The real answer (in my network-centric view of the world) is that Instagram is worth a billion dollars, a re-filed S1 and pre-roadshow signal to noise diffusion because it makes Facebook’s advertising platform more valuable through increased context. If a picture is worth a thousand words, then context about a photo is probably good for a few Gbytes in a map/reduce job.

What can you learn through Instagram? Where I take pictures. Who I share them with, who follows me and who I follow (perhaps shedding light on not just subject but style and composition). How I color-adjust the pictures provides more clues – am I nostalgic (sepia tones, black and white) or having fun (color over-saturation)? Know who is in the pictures, and where they were taken, and there’s significant weighting inferred for the edges in my page, group and friend social graphs. The data available to advertising campaign management is increasingly rich and timely — if your business depends on campaign generation, then creating richer campaign marketing data is nominally a high return investment.

I’ll be blunt: Facebook can do with Instagram what Yahoo! might have done with Flickr. It’s not about the content, it’s about what the content construction and conversation tells you.

So yeah, I can see why Facebook would spend a billion dollars on Instagram. Andy Balo (Kickstarter principal) provides some other metrics for measuring how far a billion dollars goes, but they’re all trailing indicators. An incremental $40 million in advertising revenues puts $1 billion of market cap back into a company that will be (supposedly) trading for roughly 25x annual sales post-IPO. That’s a leading indicator.

Maybe I’m being way too optimistic, but if Facebook can trawl through my Instagram photo data, then perhaps I’ll stop seeing ads that offer dental insurance to employees of a former employer.

The common wisdom is that digital photography killed Kodak. Digital images were the secondary effect. Networking was the primary. Kodak’s consumer business is about narrative: they thrived because people wanted to tell stories through snapshots of their lifes. The places I remember, to quote the Beatles. Kodak’s tag line was “Take Pictures – Further” for quite some time, a snapshot of both imaging and sharing the thousand words to do justice to the picture.

Kodak had the first digital camera (I had a consumer version of it; it used a floppy disk and took almost ten seconds per VGA quality image). They own a truckload of patents in digital imaging science, color science, and image manipulation. But their business model was predicated on taking pictures, having them developed, printed, and mailed to relatives in Iowa. It wasn’t just film; it was chemicals, paper, and the photofinishing “mini labs” that popped up in every chain drug store, camera shop and mini mall. As soon as that entire vertically integrated business was challenged by kids with smart phones posting pictures to Flickr, Photobucket, and now Facebook, the consumer business entered its denouement. Doesn’t matter that Kodak invested in Ophoto for digital image sharing, or that they make a really nice waterproof digital video camera. The higher end camera companies were able to continue to push professional grade innovation down into the consumer space, and for hack photographers like me, better glass and effectively zero cost of “wasted frames” meant that I began taking many, many more pictures than before. Every picture I take goes into an email, through MMS, up on SmugMug, or onto Facebook. Kodak adds no value to those processes, so I became a Kodak non-consumer.

Kodak bet against networking. Their business model was not predicated on telling analog stories using digital images. Adobe (Lightroom and Photoshop, not to mention the rest of their suite) and Smugmug (for high volume sharing and archival) represent the endpoints that Kodak could very well have defined had they bet that broadband networks would be cheap, ubiquitous and intimately attached to the vast majority of imaging devices (read: camera phones). They would have created a vertically integrated value chain from image capture to context (borders, ribbons, tags, clean up, editing) to archival to personal narrative. It’s not just the consumer business — Kodak also had a large medical business (X-rays and medical films). If you’ve read stories about remote radiology or remote diagnostics, you’ve seen how networking and digital imaging conspired against Kodak there as well. Both aspects are necessary; simply having great digital imaging but no networking capability means you’re making analog prints and using FedEx as your network layer to get a second opinion.

Moral of the story: You can’t stop Moore’s Law and Metcalfe’s Law from disrupting businesses. If your business model changes as a result of netowrking, you need to figure out how to deal with it. Once the publishers realized that amazon.com is a re-intermediator, not a dis-intermediator, and that building marketing, pricing and distribution relationships with amazon.com would actually increase sales of their entire front and back catalogs, they survived. Everyone who had a Brownie camera, who waited patiently for the fat picture envelope to return from Rochester, New York, is a bit sadder that the Kodachrome is being taken away.

