Working at Microsoft

I'm joining Microsoft! I'll be part of the excellent Analysis and Experimentation team, helping people learn from data. I'm excited!

I've been geeking out with big data from before data science was a thing and before being a geek ever could be considered a compliment. For two decades, I've enjoyed looking at the paths people take online, where they find success and where they become annoyed, and how changes can help more find success.

Sometimes this is prioritizing things people like and find useful. Sometimes it is changing or eliminating things that, despite the good intentions of the developers and designers, don't work for people. Sometimes it is anonymously sharing things that only some people found with others who haven't found it yet. And sometimes it is having humility about being able to guess what will work and deciding to try many things to discover what actually does work.

If you're at Microsoft, whether an old friend, a team looking to talk about recommendations, personalization, data science, and experimentation, or just looking to chat, please get in touch! I'd love to hear from you.

Quick links

What has caught my attention lately:

• Tog (of the famous Tog on Interface) says Apple has lost its way on design: "Apple is destroying design. Worse, it is revitalizing the old belief that design is only about making things look pretty. No, not so! Design is a way of thinking, of determining people’s true, underlying needs, and then delivering products and services that help them." ([1] [2])
  
  
• Good advice on adding features to a product: "'Great or Dead', as in, if we can't make a feature great, it should be killed off." ([1])
  
  
• Great data on smartphone and tablet ownership. Sometimes it's hard to remember that only five years ago most people didn't have smartphones. ([1])
  
  
• Advice for anyone thinking of doing a startup. Here's the conclusion: "So all you need is a great idea, a great team, a great product, and great execution. So easy! ;)" ([1])
  
  
• Related, a Dilbert comic on the value of a startup idea ([1])
  
  
• "People might think that human-level AI is close because they think AI is more magical than it actually is" ([1])
  
  
• "VCs hate technical risk. They’re comfortable with market risk, but technical risk is really difficult for them to reconcile." ([1])
  
  
• Google finds eliminating bad advertisements increases long-term revenue, concluding: "A focus on user satisfaction could help to reduce the ad load on the internet at large with long-term neutral, or even positive, business impact." ([1] [2])
  
  
• "Crappy ad experiences are behind the uptick in ad-blocking tools" ([1])
  
  
• On filter bubbles, a new study finds algorithms yield more diversity of content than people choosing news themselves ([1] [2] [3])
  
  
• Facebook data center fun: "The inclusion of 480 4 TB drives drove the weight to over 1,100 kg, effectively crushing the rubber wheels." ([1])
  
  
• Great data on who uses which social networks ([1])
  
  
• "One of the great mysteries of the tech industry in recent years has been the seeming disinterest of Google, which is now called Alphabet, in competing with Amazon Web Services for corporate customers." ([1])
  
  
• "Maybe part of AWS value prop is the outsourcing of outages: when half the net is offline, any individual down site doesn't look as bad." ([1])
  
  
• "87% of Android devices are vulnerable to attack by malicious apps ... because manufacturers have not provided regular security updates" ([1])
  
  
• Fun maps showing where tourists take photos compared to locals ([1] [2] [3])
  
  
• Multiple camera lenses, an idea soon coming to mobile phones too? ([1])
  
  
• Another interesting camera technology: "17 different wavelengths ... software analyzes the images and finds ones that are most different from what the naked eye sees, essentially zeroing in on ones that the user is likely to find most revealing" ([1])
  
  
• And another: "Take a short image sequence while slightly moving the camera ... to recover the desired background scene as if the visual obstructions were not there" ([1])
  
  
• Useful to know: "Survey results are mostly unaffected when the non-Web respondents are left out." ([1])
  
  
• Surprising finding, meal worms can thrive just eating styrofoam: "the larvae lived as well as those fed with a normal diet (bran) over a period of 1 month" ([1])
  
  
• Autonomous drone for better-than-GoPro filming? ([1] [2])
  
  
• "We see people turning onto, and then driving on, the wrong side of the road a lot ... Drivers do very silly things when they realize they’re about to miss their turn ... Routinely see people weaving in and out of their lanes; we’ve spotted people reading books, and even one [driver] playing a trumpet." ([1])
  
  
• A fun and cool collection of messed up images out of Apple maps. It's almost art. ([1])
  
  
• SMBC comic, also applies to AI ([1])

Not working at Google

It was a surprise, to me at least, that I wasn't able to find a good fit at Google Seattle.

Google nowadays is different than I expected, and, after four months of trying hard to find any way to make it what I wanted, I resigned.

I'm saddened and disappointed. On the bright side, I did get a chance to work with many remarkable people, which I think made it worthwhile.

Working at Google

I joined Google a few months ago. I've wanted to work at Google for a long time. I first interviewed there back in 2003!

