Deep Learning & Digital Medicine The National Health Service of the United Kingdom recently entered into an agreement with one of Google’s deep learning divisions, Deep Mind, to share heavily anonymized data on 1 million patient eye scans. The goal is to use the pattern recognition capabilities of deep learning technology to try to spot common eye diseases earlier. This is an ideal use of deep learning. You have a large amount of structured data – meaning that you know what you’re looking at (eye scans) and you know outcomes (the eventual presence or absence of eye disease). Because deep learning is, to a large extent, a black box, it is difficult to know what process the system will use to detect disease. It is possible that the system will develop new ways to spot incipient eye disease through a quasi-evolutionary process of trial and error. The problem may be that even a million eye scans may not be big enough data to yield a positive result. Deep learning typically depends on massive data sets which is why Google and Baidu have excelled at attracting deep learning talent. Application to Marketing: Digital medicine is a hot area right now. Many pharmaceutical companies (and pharmaceutical agencies) are looking at the stronger IP protection offered by digital products as a way to create more value and loyalty than medicines with limited patent protection. An algorithm for eye disease detection will not go off-patent in a few years. At first, deep learning looks like an effective way to generate these technologies. However, deep learning is far more effective on gigantic data sets. The sample sizes of the drug trials pharmaceutical companies run look puny by comparison. But the key to deriving valuable IP from deep learning is backpropagation, which allows you to examine, evaluate, and correct the steps that a deep learning system has used to reach a conclusion. If a deep learning system creates a more accurate way of evaluating eye scans, backpropagation allows a company to understand and patent the algorithm. Next Steps: If I were working at Novartis, I would partner with a large healthcare services provider like Kaiser Permanente in order to acquire and evaluate vast amounts of anonymized structured data like electrocardiogram test results, licensing deep learning technology from a company with strong backpropagation capabilities that is looking to monetize like Clarifai and then patent and sell the resulting algorithm under a three-way revenue sharing agreement. Assuming you could make it all work under HIPAA. Not that I’ve given this much thought... Read More Black Swans & White Whales Black Swan Data, a startup that claims to help companies improve their marketing by analyzing vast amounts of internal and public data recently scored £6.2 million in series B financing. (Insert compulsory Brexit pound devaluation joke here.) Big data is the blessing and curse of modern marketing. Marketers assume that there must be patterns hidden in the vast amount of data they collect. Unfortunately, the data is largely unstructured and resides in multiple locations. A company that was capable of detecting meaningful, actionable patterns would be very successful. So, is Black Swan Data this mythical beast? Or is it just another pretender tapping into Google API’s and mapping that data against internal sales figures using attractive, but ultimately unilluminating graphing libraries. Application to Marketing: I’m not sure about Black Swan Data. The usual indicators of player/pretender status point in conflicting directions. On the one hand, they have impressive clients (Pepsi, Disney, Unilever) and impressive financial backers (Albion, Blackstone, Mitsui). On the other hand, their website could not be less enlightening. Every claim is completely generic and apart from a brief mention of “machine learning” it is unclear what kind of technological breakthrough they might have achieved to solve marketing’s big data problem. One small nit that makes me suspicious of the company is their overuse of generic threejs.org animations on their website. It only looks impressive if you don’t know what you’re doing. Next Steps: It would be really nice if a small startup solved this problem instead of Google. It would also be nice if the Jets won the Super Bowl. Read More It’s in your genes DNA is small. Remarkably so. Yet it stores a vast amount of data. Even though our existing digital data storage technologies have become very efficient, you could still store the equivalent of 100 massive data centers in a shoebox if the data was encoded in DNA. Scientists at Microsoft and at Harvard have been attempting to do just that. Microsoft recently announced that they had managed to store 200 megabytes of data, including 100 literary classics in a drop smaller than the tip of a pencil. And, unlike digital storage, DNA lasts for hundreds of thousands of years if kept in a cool, dry environment. So what’s stopping this? Well, it is quite expensive to encode and decipher DNA. Although there is reason to hope that genomic testing will drive down prices, you’re currently talking about thousands of dollars. However, it isn’t hard to imagine a time in the future when all of your data – every book you’ve read or would like to read, every song in the history of music, every picture ever taken by all of your acquaintances could be stored away in a device the size of a USB stick. Application to Marketing: Data used to be expensive. It was expensive to create. It was expensive to analyze. And it was expensive to store and retrieve. The expense meant that data (stored in cardboard boxes) was considered valuable. Digital technology made data cheap, but it didn’t make data durable. Which encouraged people to believe that data was ephemeral by nature. Snapchat is the natural expression of a generation that grew up thinking of data as cheap and temporary. DNA storage allows us to imagine our data existing for forever (in practical, human terms.) Will that change our relationship to data? Next Steps: No idea. But I wouldn’t be surprised to see a startup that offers to encode your family pictures on DNA soon. Read More Sigh, yes, I’m going to talk about it. Pokemon Go. One of my designers confessed that she had spent her weekend pursuing Pokemon Go characters all over Brooklyn and Queens. At one point, she had become so lost that she had had to call an Uber to come pick her up. If you have not heard of Pokemon Go, then you must have made a conscious effort not to hear about Pokemon Go. Because it’s unavoidable. Pokemon Go allows you to create a virtual character that can catch, train, trade, or battle characters from the Pokemon universe. But (unlike in the original Pokemon games) you find these characters by walking around with your mobile phone seeing a version of the world, much like the place you’re in, populated by Pokemon characters. So, it’s a true augmented reality game, created in partnership with augmented reality powerhouse Niantic Labs which recently split off from Google. Application to Marketing: In the name of all that is good and just, please do not create a ripoff version of this game to promote some product. In a month, there will be thousands of these and not one of them will be one percent as popular as Pokemon Go. Just don’t. Just don’t do it. What you can take away from this is that IF you have a strong brand with a narrative structure and IF you partner with one of the best augmented reality companies on earth, you can create something that changes the culture. That is the power of augmented reality and it is only going to get stronger. Next Steps: Magic Leap. Read More