(1) Is innovation always data-driven? Mark makes a rather compelling argument that innovation is rarely the result of someone seeing a process or tool and automatically knowing how to make it better—it’s usually a combination of copying and trial-and-error. This leads me to think the argument is actually that innovation involves a process of evaluating how well something is working (e.g., measuring its success), and iterating to make it work better. I don’t know how much I buy it as an absolute—that all innovation is data-driven—but I suspect that it holds for the vast majority of cases.

(2) How many of us should or will be innovators?  Mark makes a compelling enough argument that any given individual in a society relies more on copying than he does on innovating—this is not to say that innovation isn’t important. The argument seems to be that any given innovation scales easily, and it scales more and more easily the more connected a society becomes. With the Internet, he argues, we need even fewer individuals innovating, because an idea that pops up in southeast Asia can make its way around the world in no time. He also makes a point about language—so I think the advent of online translation tools are also part of his argument. He seems to be saying that most of us are organized to be copiers or consumers of other people’s innovations.

What interests me about these two points in combination is what that means for how we organize as a society—how we distribute resources and human capital.

On the point about how many of us should be innovators, I think Mark is both right and wrong. He’s right if when we talk about innovation we mean new idea creation, or what I’ll call isolated innovation. That’s the idea of innovation that two guys in a garage can build something revolutionary and change the world. We saw a lot of this type of innovation in the past century. But I think Mark is a little bit wrong if we want to talk about the type of innovation we need to see more of in the next century, which I’d argue is not going to be isolated but rather systemic innovation.

The biggest problems we face this century are systems problems—climate and environment, public health, sustainable urban development, etc. These types of problems won’t be solved by two guys in a garage. Instead we need data-driven innovation at scale: we need lots of well-funded scientists collecting, sharing and analyzing huge data sets about complex systems.

If you agree with me so far, I think this suggests a few different things.

• We need to divorce the idea of innovation from startups. Innovation is as much about existing, large institutions as it is smaller, new ones. Instead of talking about start-ups, we should focus on R&D policy—and make sure that it is size-agnostic.
• We need to be able to collect, share and analyze data across institutions for the purposes of innovation. This means creating open data standards, especially in the public sector. Proposals like the EU Open Data strategy and work done on Data.gov are encouraging.
• Some of the data we will need to analyze is going to be personal data, so we need mechanisms to support consent in the innovation process. This is why projects like the one John Wilbanks is leading, Consent to Research, are so important.
• Analyzing the large sets of data that will drive a lot of this innovation will mean using the cloud. It’s just not cost-effective to expect everyone to run their own data centers for this type of computation. We need to reduce barriers to access these cloud services, such as restrictions on cross-border data flow. The APEC Pathfinder project is one encouraging effort to achieve this goal.

Just a few thoughts on a very big subject. 

My colleague Nicklas Lundblad has a good post this weekend on the virtues of teaching computer science. I felt so immediately jarred by a missing piece of the argument that I was compelled to sign in and blog for the first time in months (an absence I feel tremendous guilt about). His argument is spot-in, but misses a very simple yet critical addendum.

My own education in computer science of course influences my views on the matter. I often tell people, somewhat jokingly, that despite beginning my technical studies at an all-women’s college, I have an ex-boyfriend to thank for the decision. My high school sweetheart was desperately passionate about technology, and in my freshman year of college I made an attempt to bridge the physical distance between Boston and California by bridging an intellectual one instead.  I endeavored to take the introductory CS class at Wellesley in large part because it was what he was so passionate about….but I promptly fell in love with the subject myself.

CS is a liberal art. This was a time-honored debate our professors engaged us in. Many of them came from engineering schools but wound up teaching the subject at a liberal arts school, and so had strong views on the subject. As do I, now. It is absolutely a subject that teaches you how to think—not just how to build. Building is a side effect of the discipline, a very useful one, but nonetheless a side effect.  So here, I agree 100% with Nicklas when he says:

Computer science offers a new way to understand the world, to think about it as algorithms and data structures and data sets. That is extremely powerful. So should we teach kids coding instead of teaching them to cut and paste in word processing software? It does not seem to be a very hard question does it?

Where I disagree is that I don’t think coding is enough of a prerequisite. Like advanced biology or chemistry it should be included in the standard curriculum. But the core class we aren’t teaching our kids that we need to be is Internet Architecture—that should be like government or civics classes are today: a prerequisite for graduation.