Midway through 1995, Sun Microsystems was riding a wave that wasn’t quite yet attributed to Internet surfing, but had its origins in the tidal forces that made the at sign part of our vernacular. We had introduced a new programming language called Java that May, and during one of our engineering conferences, I sat at rapt attention while our keynote speaker — Steve Jobs — introduced his talk with the trailer for Toy Story. Having learned a small fraction of the interactive computer graphics canon (enough to say “Gouraud shading” with a straight face), I was immediately blown away. The texture mapping, the motion blur, the quality of the rendering — it didn’t seem possible within what was “normal” data center architecture, until Jobs told us how it was done and what part Sun Microsystems had had in the rendering farm that generated the movie, frame by animated frame. The “thank you” poster with Buzz Lightyear and a Sun logo that we received as a party favor at that conference still hangs in my home office. Up to that point, computer animation was an interesting experiment, but it hadn’t entered the mainstream; four months later I was taking my then-four year old daughter to see Toy Story at its Thanksgiving weekend premier, and stayed until the very end to see the sysadmin and server credits.

Toy Story’s release marked the point at which it became socially acceptable to be a nerd. Email, the web, blogging, social media and wireless client devices bled nerd colors onto everyone else, but the summer of 1995 was definitely the tipping point. And Steve Jobs pushed us front and center. Family conversations that opened with “What do you do?” no longer involved companies named Xy-, Mega- or something-ix; they revolved around talking about what computers could do in the social mainstream.

Of course, Steve Jobs took things to the next level of abstraction. After making entertainment a function of computers, he made computing a function of fashion. We love our iPods, iPhones and Macs because of their elegant design; celebrities talk about what kind of phone they use and we follow them on social media sites. We’re all nerds now. We even have our own TV show (and I’m referring to Big Bang Theory, not Eureka or The Guild even though those are equally outstanding answers).

For every comment about Jobs’ style as CEO and engineer, consider this: What if most executive boardrooms, state legislatures, and our Congress functioned with the same ruthless passion? What if a design — for a bill, a strategy, a foreign policy — that was so obviously underwhelming was simply met with “This is stupid” and forced into re-work? What if we worried incessantly about the design and experience of our work product, and let the profits come as a result of a job well done? There are management and leadership lessons beyond those recorded in the Harvard Business Review, but significantly more valuable.

Thanks, Steve, for the impact you’ve had on nerds everywhere. To infinity, and beyond.

Background: Facebook will pre-define lists for you based on your education, work, geography and other easily sortable criteria. It then suggests people to add to those lists, usually based on mutual friendships with those already in the list, or common data such as both having attended the same school. It’s a nice big JOIN problem at its finest.

Problem: I have a friend who is married to one of my fellow Tigers. She and I have mutual friends who are all alumni as well, and I tend to think of her in the “Princeton” category. So I put her in the Princeton list (at Facebook’s suggestion, I should add, again, probably based on common edges in our social graphs).

The Leak: My friend was notified of the list addition and asked to confirm “Princeton” as part of her education (again, assuming that was the criteria that generated the suggestion). Major, major privacy #fail: The fact that I add anyone to a list is private; it’s how I sort my friends and acquaintances and my criteria and grouping algorithms are completely and totally my business.

Unless Facebook would like to tell us that list additions are communicated to the other party, this seems to intrude in my own categorization. What if I create a list called “Ignored Former Coworkers” and add people to it, mostly so that I can avoid their updates that I find distracting? Are they notified of my feelings toward our time together? My guess is that if the list title doesn’t match one of the criteria used by Facebook, there’s no additional information leakage, but my two simple experiments to confirm this weren’t conclusive.

Advice: Create a list for your close friends and family members, and ignore or delete all of the other ones until Facebook figures out how to avoid leaking our non-obvious inclusion criteria.

Turns out the Image Capture tool (built into Mac OS X, in the “Applications” folder) will see your iPhone as a camera device, and let you download some or all of the photos to a selected folder. This is now right up there with Senuti as one of my favorite “gimme back my stuff” applications.

The more we share, though, the more intriguing it is to see who might be consuming that information. Who doesn’t want to know the people who visit their Facebook profile the most often, looking at pictures or wall postings? It’s the adult equivalent of finding out who put you on the list of the “cute guys” in 7th grade (this never happened to me; I’m guessing here). If we post things in public to share and elicit reactions, why not track the quietest actions?

