More on this book
Community
Kindle Notes & Highlights
Read between
June 23 - August 10, 2017
A complete list of them would both bore and depress you; there are 99 chapters in Rolf Dobelli’s book on the subject, The Art of Thinking Clearly, and 175 entries (at last count) in Wikipedia’s “list of cognitive biases.” Buster Benson, a product manager at the software company Slack, came up with what we think is a great way to group these biases and keep in mind the problems they pose for us:# 1. Information overload sucks, so we aggressively filter. . . . [But] some of the information we filter out is actually useful and important.
The judging, powered by System 1, happens almost instantaneously. It’s then justified in rational and plausible language supplied by System 2.†† This subterfuge often fools not only other minds, but also even the one that came up with it. In fact, we are often “telling more than we can know,” as the psychologists Richard Nesbitt and Timothy DeCamp Wilson put it. The behaviors we label rationalization and self-justification, then, are not always exercises in excuse making. They’re also something much more fundamental: they’re System 1 at work.
In 2006, Avinash Kaushik and Ronny Kohavi, two data analysis professionals who were then working at Intuit and Microsoft, respectively, came up with the acronym HiPPO to summarize the dominant decision-making style at most companies.
Toward a New Mind-Machine Partnership How can we make use of all this knowledge about biases and glitches in System 1 and System 2?
Advertising agencies have long helped their clients not only with the creative work of coming up with new television commercials, but also with the task of figuring out exactly when and where to show them: identifying which TV shows, geographic markets, and times were the best match for the advertisers’ goals and budget. Data and technology have long been used for this work—the agency at the center of the hit TV drama Mad Men gets its first computer, an IBM System/360, in 1969 to help it better place commercials (and impress clients)—but it has remained driven largely by the judgments and
...more
Algorithms Behaving Badly A real risk of turning over decisions to machines is that bias in algorithmic systems can perpetuate or even amplify some of the pernicious biases that exist in our society. For instance, Latanya Sweeney, a widely cited professor at Harvard, had a disturbing experience when she entered her own name into the Google search engine. Alongside the results appeared an ad that read,
In an article in Nature, Kate Crawford and Ryan Calo noted the danger “that in some current contexts, the downsides of AI systems disproportionately affect groups that are already disadvantaged by factors such as race, gender and socio-economic background” and highlighted the importance of considering the social impacts of these systems, both intended and unintended.
This feedback is critically important because it’s how System 1 learns and improves. As Kahneman and psychologist Gary Klein write, “You should never trust your gut. You need to take your gut feeling as an important data point, but then you have to consciously and deliberately evaluate it, to see if it makes sense in this context.” The best way to improve the accuracy and decrease the biases of System 1 is to show it lots of examples and give it frequent and rapid feedback about its accuracy.
But the evidence on this subject is so clear as to be overwhelming: data-driven, System 2 decisions are better than those that arise out of our brains’ blend of System 1 and System 2 in the majority of cases where both options exist.
Beginning in 1984, the political scientist Philip Tetlock and his colleagues undertook a decades-long project to assess the accuracy of predictions in many areas, such as politics, economics, and international affairs. Here again, the conclusions are both clear and striking. In a test involving more than 82,000 forecasts, Tetlock found that “humanity barely bests [a] chimp” throwing darts at the possible outcomes.
In many other cases, forecasts are implicit within a proposed plan of action. A website redesign, for example, contains the implicit prediction that visitors will like it better, as does the redesign of a bank’s branch offices.
Predict Less, Experiment More The existence of superforecasters aside, our most fundamental advice about predictions is to rely on them less. Our world is increasingly complex, often chaotic, and always fast-flowing. This makes forecasting something between tremendously difficult and actually impossible,
These organizations follow the computer scientist Alan Kay’s great advice that the best way to predict the future is to invent it. They
Through rigorous A/B testing—a common online experimentation protocol in which half the visitors see option A when they visit a site while the other half see option B—lingerie company Adore Me found that having models pose with a hand in their hair instead of on their hip could double sales for some items. Instead of spending weeks, days, or even hours having experts analyze and debate a proposed change, it’s usually faster and more accurate to simply test the options online. Often the results will be surprising.
As Mark Forsyth writes in his book The Elements of Eloquence, “adjectives in English absolutely have to be in this order: opinion-size-age-shape-color-origin-material-purpose Noun. So you can have a lovely little old rectangular green French silver whittling knife. But if you mess with that word order in the slightest you’ll sound like a maniac. It’s an odd thing that every English speaker uses that list, but almost none of us could write it out.”
right. Attempts to codify all relevant rules for complex things like languages or furniture into computer systems, and to get the systems to do anything useful, have been largely unsuccessful. As the computer scientist Ernest Davis and neuroscientist Gary Marcus write, “As of 2014, few commercial systems make any significant use of automated commonsense reasoning . . . nobody has yet come close to producing a satisfactory commonsense reasoner.”
the popularity of modern video games has also been a great boost to machine learning. The specialized graphics processing units (GPUs) that drive popular gaming consoles turn out to be extremely well suited to the kinds of calculations required for neural networks, so they’ve been drafted in large numbers for this task. AI researcher Andrew Ng told us that “the teams at the leading edge do crazy complicated things in the GPUs that I could never imagine two or three years ago.”
