Confirmation Bias
= Search for and interpret information in such a way that confirms pre-existing beliefs
Example:
Have a higher trust in an algorithm if the result confirms a belief or assuming bugs when the algorithm provides results that are against my belief
Anchoring Bias
Relying too heavily on the first piece of information that is given
Example: Benchmarking of algorithms - 70% accuracy seems high when you came from 30%
Unconscious Bias
when a Behavior shows bias, but they are not conducted with intent
I believe that people from all areas are equally capable” -> Data also shows that this person is
not hiring people from area Offenbach
In Data Science: Training a machine learning algorithm that learns biases from humans
Priming Bias
Priming bias is a subconscious mental shortcut where exposure to one stimulus influences how you respond to a subsequent stimulus.
Think of an word that starts wie S. When previous talking surrounded around astronomy you would most likely think of SUN, if we talked about cooking you would most likey think of SALT
In Data Science: Priming can be used to provide useful context, eg. with previous accuracy results. It can also be used harmful when when you setting wrong expectations.
Hindsight Bias
when you perceive past events as having been more predictable than they actually were
Example: stock market
When you in hindsight believe you knew that a certain stock value would go up and that you should have bought it
In data science: Data leakage -> accidentally including the target parameter in the list of input parameters
Self-Serving Bias
Status Quo Bias
= the preference for the current state
Example: Regret avoidance (because of loss aversion) -> When you are at a restaurant, considering new dishes but still end up ordering the same one as always
In data science: “Knowing” that a tool A is good to solve a task X and not being open for innovative new tools
Framing Bias
Automation Bias
Belief Bias
Ikea Bias
Information Bias
Last changed4 days ago