Big Data, Plainly Spoken (aka Numbers Rule Your World)

Artificial human intelligence

When we use the words artificial intelligence, we typically mean artificial machine intelligence, training machines to act like human beings. What is actually happening is the opposite - we are developing artificial human intelligence, as in, humans are being trained to think like machines.

Example 1: I recently called a cab company and told them I was at Union Square. The despatcher was taking a long time to respond, and eventually asked me whether I am in the east or west side of Manhattan. MIffed, I wondered how it is possible to not know where Union Square is. The guy explained that the system required him to enter an exact address.

When we didn't have the intelligent software, the cab drivers responded to Union Square, or an exact address, or a set of cross streets, or whichever way the passengers choose to describe their locations. Now, we must figure out what the software knows, and then speak to it in its language. After many years of such training, I doubt that these despatchers can locate Grand Central on a map!

Example 2: You call the customer service line of your bank, and get to talk to the intelligent service agent (a bot). She asks you to explain your need. You speak to her as if speaking to a human. She doesn't understand you, and wants you to try again. After a few fruitless exchanges, you learn that the only way to get through to her is to figure out her vocabulary. She only understands maybe several dozen words. So you map your real need to the "nearest neighbor" in the space of the bot's vocabulary. Next time, you call the same bank, you start using the bot's vocabulary.

Example 3: I have observed how some students learn coding. Their lifeline is Google which leads them to StackExchange or similar Q-and-A websites. They find the code they want, and copy and paste it to their screens. They press the Run button. It fails, with an error message. Suspecting that the error may have to do with the placement of a parenthesis, they move the parenthesis to a different spot. They press Run again. It fails again. They move the parenthesis to a third spot, and press Run. It fails again. Eventually, it is in the right slot and the program runs. This brute-force, try-all-possibilities paradigm is frequently used by machines because they can run calculations really fast. Now, humans are learning in the same way.

***
Alarm has been raised over the possibility that machines will eventually control humans. This is possible if humans adopt artificial human intelligence, that is to say, learn to think like machines.

 

 

Book review: Everybody Lies by Seth Stephens-Davidowitz

Seth Stephens-Davidowitz has written a fascinating book calling for social scientists to use data collected by Google or Facebook in their research. This is a controversial issue, and if it weren’t so, it wouldn’t warrant writing a full-length book about it. Google does not release publicly its search data, but provides some pre-processed and aggregated statistics through services such as Google Trends and Adwords. Researchers who use this data do not have control over its collection or processing. (Seth previously worked at Google, and has written some columns for New York Times.) Facebook does not publish its data either, although social media users have a lower expectation of privacy.

Such big datasets come with a set of knotty problems, which I have previously summarized as OCCAM. In addition to having little control over its origin, the researcher’s purpose typically diverges from that of the data collector. The data is found or observed and not usually experimental. It is often treated as “complete” or essentially complete by the researcher, which is an assumption, not a fact. In stages of merging in other data, the researcher introduces inevitable errors. Here is my previous post about OCCAM datasets (link). It is not surprising that classically trained scientists have reservations about such datasets, especially if they are interested in causal mechanisms. Nevertheless, I agree with Seth that we can make progress on solving these problems if we start taking them seriously.

In writing the book, Seth carried out a number of mini-studies using mostly Google Trends data. Here are 8 things I learned from reading Everybody Lies:

1. Some people use search engines as confessionals. They type complete sentences like “I am sad.” or open-ended questions like “Is my daughter ugly?”
2. People assume machines (like the Google search engine) will keep their secrets. For sensitive topics, Google may generate more honest data than surveys. There are many questions asked to Google that I’m sure people won’t pose to a librarian.
3. Google searches for “Obama” is frequently paired with “kkk” and the “n” word. The prevalence of racist searches does not exhibit a North-South divide – it’s East-West.
4. As President of Harvard, Larry Summers spent quite a bit of time brainstorming with Economics PhD students on how to beat the stock market using new data. (And they came up empty-handed, or so they say.)
5. Anthony Weiner got rejected from Stuyvesant High School (famous NYC public school), missing the cutoff score in the admissions test by one point
6. Some economists found that going to Stuyvesant conferred no meaningful benefit to one’s career – at least, this is the case for those who attain a score close to the cutoff in the admissions test.
7. There are 6,000 searches on Google a year for “how to kill your girlfriend” while there are 400 murders of girlfriends.
8. “Big data” does not provide any insights that surveys can’t at the aggregate level so people slice and dice the data to examine “micro” segments, which means they are analyzing a huge collection of small data sets

The research mentioned in Seth’s book come out of the Economics discipline primarily, and can be considered in the tradition of Freakonomics (2005). There are examples of natural experiments, as popularized by Steven Leavitt. Seth brings the coverage up to the current trends, describing regression discontinuity, field experiments, and other techniques favored by econometricians at the moment. For those interested in what happens after Steven Leavitt, this is a good place to start.

The Times agrees on privacy and kind of on fake news business

The New York Times Magazine has been publishing some pieces that directly relate to a couple of my blog posts.

In this article, Amanda Hess noticed that "privacy became a commodity for the rich and powerful." This echoes my blog post on "Data is the next frontier of equal rights." Hess discussed the asymmetry and hypocrisy of the situation whereby the same businesses and business executives that are wantonly stripping us of our privacy are super-protective of their own "private" information. In my blog post, I pointed out that the selling out of customers' private information started long before the Internet came of age.

In another article, basically a profile of Facebook CEO Mark Zuckerberg, Farhad Manjoo tackled the problem of social media and fake news. I encourage you to read this article alongside my post on "This Fake News Business." The article contains insights into Facebook's reaction to many of the points I raised in my blog post, plus Manjoo's reaction to Facebook's reaction.

LIke Manjoo, I find that Facebook, Google and many tech companies live inside their own bubbles. They are earnestly justifying their decisions, even when such actions prove unpopular. At one point, Manjoo was told that Facebook thinks its algorithms have solved the "clickbait" problem. It takes a lot of effort to transcend the bubbles. Is it possible that the fake news challenge may do the trick?

 

That fake news business

When Hillary Clinton unexpectedly lost the election in November 2016, it was high time for pundits to espouse their pet theories for the shocking losses suffered by the Democrats at all levels of government. The usual suspects were put on parade, such as “Bernie Bros” who abstained from voting and the “deplorables” who voted against their own interests. For the first time, two unlikely entities faced scrutiny: the social-media giant, Facebook, and the search-engine giant, Google.

Facebook and Google were accused of hacking the election by promoting and propagating so-called fake news. People dawned on the extraordinary power held by these Internet giants to shape political conversations. In response to the firestorm, both companies have taken a series of actions to combat “fake news.” These efforts are mere window-dressing. A real solution to the problem requires tinkering with their core business models. Over the years, Google and Facebook have transcended their original missions to become media organizations, while invoking a quiet revolution that has blurred the line between news and spiels. Today, both corporations are enormous advertising agencies that obtain virtually all of their revenues from advertisers. Their advertising customers flourish under the same set of conditions that allow fake news to prosper.

Critics of Facebook charged that its Trending Topics section makes fake news viral. According to an investigation by Buzzfeed, the following false items were among the most popular on Facebook in 2016:

• “Obama Signs Executive Order Banning the Pledge of Allegiance in Schools Nationwide”
• “Pope Francis Shocks World, Endorses Donald Trump for President, Releases Statement"
• "FBI Agent Suspected in Hillary Email Leaks Found Dead in Apparent Murder-Suicide"

Rolled out in 2014, Trending Topics is a feature designed to expose Facebook users to frequently shared articles that have not appeared on their personal Newsfeeds. The choice of content is supposed to be computed by an algorithm. Google’s search algorithm has also been caught red-handed in elevating dubious content. In its Featured Snippets section, found on top of many search results since 2014, it offers what the algorithm considers the best answer to a user’s question. According to a report by The Outline, the following made-up facts were cited and amplified by Google’s search engine:

• Presidents McKinney, Wilson, Harding and Truman were members of the Ku Klux Klan
• Obama was planning to declare martial law
• Proposition 63 is a deceptive ballot initiative that will criminalize millions of law-abiding Californians.

