Unit 1 - Problem solving - Algorithms

Cards (9)

  • name 5 examples of causes algorithmic bias.
    One example is malicious data manipulation
    Another is Data reflecting pre-existing biases
    Another is unbalanced classes in the training data
    Another is inaccurate data
    Another is data amplifying bias through a feedback loop
  • Malicious data manipulation has been found especially through coordinated troll attacks such as that of tay ai on twitter
  • Data can reflect existing biases such as more images being of white people as there is more white people and therefore the image sets that AI image generating models use, bias the models towards generating white individuals
  • Unbalanced classes is where data reflects existing biases such as more images being of white people as there is more white people and therefore the image sets that AI image generating models use, bias the models towards generating white individuals
  • All algorithms are biased
  • Some datasets are biased as a result of human bias such as certain genders being overrepresented in a societies view of a certain career. Such as nurses being mainly female but media also representing and amplifying this leading to datasets based off internet data assuming that a nurse would be a female
  • Correlation is not causation, however algorithms fail to differentiate and therefore any pattern is as viable as any other. Therefore a human user may be needed to check the data.
  • Data can be amplified through a feedback loop such as more ice creams being sold in an area therefore more ice cream vans are put in that area, leading to higher ice cream sales in that area. This occurs when a dataset is able to feed back in on itself, and also doesn't consider the causes of said change
  • Inaccurate data is a simple form of bias and is where incorrect data such as data that says that if ice cream is 10% off only 2% more people would buy it. However the data is incorrect as a factor of 10 error occurred when calculating the percentage and in fact 20 percent more people had bought ice cream