Because you can’t. Those applications that claim to show you who visits your profile or your “favorite friends” are virus vectors. Here’s how it works: Someone creates a Facebook application with an appealing name. The application promises to give you some personal and revealing knowledge, and in return you grant it permission to look at your personal information. The problem is that the personal information accessible to Facebook applications isn’t sufficient to determine who has visited your profile on that site. It’s not available to applications, and making it available would create privacy headaches that dwarf anything Facebook has in play.

When you grant a “Profile Tracker” (or stalker, or creeper, or anything else that implies unwanted online hovering) application access to your personal information, it typically does the following:

Grabs your email address(es), pages that you like, and list of your friends.

Posts an item on your friends’ walls so that the virus spreads itself.

Uses the information gathered to build a better phishing scam, social engineering attack, or brute force attack. If you, say, use digits from your birthday as a security code like an ATM PIN, or online banking PIN (which is a horrible, very bad, egregiously dumb idea, by the way), and your birthday information is in your Facebook profile, then you have provided more useful information to a potential identity attacker. That’s what these viruses do – they gather information to make more calculated, and therefore more fruitful, attacks on other parts of your online life. Someone with a Yahoo email account is more likely to bite on a Yahoo phishing scam asking for password information for “security purposes” than someone who only uses Gmail.

Bottom line: Be very careful giving any Facebook application access to your personal information. If you think you clicked through on something you shouldn’t have, check your privacy and security settings on your Facebook account. Don’t use easy to derive passwords, especially if you share a lot of your personal information (How many banks use “Name of your pet” as a security question? – I can probably name random people’s pets by looking at their Facebook public pictures; couple that with some banking services page “likes” and it’s much easier to brute force a password reset).

Facebook does track the pages, interests, likes and profiles you visit, and uses that information to govern the advertisements you see on the sidebars, the items that show up in your news feeds, and the ordering of various search results. It’s all very you-centric – not based on what others are clicking through and to, and not revealing anything about others’ browsing histories. It still think it’s creepy that I can comment on someone’s post, and the sidebar advertisement changes to reflect a keyword I might have used in a different context. But that’s the nature of online advertising; removing the annoyance factor also reduces our ability to use social networks as recommendation engines. I’ll keep my eyes front and center on the screen in return for useful information from the well-behaved pages that I follow, and I don’t click on things if I don’t know where they came from.

That’s probably the way we were all taught to behave in public.

Delicious is dying because it’s not generating revenue for Yahoo, and therefore didn’t fit into the pareto of projects that continue to be funded. That’s a business decision. Cries for it to be made into an open source project are just plain dumb. It’s not the code base that people care about; it’s the site and its aggregated content. If Yahoo, a company that understands scale and services, decides it can’t be run profitably, what organization would possibly fund it as a free service? Throwing code over the wall doesn’t “rescue” a project without a strong community to manage the code; throwing del.icio.us over the wall means it’s likely to hit the pavement, hard.

I haven’t used Delicious in more than two years. I stopped using it because I found that URL shorteners were performing much the same service when coupled with my usual sources of context: Twitter, Facebook and blogs. If I can see how has clicked on a link that I’ve shared, or share it with a comment, note or longer explanation, that’s far more useful to people than simply stuffing a bookmark into a list with a note. With a variety of tools to keep your bookmarks synchronized across multiple devices, the utility of Delicious simply declined over time.

And there’s the rub: Yahoo is in the business of “connecting people to the things they love.” They started off in the information organization business (sounds like their up-the-101-neighbor now) and realized that the business of prioritizing delivery vehicles to readers is called advertising. Somewhere along the way, Yahoo tried to become a content company, and instead of building on the immense value of something like Delicious, they focused on the targets of Delicious links rather than the links themselves. If you know context and destination, those are very powerful ingredients for divining what is an implicit preference of mine — pretty powerful grist for the advertising mill. How on earth could Yahoo not turn Delicious into an amplifier for their advertising business, unless they were focused on links as content instead of links as indirection to the things we love?

I’m fearful that this move foreshadows something even less delicious in Yahoo’s future: a downward spiral of trying to cut your way to increasing profit. If the cuts aren’t focusing investment on areas of growth, you never exit or reverse the spiral.

