Woof-woof dog Imagine a young child who had never seen a dog before. Show the child a picture of a dog and, in all likelihood, she will recognized dogs in the future. That is the unparalleled power of human learning – a single exposure can lead to permanent recognition. Machine learning systems have always been different. To get a machine learning system to recognize a dog, the system must be fed vast amounts of labelled data. (In this case, pictures of dogs labelled “dog” and pictures of not-dogs labelled “not dog.” After processing this data, the system will begin to be able to recognize pictures of dogs with almost-human accuracy. However, Google DeepMInd has perhaps found a way around this. By attaching a memory component to a deep learning system composed of a large number of neural nets, they have demonstrated machine learning capable of recognizing things after a single exposure. Application to Marketing: As someone who depends on the financial well-being of the digital advertising ecosystem, I am concerned about bot traffic and ad impressions. Digital advertising has a major problem with phony traffic and our attempts to identify and eliminate bot traffic have failed miserably. However, when I look over the Google Analytics for one of my clients’ sites, it is easy to identify fake traffic based on geography, behavior, source, and other factors. A human being can recognize which visitors are likely bots. If a machine learning system were capable of quickly learning to identify bot traffic and perform these operations on an industrial scale, I believe we could salvage the digital ad marketplace and return value to advertisers and publishers. Next Steps: It’s always worthwhile to keep abreast of DeepMind. Those guys are wicked smart. Read More Chromeland Security Google needs the internet to be secure. Unlike Apple or Amazon or Facebook, Google makes money when people feel safe using the open internet. However, almost half of the internet currently offers no encryption, even for some ecommerce sites. Google’s Chrome Security Team has been working to get more sites to switch from the relatively unsecured HTTP protocol to the encrypted HTTPS protocol. This is in advance of an announcement that Chrome will start attaching a warning to unsecured sites in 2017. The security team at Google has observed that the icons they have used for secure vs. unsecured sites (a little lock in the URL window of the browser) didn’t have much meaning to many of their users. And comprehension fell off even further in other countries. So the Chrome team has started to alert users more aggressively about unsecured or compromised sites, earning them the ire of small and large businesses that depend on site-generated income. Application to Marketing: Go to your website (or your client’s website). Does the URL start with HTTPS? Congratulations! You’re ahead of the game. If the site says HTTP, I’ve got some bad news for you. You’re running out of time to get that updated. And, contrary to what you may have heard, it’s not as simple as just getting a certificate. You need to make sure any third party data or plugins you use are also encrypted. For most sites nowadays, that’s a real challenge. This change is every bit as important as the switch from standard definition to HD. Don’t be caught unaware. Next Steps: Update your site. Read More Computers, who needs them? Benedict Evans, venture capitalist and astute observer of internet-y stuff, suggests that the time may have passed where we should be thinking mobile-first. Perhaps, now we should be thinking about mobile-only. The proliferation of smartphones has created a mobile internet that is more extensive, richer, and potentially more immersive than the desktop experience. So why bother with the desktop? The desktop does not feature an excellent image input mechanism, a motion detector or a GPS. Evans suggests that the desktop is holding back the internet from reaching its potential. And, while I respect Evans’ perspective, I am only partially convinced. I appreciate that digital natives are more adept with the mobile keyboard than yours truly. But I cannot be alone in believing that while mobile is an excellent place to consume the internet, it is a flawed input mechanism. Application to Marketing: Evans is correct that, for many companies, there is no need to create a desktop version of their website. Since I spend a lot of time kicking around analytics for my clients’ websites, I have noticed that some of them barely have any desktop traffic. The desktop site seems to exist largely for an internal audience (typically the CEO), who want to be able to see a pretty site on a large screen. So I agree that mobile-only may make sense for some companies. However, mobile sites are necessarily less decorative than desktop sites because there is less of a canvas to work with. If you’re going to truly commit to mobile, you need to start asking yourself what you’re willing to take away. Next Steps: Check to see how much of your traffic is coming through mobile. If you’re getting into to 70+% range, it might be time to think mobile-only. Read More Racist Robots Machine learning is an important advancement in computing. Image and pattern recognition and natural language processing have undergone rapid evolution due to machine learning. But machine learning has historically required large, labelled data sets in order to “train” a system so it can perform its function. In an ideal world, data sets would be objective and unbiased. In the real world, not so much. When beauty.ai held the first beauty contest judged entirely by machine learning this year, the vast majority of the 44 winners were white, a few were Asian, and only one had darker skin tones. It turns out the data set of women’s photos that trained the beauty.ai system was disproportionately white. Machine learning is not objective, it can be manipulated by its inputs and its presets. If you train a machine learning system to recognize criminals based on a prison population that is disproportionately minority, it will consider minorities to be more likely to commit a crime. Technology can be powerful, but data is the fuel of technology. Application to Marketing: In the past few years we have seen the advertising industry begin to critically examine a racist and sexist industry culture. I would prefer to believe that no one wakes up in the morning and looks forward to being racist or sexist that day. The easy and obvious examples of racism or sexism can be identified and then addressed. But these events, while they attract the lion’s share of press, may blind us to the subtle bias that pervades our assumptions and shape our data sets. Do not assume that the data you have collected is objective. It has been shaped by the assumptions of the (no doubt well meaning) people who may not be aware of their own biases. Next Steps: In the next few years, machine learning companies will approach you offering to analyze your data. Your first step should be to critically examine the assumptions your data is making. Read More