Biases in artificial intelligence

Why is this important?

The second wave of machine learning, powered by neural networks and deep learning, is accelerating the development of what artificial intelligence can do. However, it is useful to keep in mind that what the artificial intelligence is learning reflects the biases of our past and present. Because the learning is based on existing datasets, artificial intelligence repeats the behaviour that is inherent in that data.

Things to keep an eye on

When using machine learning it is crucial to understand what data is being fed into the algorithm and what kind of biases it could have. Also the power of people just messing around should not be underestimated, when it comes to releasing an artificial intelligence “into the wild” to learn, as Microsoft’s Tay demonstrated. However, we are not doomed to repeat our mistakes: the racist, misogynous biases can be detected and corrected.

Selected articles and websites

Can computers be racist? Big data, inequality, and discrimination
Artificial Intelligence’s White Guy Problem
Biased bots: How do we teach them ‘good values’, and is that even possible?


Mikko Dufva

Research Scientist VTT Technical Research Centre of Finland Ltd