James Arvantitakis, Andrew Francis, and Oliver Obst
Originally posted June 21, 2018
If the recent Cambridge Analytica data scandal has taught us anything, it’s that the ethical cultures of our largest tech firms need tougher scrutiny.
But moral questions about what data should be collected and how it should be used are only the beginning. They raise broader questions about who gets to make those decisions in the first place.
We currently have a system in which power over the judicious and ethical use of data is overwhelmingly concentrated among white men. Research shows that the unconscious biases that emerge from a person’s upbringing and experiences can be baked into technology, resulting in negative consequences for minority groups.
People noticed that Google Translate showed a tendency to assign feminine gender pronouns to certain jobs and masculine pronouns to others – “she is a babysitter” or “he is a doctor” – in a manner that reeked of sexism. Google Translate bases its decision about which gender to assign to a particular job on the training data it learns from. In this case, it’s picking up the gender bias that already exists in the world and feeding it back to us.
If we want to ensure that algorithms don’t perpetuate and reinforce existing biases, we need to be careful about the data we use to train algorithms. But if we hold the view that women are more likely to be babysitters and men are more likely to be doctors, then we might not even notice – and correct for – biased data in the tools we build.
So it matters who is writing the code because the code defines the algorithm, which makes the judgement on the basis of the data.
The information is here.