After college I somehow found my way into MIT’s Technology & Policy Program. Like most other new graduate students, I spent the first few weeks of my time at MIT on the job market looking for a RA or TA position to help pay the bills. In our case, the TPP program required that we find a research advisor also, which usually dovetailed with finding an RA position. I stumbled into Dave Clark’s office one day to ask for a job, completely unaware of his stature in the Internet community. He asked if I’d read his paper about the end-to-end principle and I said no, could he tell me about it? What followed from that initial conversation was by far the most meaningful educational experience of my life.

For the next two years I would work with Dave on research topics relating to identity and net neutrality. Each week, I would sit down with him for an hour and get an individual tutorial on how the Internet worked and the history of its design. I learned early on what an IP address was, and how TCP/IP worked—amazingly, I wound up graduating college with a CS degree without understanding those concepts. I could code when I left college, and I could think about problems through this lens—it had undoubtedly changed the way I thought about the world. But despite a college degree in CS I fundamentally had not learned the basic governance principles of the Internet.

To appreciate the mechanisms through which information can be exchanged and manipulated is to appreciate the mechanisms through which people are able to organize and communicate. To appreciate the processes through which technical standards are agreed upon is not unlike appreciating the processes through which laws and regulations are agreed upon. We teach all high school students how a bill becomes a law. Why don’t we teach them how a packet becomes an email?

I went to college with a lot of women who were interested in politics and economics, who aspired to careers in public service, and whose interests lay in questions of how society organizes itself. The women I know who are now working in these fields are having and will absolutely continue to have an impact on the world. But I would dare any of them to name a field of public policy not currently impacted by or disrupted by the Internet. Understanding how to write code that builds an isolated piece of technology is like understanding how to read and write, or knowing the ins and outs of a particular subject like Biology. But understanding how the Internet works is like understanding the way society is governed. The architectural design should be taught in high school, the same way we teach about the design of the US Constitution.

This might seem like a radical recommendation, but there it is. Lessig wrote “Code is Law” which is basically the same idea, but had the unfortunate side effect of focusing everyone’s attention on coding. And I suppose that most engineers get to a place where they understand the architecture of the net by starting out writing code. But I don’t believe that being able to write code is a prerequisite to understanding the design of the architecture, and therefore in my view teaching the architecture is the mandatory prerequisite we don’t have yet. We should be equipping people to answer the simple question: how does the Internet work?

I suspect we’ll be seeing more and more applications like this. Those in the personal healthcare arena—e.g., personal health tracking that predicts risk for disease—will seem largely uncontroversial at first, but may pose the hardest questions in terms of how to ensure fair access to insurance and preventative care. Others like this that are more focused on public healthcare and/or social behavior modifications I suspect will provoke more ire, uncertainty, fear, and concern. For example, it probably doesn’t take much imagination to think of a world in which the social graph and one’s behavior in a social network are used as predictors of an individual’s likely political tendency—or, put another way, the likelihood that one is a terrorist from the point of view of a given state.

It would be interesting to see, for a given use case, a mapping of the pre-digital algorithm to the post-digital algorithm. That is, what inputs are used to screen for alcohol abuse today, and how do those inputs change in a digital era.

On an unrelated topic: as I went to document key takeaways from Incognito tonight, I realized how poor I find the digital medium for note-taking in books and generally for any sort of non-fiction reading that you may want to flip back through. I’ve noted this observation elsewhere in the past, but was struck again by this drawback tonight.

I’d welcome suggestions on better ways and tools to document said bibliography, but for now I fear it’s just going to be a semi-disorganized list.

The quantified self trend is one which I’ve come to think is the future of healthcare—data-driven research and self-improvement. What I find most compelling about some of the QS examples are the way in which otherwise irrelevant, mundane observations about one’s life can lead to sharp insights and improvements. Which is why I find a new project from John Wilbanks particularly interesting—as he describes it:

 The idea behind CtR is simple: make it easy for people who want to share data about themselves for scientific, medical, and health research to do so. It’s not centered on intellectual property, though it does touch on it. It’s more about privacy, and in particular, about making it possible for people to get informed about what is possible with their data and how beautiful research can emerge if enough genomes, enough biosamples, and enough other kinds of data can be shared and connected.

When we think about the future of scientific, medical and health research it’s easy to stop at the types of medical data we collect today—blood pressure, genomes, drugs consumed, etc.—but the QS movement would suggest we could learn more by going farther into the mundane and unexpected sources of data.