I've written on this blog since 2004, during Findory and beyond, but, like many blogs, posts have slowed in recent years. Unfortunately, I don't expect to be able to post much here in the coming months either.

Thanks for reading all these years. I hope you enjoyed this blog, and I hope to be able to post frequently again at some point in the future.

Quick links

Some of the best of what I've been thinking about lately:

• Amazon now has 109 warehouses and 165k employees. Wow. ([1])
  
  
• Amazon cloud computing has 17% operating margins, surprisingly high given all the competition ([1] [2])
  
  
• Microsoft appears to be claiming they're going to be bigger than Amazon AWS in three years ([1])
  
  
• But Amazon's Andy Jassy says, "One of the biggest surprises around this business has been how long it took the old guard companies to try and pursue an offering. None of us thought we would get a seven-year head start.” ([1])
  
  
• Apple is the iPhone ([1])
  
  
• Great article on the history of YouTube: "It's easy to forget YouTube almost didn't make it" ([1])
  
  
• Mobile ads still aren't targeted (unlike Web ads) ([1] [2])
  
  
• Browsers are disabling Java and Silverlight by default, and Flash's days might be numbered ([1])
  
  
• Surprising how few people use their mobile to get directions, look up public transit, or request a taxi ([1])
  
  
• A major predictor of how much people like a picture of a face is how sharp and clear the eyes are in the photo ([1])
  
  
• Successful tests of a bullet-sized guided missile, cool but very scary ([1] [2])
  
  
• "If an election was hacked any time in the past, we will never know" ([1])
  
  
• "Maybe this head-up display for your life starts as a head-up display for your car" ([1])
  
  
• Beginning of the end for radio: "Norway the first country in the world to 'decide upon an analogue switch-off for all major radio channels'" ([1])
  
  
• A new trend in biology, collecting large amounts of data and doing A/B testing ([1] [2])

Interview on early Amazon personalization and recommendations

Amazon.com in late 1996

Amazon.com in mid-1997

I have a long interview with the Internet History Podcast mostly about Amazon around 1997, especially the personalization, recommendation engine, and data-driven innovations at Amazon, and the motivation behind them.

I think the interview a lot of fun. It gives a view of what Amazon was like way back when it was just a bookstore only in the US, had just one webserver, and we barely could keep the website up with all the growth.

Lots of history of the early days of the web, well before CSS and Javascript, before cookies were even widely supported, and before scale out, experimentation and A/B testing, and large scale log analysis were commonplace.

Give the podcast a listen if you are interested in what the Web looked like back in 1997 and the motivation behind Amazon's personalization and recommendations.

Quick links

What I've been thinking about lately:

• "The chip is so low power that it can be powered off energy capture from the body ... 35 microamps of power per megahertz of processing ... and less than 200 nanoamps ... in deep sleep mode" ([1])
  
  
• "Forgetting may be nearly as important as remembering in humans" ([1])
  
  
• Only 40% of people use maps on their smartphone ([1] [2])
  
  
• OkCupid and Dataclysm: "In the age of Big Data, the empirical has deciphered the intimate" ([1])
  
  
• Cross functional teams might seem slower when you're in them, but, long-term, are more productive ([1])
  
  
• Very good article on mostly evil uses of personalization ([1] [2])
  
  
• "Fake accounts are given a veneer of humanity by copying profile information and photos from elsewhere ... [and] a picture of a beautiful woman" ([1])
  
  
• "Because almost no one patches their BIOSes, almost every BIOS in the wild is affected by at least one vulnerability" ([1])
  
  
• Cracking by forcing non-random memory errors, just about all RAM chips currently used are vulnerable ([1] [2] [3])
  
  
• Computer security "backdoors will always turn around and bite you in the ass. They are never worth it." ([1] [2])
  
  
• "Facts can only do so much. To avoid coming to undesirable conclusions, people can fly from the facts and use other tools in their deep belief protecting toolbox" ([1])
  
  
• Why TV is losing viewers, the ads are annoying: "Decline caused by a migration of viewers from ad-supported platforms to non-ad-supported, or less-ad-supported platforms" ([1])
  
  
• "The same dysfunctional folie a deux playing out between credulous tech media and even more credulous VC investors" ([1])
  
  
• Does the difficulty of building intelligent systems grow exponentially as we make progress? That question has big implications for whether we should expect (or fear) an AI singularity. ([1])
  
  
• Very fun version of Family Feud using Google search suggestions ([1] [2])
  
  
• Do you know what you don't know? Try this confidence calibration quiz. ([1])
  
  
• Love this quote: "I have thrown away a number of successful careers out of boredom" ([1])
  
  
• Humor related to recommendation systems: "An exciting new system that takes all the bother, all the deciding, all the paying—all the shopping—out of shopping." ([1])
  
  
• Two SMBC comics related to AI ([1] [2])

Data Maven from Crunchzilla: A light introduction to statistics

Crunchzilla just launched Data Maven!

