The Rigor of Data In “The Rigor of Science” Argentine writer Jorge Luis Borges imagined a land in which the science of cartography advanced to such a state of exactitude that only a map made to the same scale as the land itself would suffice. This is a charming notion – a 1:1 map that overlies the country it represents like a blanket. Reading an article from the Deloitte University Press entitled: “Dark analytics: Illuminating opportunities hidden within unstructured data” (link below), I was reminded of Borges’ map. The Deloitte article suggests that recent advancements in computer vision, pattern recognition and cognitive analytics will allow businesses to spot patterns in the vast quantity of data they collect. The article’s authors suggest it is exactly the endless amount of unstructured data collected by video cameras, text messages, apps, websites and, in the future, IoT devices that will unlock hidden patterns of consumer behavior and profit. They suggest that a firm capable of deciphering these “deep web signals” from the noise of unstructured data will necessarily have a competitive advantage. I assume that the author’s intention is to reassure senior executives that their endless collection of consumer behavior will eventually yield a return on investment, even if the data now sits idle and unexamined in cloud storage. Why does this matter? Homo Sapiens have evolved to look for and recognize patterns. We process the world using the assumption that there are deeper patterns just beyond our perception and, were we just to grasp the right data points, the deeper patterns of messy reality would be laid bare. Borges’s story about the map suggests to me that there is a folly in trying to know too much. When we try to know the world with total exactitude, we become overwhelmed with data points. We are not extracting signals from the messy noise of reality. We are simply recreating the noise. In terms of data collection, companies today exceed Borges’ map. A hapless consumer buying a phone charger is subjected to a totalizing surveillance that would make the NSA blush, all in the hopes that he or she might tip his or her hand and reveal the underlying pattern that caused them to buy this phone charger from this company at this time. Small matter that we might simply ask a representative sample of customers to share their thoughts. This approach is dismissed as hopelessly naive. We assume that the pattern is there, that it is hidden and that it can be pried loose by a neural network of sufficient complexity. If Borges were writing today, he might imagine a land covered with server farms striving to collect consumer data on a now-displaced population. In a nutshell: Too much data simply reproduces the noise of reality. Read More See the pretty girl in that mirror there. There are a number of theories about what makes a person physically attractive. Our mothers might suggest to us that physical attractiveness is the product of an inner strength – confidence, a sense of self. Some scientists have suggested that attractiveness is reducible to an algorithm – certain proportions between eye, nose and lip, reflective of genetic health and fertility. Beauty.AI is a company that aims to crack the mystery of attractiveness through the latest machine learning technology. The company became famous (or infamous, if you prefer) for organizing online beauty contests to be judged by an algorithm that they had trained using celebrity faces, among others. The contests became controversial when the “winners” were found to be overwhelmingly young and white, revealing an underlying bias in the algorithm. Now Beauty.AI has produced a new product called Map My Beauty that claims to use facial zone recognition algorithms to objectively assess attractiveness. Map My Beauty makes heavy use of a new Beauty.AI algorithm called RYNKL which judges the extent and severity of wrinkles on various parts of the app user’s face. I will allow Beauty.AI scientist Alex Zhavoronkov to explain the approach: “The fewer wrinkles you have for your age, the more beautiful you are.” Well, that’s settled then. Why does this matter? Beauty isn’t a thing. If it was, we would see total unanimity of judgement through different times and different cultures about what constitutes beauty. So, to make the obvious (and well-trodden) counter-arguments: The ancient Aztecs thought crossed eyes were the height of beauty. The Kayan people of Myanmar artificially elongate their necks with rings because a long neck is considered more beautiful. During the Song dynasty, foot-binding was imposed on children because little feet were considered attractive in women. And even in the West, our shared conception of beauty has ranged from the voluptuousness of the Rubens model to heroin-chic. The problem with suggesting that beauty may be measured by an algorithm is that it becomes a self-fulfilling prophecy. The poor, benighted teenager who subjects their face to Map My Beauty will internalize this supposedly objective result. Confidence will be reinforced or crushed depending on the false-objectivity of demonstrably flawed algorithms. Not to belabor the point, but machine learning is only as good as the data used to train it. When you have a cultural construct like “beauty,” the labels you attach to your data should reflect cultural understandings that vary from place to place and through time. But Beauty.AI’s algorithms might generously be described as culturally constrained and, in any case, retroactive. The bias is not a flaw of the algorithm, it’s the whole point. Because beauty is subjective. In a nutshell: Algorithms applied to subjective value-judgements do not become objective. Read More Designers who code, coders who design John Maeda has enjoyed a long and successful career at the intersection of technology and design. So his words tend to carry weight in both design and technology. Recently, Maeda has been emphasizing the need for designers to learn to code. He makes the argument that these type of multi-skilled individuals tend to be present in highly innovative companies. While I personally would love to hire a first rate designer who was also a first rate developer, my own experience is that these people are rarer than unicorns. (The mythical beasts, not the over-valued startups.) While I have employed many designers with some coding skills, their code compares unfavorably with the work of a professional developer. I have also employed developers with design skills. But if you compare those design skills with that of a polished designer coming out of one of the top design colleges, well, it’s not the same. While I agree that a designer should know enough about code to have an intelligent conversation with a developer, I think it is deeply unrealistic to imagine that you can combine these roles. The result, I’m afraid, will either be sloppy code or bad design. Perhaps both. Why does this matter? Focused expertise is out of fashion. Everyone is supposed to have hybrid skillsets. Unfortunately, a great developer is (in my experience) a great developer. Encouraging designers to code is a great way to mess up your frontend code. And having someone with Maeda’s scope and influence advocate for mixing these two, radically different skillsets is going to lead a lot of startups down the primrose path. It would be great if designers could code and handle the accounting and wash the windows and be a certified masseuse. But good designers design. That’s more than enough. In a nutshell: Ask your designer to code and you’ll probably get bad code. Read More