First, Second and Third-Order Inferences

Privacy Plus+:

Privacy, Technology and Perspective

First, Second and Third-Order Inferences. Start with the facts:

• the facts of what a Human Being browses, buys, throws away, pauses to look at, and throws away;

• where they go, when, how, for how long, and where they go next and after that;

• maybe even what emotions they feel in each stage of each transaction (which 3-D facial-recognition software is increasingly able to turn from a parlor trick into a serious capability).

With these facts or “data points,” you have the raw materials from which to predict, or infer, what your Human Being is like and will like, will support or oppose, will vote for or against, will buy or turn away. Presumably, the more data points you are able to collect and add to the mix – not only about your Human Being, but also about the behavior of as many other people under as similar circumstances as you can gather – the more accurate your prediction/inference will be, and the more valuable it will be.

Let’s call what you’re creating a “First-Order Inference,” since you’re starting with the raw materials – the facts / data points – and drawing inferences directly from them. You know that “inferences” are expressly included within the definition of Personal Information under the California Consumer Privacy Act (CCPA) and other state privacy acts, so it would certainly seem that those privacy laws apply to the First-Order Inferences that you are creating. So, perhaps muttering and grumbling, you comply.

From this point, however, it gets harder.

Suppose you wish to draw a further inference, starting from the First-Order Inferences that you already have drawn. Now that you have concluded, inferred, profiled, what your Human Being will likely enjoy, buy, support, or do, you want to infer what your Human Being will likely do next. Think of your answer to this as your Second-Order Inference, an inference drawn from the First (perhaps with some extra facts / data points or interpretation thrown in). From it you may draw a Third, and on and on.

For example, suppose you own a casino in Vegas. You collect (properly of course) a Human Being’s browsing history, purchases, and maybe records of TV shows, and deduce from those facts / data points that they might enjoy a trip to Vegas. That is a First-Order Inference – and monetizable, too, since well-directed ads and perhaps a coupon or two might well entice them out there.

Now suppose your Human Being actually comes into your casino because your First-Order Inference, through targeted advertising, got them there; but now you want to build on that. You already do everything imaginable to get customers to bet more and more – pumping oxygen into the casino, adding bright lights, stimulating sounds, and taking out the clocks – and every inch of your casino is already covered with fraud-detection cameras. But suppose you attach your cameras to sophisticated 3-D facial-recognition software that can read emotions? Could your software use inferences drawn from your Human Being’s evolving emotions to signal dealers when to pour it on, or when to back off a little and suggest the Human Being take a short break? (For that matter, with the right databases and algorithms, could it also read your Human Being’s “tells” and signal them to dealers at the tables, thus increasing the house odds even more?) These would be “Second-Order Inferences,” drawn from the First-Order ones you used to attract your Human Being into your casino in the first place.

You will know when your Human Being’s had enough, because they will leave. But you would like them to come back, so you want to know: Why did they leave? Was it because the convention was over, or was the experience ruinous? This might be a Third-Order Inference. And suppose you discern – from your processing of properly-collected browsing and purchase histories after the trip – that your Human Being would benefit either from for-profit lessons on how to play, or encouragement on how to reach the next level in their brand new vice!, or from counseling? Might this be a Fourth-Order Inference, drawn from the First-, Second-, and Third-?

This hypothetical is limited, because it involves only one actor (you) and one data subject (your Human Being). It also follows a straight, linear progression as you act to influence your Human Being’s life, from a state of Vegas-innocence to the aftermath. The inferences all build on one another, as predictions turn into facts, and “will like” becomes “did act in the following way.” Presumably the CCPA and its progeny would extend to all these later-order inferences you are drawing, at least as long as they are identifiable to a particular household or individual.

But a third-party market might develop in later-order inferences. (We would be surprised if it didn’t.) Tutoring classes, financing offers, travel agencies, even counseling services could be interested. Let’s work on drawing our own inferences about how privacy laws would apply then…