Quite honestly, I was not greatly surprised by the article on BBC, that Artificial Intelligence (AI recruiting did not manage to overcome human bias (see link to the right). Yet, it got me interested in the question, why we are not able to overcome bias in recruiting with AI 

Just as a reminder: all of us have biases. I like to call them the more subtle siblings of prejudices. Biases operate like filters for information for us, and actually help us in being more efficient with our attention. However, if you want to work for example towards a more egalitarian approach to hiring, they may operate against your intentions.

While I am a little conservative when it comes to AI, I have been surprised by some news in the field. For example, I recall findings that computers can become better in playing chess than human beings or can give better diagnosis for patients than physicians.

I can't help wondering, if the key difference lies in the nature of the decisions related to each domain. I like the Cynefin model (see right) to classify situations.

With both chess and medicine there are clear constraints for the decisions - either rules of the game and the moves of the partner or there are the symptoms that a patients shows. I argue, therefore, that these are either clear or complicated situations. Recruiting falls into a different category. To me it is typically complex (hopefully not chaotic). Even if it is an established role in an established company, there are a lot of unknown factors, for example, whether or not the candidate and the existing team will collaborate well.

As the nature of the situation is so different, I wonder, if we understand enough about human heuristics, that is the way we make predictions based on previous experience. Do we know, which heuristics work best in recruiting? A common one is to anticipate future behaviour from past behavior of the candidate. 

But there must be more. As the employment world is changing so rapidly and new profiles are constantly emerging, there may be jobs for which no one has matching experience. So which heuristics can help then? I personally don't know, but I invite  your insights in the comments section.

https://www.bbc.com/news/technology-63228466