(Google reacts quickly to such discoveries, replacing the offensive snippets.) The same system also generates answers for Google Assistant, the “intelligent” personal assistant.

The spread of fake news is an unintended consequence of the well-oiled machines that Google and Facebook have created for their advertising customers. To see this, we revisit the humble beginnings of the two Internet giants.

In the late 1990s, Google’s founders penned their famous “Don’t Be Evil” motto. That was the era of “eyeballs,” when technology startups earned gold stars for amassing users while anyone mentioning revenues, let alone profits, was heckled and shamed. Competitors such as GoTo.com which sold search real estate to the highest bidder were definitely in the evil category. When Facebook went public in 2012, it boasted over 800 million users but carried an unproven ads-only business model.

The venture-capital backers eventually made good on their investments in both startups through record-setting IPOs. Shareholders of Google and Facebook then pushed management to abandon the carefree, cash-burning culture. Both tech giants in short order transformed themselves into high-powered profit engines, becoming two of the greatest advertising agencies in the world, each earning $20 billion of ad revenues per year. Advertisers became their best friends, buying the argument that the performance of digital ads is measurable, and thus more efficient, than traditional forms of advertising.

Google’s bread-and-butter product are the paid search ads shown around its search results. Advertisers bid for placement and pay Google on a per-click basis. (Yes, they adopted the GoTo.com business model, if you’re wondering.) The cost-per-click (CPC) schedule is arguably more accountable, as advertisers only pony up if someone interacts with their ads. It wasn’t long before clicks became the preeminent currency in the digital marketing universe. Facebook also makes most of its revenues on a CPC basis.

The business models of Google and Facebook are to deliver clicks to advertisers. The CPC bidding process aligns all parties toward the common goal of maximizing clicking. Statistics show that few Google users ever roll over to the second page of search results. The top section becomes the most valuable real estate in search. The Facebook newsfeed, another coveted online property, is designed just like a page of search results. Advertisers compete for placement by bid prices and by something Google calls the “quality score”. The key determinant of “quality” in Google-verse is the click-through rate. (Facebook uses a similar device called “relevance score.”) Both companies provide tools for advertisers to test different ad formats and designs to evaluate which ones generate more clicks. A 2012 study by Facebook scientists demonstrated how ads get more clicks if the advertiser inserted the name of a friend who Liked the ad. (Bakshy, et. Al., “Social Influence in Social Advertising: Evidence from Field Experiments,” which I discussed in this old post)

Soon, enterprising entrepreneurs went hunting for clicks. All types of “click bait” sprouted. For example, ads are placed next to scroll bars to take advantage of accidental clicking. On the mobile phone, there are “self-clicking” ads that initiate unauthorized app downloads. But the term click bait is most often associated with sensational headlines designed to make us click. In the last decade, several new media outlets owe their success to a posse of headline writers who excel at writing click-worthy copy. Business Insider, Mashable, Gawker, and Buzzfeed come to mind.

Peak click bait arrived with the invention of “content farms.” Demand Media and Associated Content pioneered this hyper-efficient mode of generating clicks. They research the most common search queries, and hire minimum-wage, hourly workers to tailor-make content to answer those queries. Because such content is directly relevant to popular queries, Google rated it high “quality,” and so, these content farms dominated the first pages of search results. Demand Media, for example, owns and runs the eHow website, which specializes in how-to videos. Users who click on such links - thereby inflating their quality score - frequently find the content insufficient and repetitive, since the hired hands have little to no expertise in the subject matter.