Which is why I’d be happy if it’s proven that P != NP, and in fact there are computing problems that remain unsolveable, no matter how much iron is dropped on their algorithmic heads. At first glance, my opening aside and this statement seem contradictory – how could I possibly take the “under” on a problem-solving challenge? The answer is that I think there’s more potential good computer science that came stem from knowing there are problems whose domains lie outside of the procedural programming approach. I’m taking the over on computer science creativity. We’ve seen the proof of Fermat’s Last Theorem in the last twenty years, and settling that debate didn’t dampen mathematics research one epsilon.

As humans, and essentially analog beings, we deal with boundaries all of the time, from the speed of light to the incompressibility of matter (beyond a point), to the fact that gravity acts on all ages equally. We similarly deal with non-determinism daily as well, whether it’s variation in the fruit we buy in the store, a flaming red sunset, or the benefits and difficulties of genetic variation. I was initially going to point to Greg Bear’s Darwin’s Radio/Darwin’s Children series for further reading, but this week’s discovery of arsenic-loving bacteria made weird life resemble weirder art in a good way.

So we live in a world that has computational and physical boundaries and is rife with non-determinism. Haivng P and NP in distinct computational classes just sort of fits in.

I started writing down this argument based on a question in a job interview. The engineer screening me had read my blog entry about P=NP and asked me if I thought cloud computing would help in the answer. I took the liberty of answering a related question, namely, that I don’t think P=NP, and that cloud computing will necessarily improve our ability to attack problems that lie on the NP side of the Venn diagram.

As I’ve been plugging along in Dick Lipton’s book about P=NP, I’ve been struck by the number of times randomization, average case versus worst case, and counting or sampling mechanics come into play. These are the workhorses for handling problems that refuse to fit neatly into a deterministric, Turing-machine defined box. If P != NP, then it forces us as computer scientists to derive new problem solving techniques that involve approximating algorithms, using randomness to find first-fits, sampling and Monte Carlo simulation, and cross-science applications like simulated annealing (injecting noise into a system to try to overcome local minima and find a better, global solution). The real-world target domain for many of these problems are themselves (as I mentioned earlier) non-deterministic and highly varied. It would be great to use an NP algorithm to project a protein folding that would crimp a strand of the AIDS virus; but the virus itself is mutating and evolving. It just has to be good enough to work, and hopefully work across an ever-changing set of uses cases. Algorithmic elegance or completion is prettier but not economic in many senses.

This is where cloud computing comes into play – many of these approaches require large data sets, or large sample sizes with a few operations per item – and that maps well into the highly elastic infrastructure provided by clouds. I believe one of the primary impediments to cloud adoption is not converting today’s applications to be cloud-aware but instead finding out that we haven’t written good, elastic applications. Yet. Cloud computing isn’t going to help us solve P=NP (that I can tell) but it will act as a proving ground for the very real solutions that we attempt to converge into a sense of “good enough.”

The challenge of having P != NP — of the ring just outside the researcher’s grasp — stimulates creativity, collaboration and what I think is a quintissentially human quest for something just beyond the proven horizon. That’s an over bet that I’ll always take.

Finally: The question was asked by someone at Juniper Networks, where I’ll be starting as the Chief Technologist of the Junos Space business unit. No sooner had the question been posed then I knew I wanted to work for this company; I’m thrilled that they felt the same way and I can start the new year in a new gig.

Computer scientists call this “The P=NP Problem”. I started thinking about it again recently, after a 27-year hiatus, due to a “plate o’ shrimp” set of events: A huge buzz in the computing community around a proposed solution to the problem, thinking about my former CS-conspirator Tom (who actually got this stuff) and the publication of my former thesis advisor’s book on the problem. Thesis advisors retain an avuncular relationship with their students for decades, whether or not the advisors know it — and in the last year when the Princeton community lost a number of professors who had advised the classes of the mid-80s, I saw an enormous outpouring of emotion from my friends. Seeing Dick Lipton’s name in the press again was equally exciting and challenging, triggering some of those feelings, making me shell out nearly a Benjamin for his book. I started reading two days ago, thinking that I’d do an amazon.com review, but got thinking about hard CS again and here we are.

What is the P=NP problem? It’s been unsolved for the better part of fifty years. It involves a famous mathematician (Kurt Godel, attracted to Princeton by Einstein) and the father of the modern computer (John von Neumann) exchanging notes about things so complex and divisive that any news about a potential solution immediately accretes a small conference to explore the idea. If you can solve it, you win a cool million dollars as P=NP is one of the Millenium Problems.

