This post is a follow-up to one of my talking points in this week’s keynote at the 2013 Adult Student Recruiting and Retention Conference
We are fast approaching the top of the hype cycle when discussing learning analytics and big data. The use of data to improve student outcomes is not a new phenomenon but, like in other areas, the promise of what is now possible is staggering. The operative word in that last sentence is “promise.” While analytics are rapidly changing the business world, systematic use of student data in education is still in its infancy. (note: I recommend reading Michael Feldstein’s recent post A Taxonomy of Adaptive Analytics Strategies for more)
If you use Amazon to shop, you have likely seen the recommendations made just for you. Many times, these make complete sense, like when I am told that author William Gibson has a new book out that I might want to try. But occasionally I am shown a very odd item. Amazon is, as most of us know by now, mining data across millions of users and billions of transactions to look for patterns. Sometimes, these patterns are not evident based on the purchasing behavior alone. Occasionally, these items are laughable and we wonder how Amazon could make such a ridiculous recommendation. Other times, however, it is eery how I am shown something that does not connect directly to any of my past shopping history that is relevant to me. While sometimes helpful, it is also just a little creepy.
One recent story about Target’s use of customer data has gone viral and is now quite well-known. This gist of the story is that Target began sending coupons to a teenage girl who, according to her shopping habits, appeared to be pregnant. The girl’s father was outraged at the store only to learn later that his daughter was, in fact, pregnant (Forbes version of this story). Subsequent stories explored the extent of data retailers now collect about ever aspect of their customer’s lives. Much of this data is freely given through use of loyalty card programs and credit card privacy agreements but this data is then cross-referenced by buying and selling through data brokers (see FTC takes aim at data brokers for a recent story). Most of us are likely unaware of the power of data brokers and how information we assume is in a silo is actually being resold and then cross-referenced, compiled, bundled for marketers. Target, like many other retailers takes advantage of these types of services to learn more about customers and provide specialized offers. This is not necessarily a bad thing for consumers but might make us uncomfortable when we realize this is happening without our knowledge (ever wonder why you might get a discount on gas with your grocery loyalty card?).
The mustard problem
Despite the amount of data Target has about me, my family, and my shopping habits, they (and other retailers) still fail in a very important way. At the moment of highest need, they cannot help me make a decision. When I am standing in front of 40 different kinds of mustard, there is no accurate guide to help me figure out what I really want. Of course, in today’s mega stores, it is not just mustard but every product seems to have a dizzying array of choices. While I might get a coupon to persuade me to make a choice, that is about the retailer and the product manufacturer trying to influence me to maximize profit and not help me match my taste bud’s desire at that moment.
In the 1997 journal article “Artificial Tutoring Systems: What Computers Can and Can’t Know” Frick outlines this exact problem. Computers are good at knowing how to do something (process information in various ways) but not good at establishing meaning. While computers might house a lot of data about me to the point of being “creepy,” they need an incredible amount of contextual and social data at the moment of need to provide accurate guidance to me about what mustard I want to purchase. A lot of progress has been made since 1997 but the same basic problems remain.
Now think of the almost endless amount of digital education content (and analog instructional options) and it makes 40 types of mustard seem like a trivial problem. We are not even doing a good job helping students personalize their learning needs with static web content. As the wealth of new instructional opportunities increase, the need to help people discover what works best for them will also increase.
While Target might know a lot about us, our schools and universities have equally vast repositories of data. Demographics, income, login patterns, healthcare issues (in the case of on-campus clinics), email volume, contact lists, web surfing and internet use (when on campus), and countless other data points are available. Note that I did not even touch on what is available through the LMS, which is interesting but only a partial snapshot of your data profile. Universities have many rules in place to isolate and segregate this data. FERPA and other laws require educational institutions to develop and follow strict data privacy standards. Our scruples and legal requirements make it hard for us to fully take advantage of the vast troves of data retailers are mining but there have to be steps in between where we are today and the “shady” back alleys of consumer data brokers.
We might not yet be at the point technologically to make perfect recommendations but we are getting closer. Taking the next steps, however, might mean using data in unusual and even “creepy” ways to determine what really works.
Frick, T. W. (1997). Artificial tutoring systems: What computers can and can’t know. Educational Computing Research, 16(2), 107–124.
Hill, K. (2012). How Target figured out a teen girl was pregnant before her father did. Forbes. Retrieved from http://www.forbes.com/sites/kashmirhill/2012/02/16/how-target-figured-out-a-teen-girl-was-pregnant-before-her-father-did/