Data Maven from Crunchzilla is a light introduction to statistics and data analysis.

For too many teens and adults, if they think about statistics at all, they think it's boring, tedious, or too hard. Too many people have had the experience of trying to learn statistics, only to get bogged down in probability, theory, and math, without feeling that they were able to do anything with it.

Instead, your first exposure to statistics should be fun, interesting, and mostly easy. Data Maven from Crunchzilla is more of a game than a tutorial. To play, you answer questions and solve problems using real data. Statistics is your tool, and data provides your answers. At the end of Data Maven, you'll not only know a bit about statistics, but also maybe even start to think of statistics as fun!

Like programming, statistics and data analysis are tools that make you more powerful. If you know how to use these tools, you can do things and solve problems others cannot. Increasingly, across many fields, people who understand statistics and data analysis can know more, learn more, and discover more.

Data Maven is not a statistics textbook. It is not a statistics class. It is an introduction. Data Maven demystifies statistics. Teens and adults who try Data Maven build their intuition and spark their curiosity for statistics and data.

Please try Data Maven yourself! And please tell others you know who might enjoy it too!

Update: Six years later, Data Maven is mostly shut down, missing from crunchzilla.com if you go to it, but still available if you go directly to the URL for Data Maven. Unfortunately, Data Maven was never popular. And it seems it may have distracted from the coding lessons of Code Monster, Code Maven, and Game Maven. So now it is mostly shut down. I hope someone else will give teaching more kids and adults about statistics. It's hard to make easy and fun, and it's important, especially nowadays.

More quick links

Some of the best of what I've been thinking about lately:

• Great TED talk titled "The mathematics of love", but probably should be titled "A data analysis of love" ([1])
  
  
• Manned submarines are about to become obsolete and be replaced by underwater drones ([1] [2] [3])
  
  
• "No other algorithm scaled up like these nets ... It was a just a question of the amount of data and the amount of computations." ([1] [2])
  
  
• What Google has done is a little like taking a person who's never heard a sound before, not to mention ever hearing language before, and trying to have them learn how to transcribe English speech ([1] [2])
  
  
• Teaching a computer to achieve expert level play of old video games by mimicking some of the purpose of sleep ([1] [2])
  
  
• "Computers are actually better at object recognition than humans now" ([1] [2] [3] [4])
  
  
• The goal of Google Glass was a "remembrance agent" that acts as a second memory and gives helpful information recommendations in real time ([1] [2] [3])
  
  
• A new trend, large VC investments in artificial intelligence ([1])
  
  
• "Possibly the largest bank theft the world has seen" done using malware ([1])
  
  
• "Users will prioritise immediate gain, and tend to dismiss consequences with no immediate visible effect" ([1] [2])
  
  
• "Crowds can't be trusted". It's "really a game of spamfighting". ([1] [2])
  
  
• SMBC comic: "All we have to do is build a trustworthiness rating system for all humans" ([1])
  
  
• Dilbert describes most business books: "He has no idea why he succeeded" ([1])
  
  
• Architect Clippy: "I see you have a poorly structured monolith. Would you like me to convert it into a poorly structured set of microservices?" ([1])
  
  
• Man kicks robot dog. Watching the video, doesn't it make you feel like the man is being cruel? The motion of the robot struggling to regain its balance is so lifelike that it triggers an emotional response. ([1] [2] [3])
  
  
• SMBC comic: "Are we ever going to use math in real life?" ([1])

Quick links

What has caught my attention lately:

• "Ads are often annoying ... [and] the practice of running annoying ads can cost more money than it earns" ([1] [2] [3])
  
  
• Robot plays beer pong, but the real story is the clever bean bag robotic gripper using the "jamming phase transition of granular materials" ([1] [2] [3])
  
  
• Good list of features a modern phone should have but does not ([1])
  
  
• "At this point, Apple is basically an iPhone company with a few other side businesses ... The iPhone accounted for ... a staggering 69 percent ... of Apple's revenue." ([1])
  
  
• "We were not building the phone for the customer — we were building it for Jeff [Bezos]" ([1] [2])
  
  
• "One of the biggest problems in organizations is that the meeting is a tool that is used to diffuse responsibility" ([1] [2])
  
  
• Pew poll on how opinions of US scientists differ from the US population, and public's perceptions of scientists ([1])
  
  
• Pair a "brash, young scientist" with a "wiser, older scientist" to maximize innovation ([1] [2] [3])
  