Basically, there are two types of problems that we look at in computer science. Some are easy to solve, and we know that we can solve them in a predictable and bounded time. If you have 1,000 numbers to sort, and your computer can sort them in 1 second, then sorting 10,000 numbers will take 10 seconds or so (Acutally less, but I won’t get into the math here). These are known as “P” class problems.

There are other problems that have answers that are easy to verify, but the answer can’t be calculated in any predictable or bounded way. Here’s a goofy example: you take your really big truck to Costco for a major snack food run. Unfortunately, you went when you were really hungry and the Giants were playing a late game, so you overloaded just a bit. Each box of candy, each pallet of soda, and each 100-count frozen burrito box has a particular weight as well as a specific, non-crushable height, length and width. You know the volume of your truck and its weight capacity, and you know the maximum interior height, length and width of the cargo space. What’s the best way to pack the truck without crushing anything, knowing you have to get chips, burritos, drinks, candy and at least one healthy alternative? You know that some of those boxes aren’t going to fit, so you have to make trade-offs. You need to worry about the total width, total length and total height of the stacked boxes, the total volume of everything inside, and the total weight of what you load on those rear wheels. It’s easy to tell if your trial packing meets the criterion – the rear door closes without any crunching sounds, and you can check off the list your friends (and spouse) gave you. Checking the solution is easy. Finding it is impossible. Which gives you a perfectly good reason to swear at your mountain of boxes in the Costco parking lot.

The reason the truck packing problem is impossible to solve (even though we can check answers very simply) is that calculating an answer boils down to determining every possible combination of box packing and stacking, and then checking each one. There’s no easy, well-sequenced way to solve this one. If you have only five boxes, there are more than 100 ways to pack them. With a thousand boxes and a photographic memory that helps you verify the shopping list in a second, you’re going to spend years finding a solution (however, the Swedish Fish will not have passed their expiration date, although the burritos will be ancient history). These are known as “NP” problems, and despite the goofy example, they exist in the real world.

What’s the point of P=NP? It asks whether the two classes of problems are the same, or fundamentally different. If they are the same, it means that NP problems have deterministic, bounded solutions that we simply haven’t discovered yet. If they’re different, then there are NP problems for which no such solution exists. There is a good news/bad news conundrum wrapped around the question of P=NP as well, because the answer challenges our existing approaches to, and applications of, computing.

Here are two problems that fit the NP category: when you are using a “secure” internet site, your network traffic is encrypted using some very large numbers to generate the encryption. Deriving the large numbers underlying the secret channel is a problem in NP, and you’re basically pretty happy that it belongs in that class. Since brute-force guess-and-test attacks will take years (as in millions of years) to solve the problem, it’s effectively impossible for an attacker to reverse-engineer the security mechanisms and tap into your personal or financial information. Having problems in the NP space is quite helpful when you want to ensure your computation only goes in one direction (hence the nickname “trapdoor algorithm” for this type of computation).

On the surface then, having P and NP be different is a good thing, and we should be cheering on the theoreticians trying to prove that result. But here’s a different NP problem: Given a long string of proteins, like a virus or genetic marker, find another string of proteins that will fold neatly into it, as a vaccine or treatment. Protein folding is way out there in the complexity spectrum. If P and NP are different, then we lose the ability to fully automate, with a higher degree of certainty and precision, some of the pressing issues in biochemistry and computational chemistry.

It may be another year, or ten years, or hundred years before we develop the computing tools and disciplines to prove the P=NP question either way. Fermat’s Last Theorem stood unsolved for almost 360 years until a bit of computer science work shed light on a novel approach. Insights needed to tackle P=NP may come from other disciplines, and that’s the benefit of computing permeating so many aspects of our social domains.

Facebook is now the largest referrer of traffic to my blog. Those are readers who would have previously read the entire entry on Facebook, without seeing what other goodies and random hollering I have in the wider universe of snowman happenings.

Blog traffic was up pretty substantially, due both to increased referrals as well as more users going deeper than one page. Decreased bounce rates are a win.

In this case, RSS no longer stands for Really Simple Syndication – the feed isn’t syndicated at all. It’s more about simple sharing, and it’s simply a good idea.