Google Trips on Attribution The Electronic Privacy Information Center (EPIC) has filed a legal complaint against Google based on their Store Sales Management program that attempts to track consumer purchasing behavior in the real world to see how it correlates with online searches. According to EPIC, the system used to anonymize user data relies on insecure technology and it is difficult for users to know how to opt out of this type of tracking. This is because Google uses location tracking as a proxy for purchase behavior. If I Google “smoothies near me” and then go to a smoothie store, Google attributes a real world sale to the online search. EPIC claims that Google could potentially purchase third party data on credit card purchases, further endangering consumer privacy. Why does this matter? Attribution has always been marketing’s biggest challenge. If I can demonstrate that advertising on my media property leads directly to sales, I can sell more advertising at a much higher price. Since Google dominates non-social digital advertising through search and display, it is understandable that they would want to nail down the attribution piece. But, honestly, it’s a mistake. First, Google’s dominance is due to the fact that their attribution on search and display is already superior to everyone else. You can run ads on Google and only pay for purchases completed after someone clicks on the ad. But consumers understand that transaction. It’s baked into the system. Tracking people as they walk around is a potential public relations disaster with little upside. It is doubtful that the slight improvement in attribution will drive that much more business. Advertisers already figure assumed attribution in the amount they are willing to pay. Second, Google doesn’t need any more privacy headaches, particularly in the EU. In a nutshell: Google wants their ad system to be comprehensive. Consumers probably don’t. Read More My Chatbot speaks French Researchers at Salesforce have discovered that they can improve the comprehension of their chatbots by first training the underlying machine learning systems to translate between two languages. Although translation is not the point of the chatbot, the act of learning two different languages and producing understandable translations between them teaches the system how language works in a way that simply training the system to fulfill its primary function does not. This confirms other research that shows that machine learning systems benefit greatly from learning things unrelated to their primary function. The benefits of multilingualism would come as no surprise to linguists and cognitive scientists. Humans who are polyglots (speak multiple languages) tend to score better on standardized tests, they are better at remembering lists or sequences and they are better at avoiding mis-spellings or simple grammatical mistakes. Whether you are a person or a chatbot, learning other languages improves performance. Why does this matter? Machine learning requires us to rethink seventy years of programming behavior. Traditionally, systems have been built to accomplish a single task. Programmers evaluated their work based upon how simple their code could be and still accomplish that task. Computers were (and are) thoughtless, so the programmer needed to account for every eventuality. Human-built systems were ideally elegant, focused, and comprehensive. Machine learning works differently. Systems built through machine learning resemble human children in that they benefit from experience, exploration and unfocused learning. Teach a machine learning system to recognize photographs and it will eventually be better at recognizing sounds than a system trained on sounds alone. Learning is cumulative. Perhaps in the future, machine learning systems will come pre-educated, having been trained on multiple, unrelated tasks before being trained on their eventual application. In a nutshell: Chatbots. They grow up so fast... Read More RAND in the Valley The RAND Corporation (standing for Research ANd Development) has always had a vaguely “New World Order” feel to it. Operating as a non profit entity, it gets its funding from various government agencies, companies and individuals. But it maintains a kind of willful opacity about its operations and interests. It was often at the forefront of technology during the Cold War when it’s research touched on areas of espionage and code breaking. Having recently opened an office in Silicon Valley, The RAND Corporation is careful to emphasize that it’s primary interest is in secondary effects of technology on society. This focus is understandable. The technological upheavals of the past twenty years have begun to shape society in unexpected and not altogether positive ways. While every society has its “haves” and “have-nots,” a society in which knowledge and computing power are unevenly distributed risks permanent, inflexible stratification. Of course, while they’re here, The RAND Corporation is also looking into a few other things for their friends in DC. Why does this matter? There is a scene in Good Will Hunting when the hero is interviewing with the NSA. The man from the NSA tries to sell Will on the job by saying that he will get access to the best technology before it is available to the general public. At the time, that was realistic. Not today. The big technology companies have innovation at least equal to that of the U.S. Government and, in some cases, their technology is clearly superior. As the intelligence world struggles to adapt to cyber security threats, deep learning and quantum code breaking, they will look to make public/private alliances with technology companies. Reading between the lines, the RAND Corporation is expanding into Silicon Valley in order to facilitate those partnerships because that is what they are set up to do. I have no doubt that they are seriously studying secondary effects of technology on society. But I would be surprised if they were not also exploring how technology companies can be drawn into a partly-privatized intelligence network. In a nutshell: The RAND Corporation is never doing just one thing. Read More