  
• Google Earth Pro is now free, lets you get high res stills and movies of anywhere on the planet ([1] [2])
  
  
• People told a placebo was "expensive" had twice the improvement as measured by physical tests and brain scans ([1])
  
  
• Blind men successfully train themselves to "see" using echolocation, and brain scans determine that they are using the otherwise unused visual centers of their brains to do so ([1] [2] [3] [4] [5])
  
  
• Rather than modeling crowds with attraction and repulsion between agents, only avoiding anticipated collisions behaves closer to real humans ([1])
  
  
• Xkcd comic: "I can't wait for the day when all my stupid computer knowledge is obsolete" ([1])
  
  
• Xkcd What If: "Getting to space is easy. The problem is staying there." ([1])

More on what to advertise when there is no commercial intent

Some of the advertising out there is getting spooky. If you look at a product at many online stores, that product will then follow you around the web.

Go to BBC News, for example, and there will be those dishes you were looking at yesterday on Overstock. Not just any dishes, the exact same dishes. Just in case you forgot about them, there they are again next time you go. And again. And again.

A few years ago, I wrote an article, "What to advertise when there is no commercial intent?". That article suggested that, on sites like news sites, we might not have immediate commercial intent, and might have to reach back into the past to find strong commercial intent. It advocated for personalized advertising that helped people discover interesting products and deals related to strong commercial intent they had earlier.

However, this did not mean that you should just show the last product I looked at. That is refinding, not personalized recommendations. Refinding is all a lot of these ads are doing. You look at a chair, ads follow you around the web showing you ads for that same chair that you already know about over and over again. That's not discovery. That's spooky and not helpful.

Personalized ads should help people discover things they don't know about related to past purchase intent. If I look at a chair, show me highly reviewed similar furniture and good coupons and big deals related in some non-obvious way to that chair and that store. Don't just show me the same chair again. I know about that chair. Show me something I don't know. Help me discover something I haven't found yet.

I understand the reason these companies are doing refinding is because it's hard to do anything better. Doing useful recommendations of related products and deals is hard. Helping people discover something new and interesting is hard. Personalized recommendations requires a lot of data, clever algorithms, and a huge amount of work. Refinding is trivially easy.

But publishers aren't doing themselves any favors by allowing these startups to get away with this kind of useless advertising. As a recent study says, "the practice of running annoying ads can cost more money than it earns." That short-term revenue bump from these spooky refinding ads is like a sugar rush, feels good while it lasts, but hurts in the long-term.

They can and should do better. Personalization, including personalized advertising, should be about helping people discover things they could not easily find on their own. Personalization should not be refinding, just showing what I found before, just exposing my history. Personalization should be helpful. Personalization should be discovery.

Quick links

Some of the best of what I've been thinking about lately:

• Tiny cheap satellites will provide near real-time imagery of the entire Earth to anyone who wants it, starting in about a year ([1] [2] [3])
  
  
• Amplifying motion and color changes in video, which allows augmented perception ([1] [2])
  
  
• Birds can hear the very low frequency sound produced by severe weather and are able to flee well in advance of incoming storms ([1])
  
  
• Nice example of blending computer science with another field, in this case genealogy, to yield big new gains ([1])
  
  
• "An energy gradient 1000 times greater than traditional particle accelerators" ([1])
  
  
• People "don't want to watch commercials, are fleeing networks, hate reruns, are increasingly bored by reality programming, shun print products and, oh, by the way, don’t want to pay much for content either. Yikes." ([1] [2])
  
  
• Everything we know Google is working on ([1])
  
  
• Funny and informative: "Riding in a Google Self-Driving Car" ([1])
  
  
• Google is rejecting security based on firewalls ([1] [2] [3])
  
  
• "Whether you call it a Star Trek Universal Translator or Babel fish, Microsoft is building it, and it's incredible." ([1])
  
  
• "Every dollar a worker earns in a research field spills over to make the economy $5 better off. Every dollar a similar worker earns in finance comes with a drain, making the economy 60 cents worse off." ([1])
  
  
• "I’m a big believer in making effectively infinite computing resources available internally ... [Give] teams the resources they need to experiment ... All employees should be limited only by their ability rather than an absence of resources or an inability to argue convincingly for more." ([1])
  
  
• "We think of it as a one-on-one tutor. It will test you and generate a personal lesson plan just for you." ([1])
  
  
• "Apparently, a sufficient number of puppies can explain any computer science concept. Here we have multithreading:" ([1])
  
  
• Fantastic to see a US president promoting computer programming to kids: "Becoming a computer scientist isn't as scary as it sounds. With hard work and a little math and science, anyone can do it." ([1])