Feedback and Explanation 1.0

Blocking: The concept derived from Pavlovian conditioning that associations or learning attributed to a stimulus will not occur if those associations are redundant or superfluous. For example, a lab animal may learn that a red light signals food. If a green light follows and just as reliably predicts food, the animal will not learn to associate green with the food, since prior learning 'blocks' the association. Blocking should not be confused with blockheadedness, which is characterized by an inability to learn new and better explanations to an event once the first explanation is fixed in your mind.

Explanation is critical, for if you only go by the correlations of nature, your predictions can go seriously awry. Oftentimes those correlations work consistently, and for our practical affairs universally (though not perhaps it may be added for the universe). Throw a ball into the air, and Newtonian mechanics will predict where it will land. Of course, Newton’s laws break down when you are considering the very tiny (Quantum physics) or the very large (General Relativity), but the Newtonian inductive (i.e. consistent un-falsifiable correlations prove the rule, as compared to the deductive approach that uses the rule to predict and falsify correlations) approach is a reliable solution to our practical problems, even though it is irrelevant to our cosmic ones.

When we get down to human nature however, explanation seems to be on the wane.  For our practical affairs, it is now the correlations that count. They are easy and cheap to derive, and with the advent of data mining, we can find correlations upon correlations that would make even Newton blush. Now even without Newtonian equations, behavior can be predicted with pinpoint accuracy through the correlations found through the brute force of our computing power. With predictions like this, who needs explanations!

And so we don’t.

This mindset is characterized by net denizens and wizards (Isaac Newton, who considered himself an alchemist first, was also a wizard), who have every confidence in their predictions, and have the gathered eyeballs and mega bank accounts to prove it. To illustrate this mindset in action, consider this recent article in by Thomas Goetz in Wired Magazine on ‘Feedback Loops’.  Getting feedback not only informs, but it motivates, and getting prompt and regular feedback can get people focused, motivated, and aroused to do what they need to do. This is a simple and reliable premise vouched from not only all human experience but all recent iterations of human experience. The internet in particular provides us with not only information, but also feedback as to the state of our behavior. Harness that power, and you can harness human motivation, presumably of course for the good. All well and good, except that there are negative correlations within the positive correlations that a data miner may overlook but a good theory or explanation predicts.

Life is full of traffic signals

Consider that blinking road sign up ahead that gives you your speed. The information is redundant, as you know your speed from a quick glance at your speedometer. Nonetheless, as the data show, this feedback motivates you to slow down, and even after the sign passes keep slowing down. However, as Goetz claims, this is due to the fact that ‘people are reminded of the downside of speeding, including traffic tickets and the risk of accident’ (as if the speedometer doesn’t!).  So whether information feedback is redundant or non-redundant, feedback works. The implicit correlation and thus prediction nested in Goetz’s article is that non-redundant and redundant feedback have a sort of equivalency. The fact is though, they don’t. Humans and indeed almost all sentient creatures do not tolerate information redundancy, indeed they don’t have the time or computational space for it. In fact, redundant information is automatically blocked out through a well known process aptly named ‘blocking’. As an illustration, consider another road sign example. Suppose you see a traffic light turn red, and then ten seconds later a second brown light also turns on. Both red and brown light correlate with ‘stopping’, but only the significance of the red light will be remembered. The information from the other light is redundant, and is therefore blocked. So when the light turns brown, you will not stop because your brain blocked you from considering it. This blocking phenomenon holds for all creatures and all events, and forces another explanation for Goetz’s data, namely the fact that people may be slowing down because they perceive that the blinking road sign does not just give information you already know, but information you don’t, namely the greater likelihood that there is a cop around the corner. I may be wrong here, but that is a good thing, because as with all predictions coming from good explanations, this premise is imminently testable. For example, put that feedback on your speed on every billboard you pass, and see what happens!

But there is another prediction that comes from explanation, namely that novel or unpredictable events provide an incentive salience or importance to moment to moment behavior that depends upon how information is arranged, or to point, its feedback function. The neural basis and explanation of feedback is that we are responsive not only to the ends of our behavior but to the novel twists and turns that get us there.  In other words, performance feedback works because it activates mid-brain dopamine systems that are sensitive to the novelty that is implicit in non-redundant feedback. But dopamine is not activated by redundant information, only novel info will do. Hence the motivating power of redundant information if refuted yet again, but this time from predictions derived from how explanations of how the brain actually works. In other words, it ain’t loopy feedback loops’, but the novelty that counts, or in the large the explanation that counts.