Efforts like his lead us to agree with Google’s Kaz Sato, who says, “It’s not hyperbole to say that use cases for machine learning and deep learning are only limited by our imaginations.”
This might mean future self-checkout machines and processes that look very different, but we predict that large-scale virtualization will arrive, despite unimpressive progress so far. When it does, it might look like Amazon Go, an 1,800-square-foot convenience store unveiled in Seattle by the online giant in December of 2016. It’s a retailer with neither cashiers nor self-checkout kiosks. Instead, in-store sensors and cameras combine with machine learning technologies and a smartphone app to keep track of everything customers put in their shopping baskets, then bill them for whatever they
...more
present, we see companies explicitly appealing to one side or the other of this self-selection. The fast-food chain McDonald’s is, like Eatsa, increasing virtualization. By November 2016 it had installed digital self-service ordering and payment stations in 500 locations across New York, Florida, and southern California and announced plans to expand the touchscreen technology to all 14,000 of its American restaurants. The Discover credit card, in contrast, is stressing the human touch. A series of ads, first aired in 2013, featured phone conversations between customers and employees played by
...more
Mcdonalds
Our conversations and investigations point to recent major developments in five parallel, interdependent, and overlapping areas: data, algorithms, networks, the cloud, and exponentially improving hardware. We remember them by using the acronym “DANCE.”
Together, the elements of DANCE are causing the Cambrian Explosion in robots, drones, autonomous cars and trucks, and many other machines that are deeply digital.
The most profound benefit of 3D printing is probably that it makes experimentation and customization inexpensive. The path from idea or need to finished, useful part no longer has to include the time-consuming and expensive steps like mold making and other conventional manufacturing practices.
Bass, the former CEO of design and engineering software company Autodesk, sees 3D printing as only one part of a much larger story. As he told us, “I think additive manufacturing is a subset of what has really transformed manufacturing, which is the use of low-cost microprocessors to precisely control machinery.”
a more useful frame for thinking about 3d printing?
There are three rules for writing a novel. Unfortunately, no one knows what they are. — attributed to Somerset Maugham (1874–1965)
Figure 1 A heat exchanger designed using generative design software. (© Autodesk)
Natural Artificial Designs The heat exchanger shown in Figure 1 is an example of “generative design,” a process in which software is used not to help a human designer create drawings, perform calculations, and explore trade-offs, but instead to do all that work itself, 100% automatically, and to come up with one or more complete designs
This part was manufactured by 3D printing. In fact, it would have been impossible to make using traditional manufacturing processes. But now that 3D printing is a reality, generative-design software is no longer constrained by older production methods and is free to imagine and propose a vastly wider range of shapes. And unlike most, if not all, human designers, the software isn’t consciously or subconsciously biased toward...
This highlight has been truncated due to consecutive passage length restrictions.
Figure 2 Race car chassis model. (© Autodesk)
stunning!
We suggest a slight tweak for health care: the medical office of the future might employ an artificial intelligence, a person, and a dog. The AI’s job will be to diagnose the patient, the person’s job will be to understand and communicate the diagnosis, and to coach the patient through treatment, and the dog’s job will be to bite the person if the person tries to second-guess the artificial intelligence.
Economic progress, in capitalist society, means turmoil. — Joseph Schumpeter, 1942
WITHIN ONE GENERATION, SEVERAL LONG-STANDING INDUSTRIES were transformed permanently and deeply by a single computer network. The business world has rarely, if ever, seen disruption at this speed and scale before. The first sentence in the previous paragraph exaggerates—the Internet had some help from other technologies as it remade sector after sector—but we don’t think the second sentence does. As we described in Chapter 1, there have been technology-driven industrial revolutions before, based around advances like the steam engine and electrification, but they took longer to unfold and
...more
Perhaps the clearest sign of deep shifts in the industry was Playboy’s announcement in October of 2015 that after sixty-two years, it would no longer feature nude photos. Founder Hugh Hefner, who in 2006 was named by the Atlantic as one of the most influential living Americans in large part because of photos of unclothed women, agreed with the move. One of the reasons for this change was that, like other publications, Playboy depended increasingly on traffic from social media, but sites like Facebook and Instagram did not allow nudity.
other examples of business disruption, memorably defined by Thomas Friedman in his book Thank You for Being Late as “what happens when someone does something clever that makes you or your company look obsolete.” Digital technologies are perhaps the most powerful tools ever wielded by clever disruptors.