Which brings me to Flu Trends, an indicator for the incidence of flu in a given area based on search behavior. It can sometimes feel, especially in some privacy circles I move in, like an overused example, but an important one. Flu Trends speaks to an issue at the core of the change we need to see in how we think about privacy—the idea that data must be used for the purpose for which it was collected, otherwise called purpose specification.

In a talk I gave at last year’s OECD roundtable on the economics of personal data I emphasized this point: improving the world in the 21st century is going to depend at least in part on drawing intelligence from vast amounts of loosely collected information. More often than not, surprisingly useful intelligence will come from the most unexpected sources of information—and we need a flexible enough policy environment to support the exploration of those sources.

I’ve been traveling for a week, something I have been doing less and less of over the past year or so. I’d flown into Boston on Tuesday night arriving 2hrs late at roughly 2am, worked from Cambridge for a couple days, then rented a car Friday to drive to a wedding in Hanover, New Hampshire. Being the amazing advanced-planner that I am (ahem, or rather, not-so-amazing) I’d taken off from Cambridge and high-tailed it through to New Hampshire without thinking twice about cash or gas. Along the way, I’d hit a toll plaza (forgot about those!) and had to apologize profusely to the toll collector while I pecked around in my coin purse to produce 96 cents, not quite the $1 he was hoping for but he kindly let me through. I realized I may have been slightly unprepared, and was now without any cash, watching a rapidly falling gas gauge, and approaching nightfall.  An hour or so later, just before 8pm and with 20miles to go, I decided the near-empty tank wasn’t going to get me all the way there and I better pull off for gas before stations closed. I pulled into an isolated gas station a few miles off the highway only to find that my credit card company had disabled my account. Lucky for me, since I had no cash on hand, I had a different card on hand—so I paid and went my merry way.

I arrived at my hotel to find a voice mail from the fraud detection department asking for a return call. After verifying my identity, they asked about a slew of what seemed like fairly obvious charges: “Did you charge $7 to United Airlines on Tuesday? $10 to the Boston T on Wednesday? $300 to Enterprise Car Rental on Friday? What about this $30 for <some random NH gas station> earlier tonight?”

Seriously guys? I bought an airplane ticket from SFO-BOS using your credit card a month earlier. I then reserved a hotel room for Friday and Saturday nights in Hanover on your credit card. Then I made a car rental reservation from Cambridge, again with your credit card. So your detection system sounded alarms because I bought a glass of wine on a flight you knew I’d purchased, bought a subway pass in a city you knew I’d arrived in, rented a car you knew I’d made a reservation for, and bought gas for the car you knew I’d rented? Was it really that hard to connect the dots? What exactly have you been investing in or innovating over the past few years, anything at all?

I spend so much of my professional life working with people who worry about privacy invasions resulting from advanced data analytics. Never do I hear people saying things like I thought Friday night: oh dear credit card company, you had the data, you should be smart enough to make my life easier by figuring out what it means. Instead, when companies do this sort of analytics well, I hear outrage over how it is that the company knew what it did, and whether what it knew was actually accurate.

When are we as a society going to get as outraged about the lost opportunities that could have been captured using data analytics as we get about the potential privacy invasions that might arise from collection and use of the underlying data sets? When will the public pressure be of the form “Why didn’t you know better?” instead of “You shouldn’t have known that?”

By setting your predictions for the future in a familiar landscape, you allow yourself to use your memories of the past to help you predict what might go wrong in the future.  If you are only able to think abstractly about the future, then you are much less likely to find specific problems that may arise.

The paper is titled “Predicting Premeditation: Future Behavior is Seen as More Intentional than Past Behavior.” I look forward to reading it in full this week.

At a high level, this led me to ask a few questions about predictions. My anecdotal observations are that in general we “freak out” a bit about predictive analytics, the idea that we might predict a future event based on recorded history. As shown in popular culture (e.g., Minority Report), we spin up all sorts of dystopian fears about what a world driven by predictive analytics might look like.

But ask a scientist who studies humans not machines, and apparently he’ll tell you that humans predict in basically the same way machines do—based on past history. When thought about in the human capacity, my hypothesis is that predictions “freak us out” a little bit less. I have one guess as to why that might be: we understand human predictions to be simply that—predictions—whereas when data and computers are involved we perceive the predictions to have more of a discrete character, and we imagine our use of those predictions to be quite different.

