to evaluate any actions it takes in this area. A health maintenance organization engaging in similar behavior might seem less creepy, so the ethical decision point process might be less important in that case.

Davis talks about ethical decision points in the context of big data, but the methodology is applicable to all sorts of situations. We all have values, and we all do something like ethical decision point analysis in our everyday lives. The speed and scale of big data technology make it essential for organizations develop the ability to carry out the process explicitly and transparently as a core competency. This reduces the risks of unintended consequences and provides a starting point for a clear and immediate response when things go wrong.

To determine the current practices of large organizations, Davis visited the websites of the top 50 of the Fortune 500 companies in the Fall of 2011. Few of these organizations understand and articulate sets of values that customers can use to interpret the organizations' imprecise policy statements. Most focus on "privacy" with little mention of identity, ownership, and reputation. Most are concerned with "personally identifying information" but fail to define that term. Most say that they don't sell personal data, but none claim that they don't buy such data. There is much more information in Davis's summary of his findings. Things may have changed for the better since this survey, but what he found was not encouraging.

Davis sees hope in the fact that the organizations he studied have many strong capabilities in place: leadership and management, strategic planning, a product development process, communication, education and training, and a process for initiating change. But he also notes that not everyone in an organization has the same values and that different roles within an organization have conflicting interests in transparency and alignment. And in the face of tactical pressures, it's tempting to kick the can down the road by avoiding ethical discussions entirely. Davis hopes to overcome these difficulties by taking a cookbook approach. The final third of this short book lays out his alignment methodology framework in more detail, complete with forms and a case study. The forms help you define value personas, an analog of customer personas, which most executives and marketing personnel understand.

This focuses on a critical area for every company that deals with big data and every person who engages in online transactions. As Davis points out, if you ask people what they want, they will tell you. By insisting on definitions of terms and explicit statements of values and how actions align with those values, he creates a framework in which people can discuss difficult issues without unnecessary confusion or rancor. I highly recommend this book to everyone.