Predicting Dependent Variables

Consider a list of metrics. Now pick the Y, the dependent variable, from that list.

A human would use their judgement.

A machine would use an algorithm.

It’s clear that human judgement varies. There’s evidence that it does.

Causing a machine to make accurate predictions about human judgement is an interesting problem because of the inherent variation of judgement within human populations.

This is a polite way of saying that some people really diverge from the median in their application of judgement.

Consider the following question:

If given a table of Gross Profit, Sales, COGS, Marketing Spend, Working Dollars, Non-Working dollars, Discount Cost, Impressions, Paid Media Impressions, Paid Media CTR, UV, V, and PV, which is the dependent variable?

The answer is pretty obvious to a single observer.

Every person reading that question believes that the answer is obvious. At least five readers will state that I forgot expected LTV (I did not. I just left it out.)

There isn’t going to be unanimous agreement. People are different. And more importantly, the context is going to be different, depending on their own view, and the context of the problem they’re solving.

People are bias.

When people together to agree on specific language, for instance the difference between an amplified engagement and a consumptive engagement, it enables progress. That language is representative of a paradigm. A paradigm means a set of standards, language, definitions, biases, and beliefs about how nature works. Paradigms are really important for progress.

Machines contain the paradigm of their creator.

(And that even goes for unsupervised learning. Even the choice to do unsupervised learning contains a paradigm.)

A good meta-optimization objective is to build a machine that caters to a clear paradigm. If there are those that don’t share that perspective, they are free to disagree.

We do need to progress. And that means making the best predictions about the Y variable that we can within the constraints of a defined paradigm.

I don’t think we should wait.