In fact, predictions form the basis of the majority of decisions we make every day. In my professional capacity I’m actually encouraged to document my predictions and the fact-base I’m basing them on so that later we can review our accuracy and revise our internal (human) predictive models in the future. What I find fascinating is this prediction of my own: as predictive analytics become more and more commonplace, we will find more and more reason to fear the outcome and discourage its use. Yet, we generally recognize that better data yields better answers—so the interesting question will be this: what do we trust more, subjective interpretations of history, or data-drive analyses of history?

Last week I came across this story about a new iPhone app that scans moles to detect likely melanoma: “It scans skin lesions, takes pictures of them, measures the diameter, and uses an algorithm to determine the likelihood of melanoma.” I found it surprising to read about (as Business Insider apparently did to write about, titling the post “HOLY MOLEY…”), but in retrospect I’m not sure I should have. It seems fairly straightforward—they always say you just look for abnormal edges, unrounded moles, or very large moles to detect ones that might warrant some additional medical attention. You could even imagine this application storing measurements of your moles and noting when significant changes occurred, identifying the other major risk factor of growing moles. Seems totally straightforward.

Except, the whole “what do you do with this little prediction” part of the equation. That’s where things get messy. Business Insider alludes to the complexity here by mentioning the possibility of future lawsuits over missed malignancies. Forget lawsuits, what about the more straightforward question of whether this is useful at a macro scale or not?

Around the world we face rising healthcare costs, and in the US we face costs that seem to be rising disproportionate to the actual improvements in quality of life or lifespan.  We have here an opportunity (maybe) to put a screening mechanism in the hands of anyone with a smartphone. You could quickly get into questions here of whether screening works and how well (I recall discussion of the efficacy of screening in breast cancer at some point in Emperor of Maladies, which I highly recommend by the way), but setting all that aside for a minute: we’re talking about putting a screening tool in the hands of a layman.

We all have different set points of anxiety—I’m one of those fortunate souls who has a very high set point when it comes to anxiety. I have had to tell a doctor that self-screening for a disease I’m not tremendously likely to have seems like a raw deal to me, because it just creates anxiety and in more cases than not prompts an unnecessary trip to the doctor to allay my worry.

If my phone told me one morning, that mole might be malignant, well I’d call my doctor. That may be a good thing, if the prediction is likely to be accurate. But the thing is, the developer of an application like this has every reason to over-communicate risk—and to bias toward false positives. As someone with Mediterranean skin whose had her share of moles removed and possible-malignancies investigated, I might wind up calling my doctor a fair bit with this tool.

At some point I may delve into comparisons between this type of directed, action-oriented risk and the longer-term, wellness oriented risk (where I might classify, for example, 23andMe)—I suspect that both the types of risk at issue and the ways predictions of risk are communicated impact the spectrum of responses available to us, and the likelihood of impact. That seems the core issue here—can this app achieve impact with its predictions, or is it likely to make poor predictions that lead to higher costs all-around? And are layman able to process and act on the predictions given to them in ways that benefit society at large?

Eagleman’s book appears to describe the science behind the claim that our brains are made up of competing rivalrous modules of activity. That alone is fascinating enough. But where it should get really interesting is the final chapter, in which Eagleman sketches out a theory of jurisprudence that bases punishment on predictions of future behavior, as opposed to factual analysis of past behavior. Note he is talking about punishment, as opposed to the basic determination of wrongdoing.

I find the general idea a really interesting one. Could we predict—based on a person’s biology, upbringing, and the context of his life at the time—whether or not he is likely to be a repeat offender? And if we could, should we?  I think Eagleman is right when he says:

Neuroscience is beginning to touch on questions that were once only in the domain of philosophers and psychologists, questions about how people make decisions and the degree to which those decisions are truly “free.” These are not idle questions. Ultimately, they will shape the future of legal theory and create a more biologically informed jurisprudence.

It seems likely, though, that it won’t just be neuroscience answering these idle questions. That is, a tremendous amount of data is required to do the type of analysis Eagleman is referring to here—the data about upbringing and life context. It’s both neuroscience and big data analysis that are leading us down this path. There is a related set of questions, then, around access to and use of data—many of those questions are being sorted through today, and crystallized into public policy in ways that will be more and more relevant to all aspects of our lives in the future.

I’ve yet to sort out exact details about when I’ll be in Boston and what my work-plan is for my time there, but in the meantime I hope to be blogging more often. I anticipate that I will use my blog in part as a repository for interesting reading on various types of predictive analytics and the ethics surrounding their use. Expect to see more of that, with some of the same old contemporary commentary thrown in.