Section Eight - Feedback loops
A story about Michael Jordan, Sam Bowie & Frenkie de Jong (2 minutes)
Suppose you write a well-thought-out-algorithm. You were also very aware of the issues with the underlying data. Now another question arises: did you also think about closing the feedback loop?
Basketball & soccer
A good data model has feedback.
Cathy O’Neil gives the following example: Imagine you have to select a new basketball player from the pool of college players. The LA Lakers, a professional team, do not select a point guard, because he does not have enough assists. The point guard is selected by the Utah Jazz and there, in his first season, he has a lot of assists. The result will be that the LA Lakers revisit their models and figure out what else they need to look at to make better predictions.
Feedback makes the model better.
The same goes for Jurgen Streppel, who as head coach of Willem II decided to part with Frenkie de Jong. Not enough talent. Mistake of course, but thanks to feedback Streppel was able to learn from it. In this way, Frenkie not only makes Barcelona better, but also Jurgen Streppel. The same goes for Jack Ramsay who, in 1984, selected Sam Bowie over Michael Jordan.
Sam who? Right!
McDonalds & Fontys University
Now suppose someone applies for a job at McDonalds.
For this he (or she) has to fill out a number of personality questions, play an online game and do an interview in which his micro-expressions are read by a computer. Based on this, a computer automatically decides not to select this person. He then goes to work at the Burger King and there he becomes a franchise manager in no time. The McDonalds system will never know. It can only improve on analyzing people it hired. Not on the applicants that were denied. So, the system remains dumb.
There is no closed feedback loop.
The same goes for our students. If they graduate we try to 'follow' them. Do they get a good job? Do they earn a lot of money? Do they make the world better? However, the moment they drop out they are off our radar. Perhaps the fastest way to success is to drop out From Fontys University.
Who knows?
Quick question: do you know why most systems have no closed feedback loop?
Quick answer: The reason is simple. Most automatic systems are not about people that are worth millions like professional athletes. Most systems are there to separate the herd. In the hypothetical McDonalds example, feedback is unimportant and it does not matter whether there is collateral damage. There are enough people to be careless. Talented staff that is not hired are a calculated risk. Uninteresting. Not important.
Unless you’re one of those people that were denied, of course.
Take aways from section eight:
- A good datamodel has a closed feedback loop (or is aware that the loop is not closed);
- Most systems do not have such a loop because they can afford not to.