The line between information and commercial spiel has continued to blur as Google and Facebook built up their advertising businesses. Take a how-to video: it can fall into either category, being informative, or being spammy. Something about these content farms shook Google management to take decisive action, altering its famous search algorithm to target and downgrade their content. Looking back, we recognize that moment as seminal, when Google, and by implication, other tech giants, exercised its power to arbitrate and filter content seen by hundreds and millions around the world.

The algorithm update dealt a blow to Demand Media and other content farms. Nevertheless, a milder form of such strategies has thrived, now known more benignly as “content marketing.” Advertisers commission copy that is written to read like information or news. Almost all commercial Web properties insert such content around their regular offering. Many reformed, out-of-work journalists have reluctantly become content marketers, and in so doing, they have accelerated the blurring of news and advertiser spiel.

For example, when I visited Yahoo! News just now, I was shown an item titled “People in Heavy Debt May Be In For a Big Surprise.” If I wasn’t looking out for it, I would have missed the disclosure, in a faint gray font, that this item was “sponsored” content. Clicking on the title sent me to a website, with an article written by marketers and a form from which I could request a price quote for their debt relief product. Although this content is commercial, not political, it is “fake news.” 

Advertisers are not bound by the same rules that govern journalists. They are in the business of shaping public opinions in pursuit of profits. Disney is where you can forget about life’s trouble and be happy. Wearing Nike products makes you a better athlete. Brand advertisers create meaning through a mixture of facts, half-truths, and opinions. They desire reach, clicks, and influence. Google and Facebook provide everything these advertisers want, and content marketing is their best product yet.

Through Google and Facebook, advertisers buy access to gigantic networks of consumers, to whom they can preach brand messages any time they want. These super-networks were originally built without the commercial imperative. Users flocked to Google’s search engine because its algorithm found useful information in the World Wide Web while Facebook made it easy for users to share pictures and stories with friends. Eventually, Google and Facebook started inserting third-party advertising messages around search results and user-generated content respectively. Over time, their engineers develop innovative products that facilitate the diffusion of brand messages. For example, the Like buttons enable Facebook to locate users outside the Facebook domain, and deliver them advertisements.

The same assets that propagate viral web pages and videos are delivering commercials to consumers. Advertisers love this new medium: previously, newsprint is limited by space and the once-a-day publication schedule while television and radio are limited by location and bandwidth. The new medium, abetted by the rise of mobile devices, is 24/7 immersion. Content marketing gives advertising copy a camouflage to hide amongst news and information. The crafted headlines are click baits. On Facebook, users freely share and like content. On Google, popular content is deemed important, and given preferential positioning on pages of results. Both networks are extremely efficient at spreading messages.

An unintended consequence of these efficient networks is the spread of fake news. Peddlers of fake news desire the same things as advertisers: reach, clicks and influence. They also write click-bait headlines that have the appearance of real news. They also want to shape public sentiments. They also rely on people passing messages to their friends and peers. They also can use experiments to optimize their designs and formats. They also consider popularity as a proxy for importance.

Google and Facebook cannot solve the fake news problem without taking a hard look at their business models. The strategies that have delivered them financial success beyond imagination are the same ones that enable fake news to prosper. If they are serious about dialing back fake news, they must apply brakes to their advertising juggernaut.

Ramp metering magic

Here is a recent article giving some history on the invention of ramp meters, used on highways to mitigate congestion. I discuss this subject in Numbers Rule Your World (link).

There is an interesting stochastic phenomenon underlying highway congestion. Ramp meters help by regulating the inflow of vehicles onto the highway, and prolonging the period of time by which the highway runs at full capacity. The key insight is that when the highway breaches its capacity, it does not run at maximum throughput - it runs at less-than-maximum throughput.

In the book, I also discuss the fascinating large-scale experiment in Minnesota some years ago that generated nice statistics to support the theory of ramp metering.