Platform economics, Moore’s law, and combinatorial innovation continue to produce developments that take industries and their incumbents by surprise.
This is a common feature of digital platforms: they can unbundle resources that used to be tightly clustered together, and therefore difficult to consume one by one.
It’s a pattern we’ll see more often in the future, we predict. We agree with the business scholars Geoffrey Parker, Marshall Van Alstyne, and Sangeet Choudary, who write in their book Platform Revolution that “as a result of the rise of the platform, almost all the traditional business management practices . . . are in a state of upheaval. We are in a disequilibrium time that affects every company and business leader.”
What are the most important digital platforms in your industry today? What do you think they will be in three years?
2005 Google bought Android, a startup about which little was known, for approximately $50 million. The technology blog Engadget commented at the time that “we only have the faintest idea why Google just bought Android, a stealthy startup that specializes in making ‘software for mobile phones.’ ” Within a few years, though, the value of a robust alternative to Apple’s platform for apps became quite clear.
User experience is a broader concept, encompassing how effective and pleasant a product is to use. The difference between the two was wittily summarized by the designer Ed Lea with two pictures: a spoon representing the user interface, and a bowl of cereal representing the user experience.
Math, Data, and Apps: Revenue Management to the Rescue “Revenue management” is the name given to the set of algorithms and technologies developed over the years to help service businesses deal with finite capacity and perishable inventory, and turn them to their best possible advantage. The fundamental goal of revenue management is to let service companies sell as much as possible of their finite and perishable inventory to those customers with the highest willingness to pay, then sell off the rest to customers further down the demand curve. Revenue management began with airlines,† spread to
...more
Our favorite label for such platforms, which we first heard from artificial intelligence rock star Andrew Ng, is “O2O,” which means “online to offline.” We like this shorthand because it captures the heart of the phenomenon: the spread from the online world to the offline world of network effects, bundles of complements, and at least some of the economics of free, perfect, and instant.
Elance and oDesk were two of the pioneering online resources for connecting freelancers and clients. In 2015 the companies combined to form a new company, called Upwork. Businesses use the site to post projects that independent freelancers or agencies then bid on. The tasks undertaken can range from web design and copywriting to accountancy and data entry. The Upwork platform matches experts from anywhere in the world—why not, if the deliverables are digital?—with the jobs they are best placed into and provides tools for, among other things, project management and payment. By 2016, Upwork
...more
Elance and oDesk were two of the pioneering online resources for connecting freelancers and clients. In 2015 the companies combined to form a new company, called Upwork. Businesses use the site to post projects that independent freelancers or agencies then bid on. The tasks undertaken can range from web design and copywriting to accountancy and data entry. The Upwork platform matches experts from anywhere in the world—why not, if the deliverables are digital?—with the jobs they are best placed into and provides tools for, among other things, project management and payment. By 2016, Upwork was
...more
final reason we’re professionally fascinated by O2O platforms is that they really weren’t possible even a decade ago. Many of the businesses described in this chapter rely on powerful mobile computing devices, and as we’ve seen the smartphone era only started in 2007 with the iPhone (and apps from outside developers took another year to arrive). Smartphones were not only the first truly mobile computers; they were also the first location-aware ones, thanks to their GPS sensors. These are indispensable complements to almost every successful O2O system. Cloud computing was also critical
Platforms can capture much, most, or even all of the value as they spread throughout an industry.
As the mathematician and author John Allen Paulos observed in the early days of the web, “The Internet is the world’s largest library. It’s just that all the books are on the floor.”
Hayek wrote, The marvel [of prices] is that in a case like that of a scarcity of one raw material, without an order being issued, without more than perhaps a handful of people knowing the cause, tens of thousands of people whose identity could not be ascertained by months of investigation, are made to use the material or its products more sparingly; i.e., they move in the right direction. . . . I am convinced that if [the price system] were the result of deliberate human design, and if the people guided by the price changes understood that their decisions have significance far beyond their
...more
Groups often behave in ways that are emergent and thus generate knowledge. As groups went online and became the crowd, innovators found different ways to detect and harvest this knowledge. Prediction markets were one of the earliest of these, and the ones that built most directly from Hayek’s insights. These are markets not for goods and services, but for future events, such as a particular person being elected US president in 2020, an upcoming movie making between $50 million and $100 million in the box office in its first week, or the official US inflation rate averaging more than 3% over
...more
Writing a New Playbook Studying Linux’s history reveals several principles that appear to be important, perhaps even essential, for bringing the crowd together to accomplish something significant. These include openness, noncredentialism, verifiable and reversible contributions, clear outcomes, self-organization, and geeky leadership.
Collaboration / working in networks. These and Wikipedia's rules are two of the best examples.
