Personalization and the Outliers

Henry Ford was an early proponent of personalization. "Any customer can have a car pained any color that he wants," said Ford in 1909, "so long as it is black."

There have always been limits to personalization. I like to wear hats, but my head is some sort of extra-large melon, so while hat manufacturers may offer choices to fit the personal size preferences of many customers, I'm an outlier. Many times I'm just SOL on a particular hat.

Imagine that all your potential customers are like a big bell curve. Aim too narrow, meeting the personal preferences of the fat part at the peak, and you will not capture as much of the market as you'd like. Or if you're up for it, imagine a line plot-- the sweet spot is in covering the cluster at the center while ignoring the outliers who are off in their own little corner. You can predict what the majority of your customers probably want-- to be sure of hitting the outliers, you would have to cover the entire charter, including areas that were completely empty. Spread out to offer every option that every customer on the diagram wants, and on the thin ends you'll be losing money. Burger King could offer pineapple and bananas as burger toppings, and somebody might even order it, but they would never sell enough to cover the cost of stocking those ingredients. Maybe nobody would ever choose banana at all, and BK would be out the money. They could advertise they were offering perfectly personalized burgers, but they'd be losing money to do it.

Personalization as a marketing strategy has to aim for the sweet spot where you are hitting maximum customers for minimum. You can only do that by recognizing that some customers have personal preferences that it will cost you too much to meet, and so those outliers must be cut loose. The only choice they get is "Buy what we offer, or do without."

Any system, any product that is going to marketed to any kind of scale, is going to have this built-in limitation.

And that includes personalized education.

Putting together banks of questions and drills and instructional activities is expensive for companies-- remember how aggressively the major test manufacturers guard their question, because it would be prohibitively expensive to have to replace a set of 50 or 70 questions. So creating the giant bank of possible modules for personalized learning programs would presumably be hugely expensive. Therefor, the module manufacturers will need to control costs and maximize return by ignoring the outliers.

There's no money to be made in creating a module bank that meets every conceivable educational need that could appear, particularly if we are talking about a program meant to serve hundreds of thousands of students.

Likewise, the artificial intelligence that is allegedly sorting out the students based on their performance cannot afford to be too precise in its sorting. If it analyzes student achievement and ability based on dozens of factors, then it will need to sort the students into hundreds of separate "bins," each with its own set of instructional modules, each of which will in turn sort students into another hundred separate bins. This system would be complicated and, more importantly, expensive as hell.

Instead, it would be easier and cheaper to collect only a couple of data points and sort the students into a half-dozen large bins that encompass a broad spectrum of students. Details of their personal educational needs will be discarded, and outliers will just be jammed into whatever bin they come "close enough" to matching.

In short, a profitable system will be no more personalized than giving all students a ten-question pre-test, then sorting-- students who get an A or B move on to Worksheet #1, students who get a C move on to Worksheet #2, and Ds and Fs move on to Worksheet #3.

And this is before we even start to work in other factors like the student's interests (a critical factor in choosing reading assignments) or the style of exercise they work best with (picture-based? long story problems? puzzles?).

To be marketable, personalized education systems have to promise that they will provide an educational program perfectly suited to each and every child. But to be manageable and profitable, they have to provide a system that discards outliers among students and just jams them in with everyone else.

You know what's good at personalized learning for outliers? Carbon based life-forms animated with non-artificial intelligence, with professional training and experience in the education of young humans. Collect a bunch of those, give them several individual small humans to instruct, and you can have all the personalized instruction you want-- and it won't even create a permanent data file. Best of all-- these carbon-based life forms can even provide personalized instruction for the outliers.