Search engine bias

How search engines reinforce social stereotypes in gender and race.

Researchers in the field of person perception explore how we reach quick judgements about the people we meet.  They suggest that we judge people on two dimensions – warmth and agency. Warmth concerns itself with pro-social intentions that can be considered as positive or negative (e.g. kind, or insecure). Agency concerns itself with competency to carry out goals and desires (assertive or rational or intelligent).

This is where gender stereotypes come in.  Women are expected to be high warmth, low agency – men the other way round.  And research shows that individuals who are perceived not to confirm with this are overlooked at work.

Testing Bing for gender bias

An article in Harvard Business Review describes research from the University of Sheffield’s Information School into whether image search engines perpetuate gender stereotypes.

The researchers image searched 68 characteristics and analysed the top 1000 results.  They collected results from four regions (India, South Africa, the UK and the US).

Key findings:

  • Warm traits with negative implications (e.g. shy) were always more likely to retrieve images of women
  • Warm traits with positive implications were more likely to retrieve gender balanced results
  • Although regional results did not retrieve the same images, the gender distribution was quite consistent.

The researchers hope that developers will think about how bias can be introduced into each stage of the development process and to prioritse developing automated bias detection methods.

Other recent research

Sources: Harvard Business Review