Bias in AI or Just Another Bad Idea?

by Dwayne Phillips

Once again someone creates all sorts of fancy explanations for what was simply a bad idea poorly conducted.

There seems to be a lot of “bias” in the machine learning area of artificial intelligence research and practice. Or can we explain the problems without using such fancy terms like “bias?”

What were the requirements?
What were the tests?
Did anyone review any of this?
Did anyone ask any questions?
Was anyone allowed to ask any questions?

It is easier for stupidity to slip through when we skip these steps. Afterwards, we create all sorts of fancy reasons for the stupidity (bias, over sampling, under sampling, data integrity, governance, provenance, etc.).

There are persons who know how to approach technical problems. There are sound and proven techniques for approaching technical problems. Many see these persons and techniques as boring or old fashioned or something. That is unfortunate.

If you are managing such advanced efforts, please, step back and use some of these old, proven techniques and maybe even bring in some of these old, proven persons as well. At least for a day.