Ways to solve for Bias in AI - thoughts?


We wrote this initial piece on Bias in AI, it’s impacts and captured global best practice into how to resolve for it.


In this blog post we explore AI bias, and point to practical ways to resolve the achilles heel.

We’re interested in AI & getting it right, which is why we are taking an interest in bias!

For those with some exposure to AI, what do you think?

How can technologists remove bias in AI?

What practical ways are working in your tech businesses?


PSA: this is going to feel a little meta if you have limited knowledge, reading, experience or exposure to AI or machine learning. If this is you - perhaps check out this AI 101 real quick and come back here for seconds!


You gotta have AI before you discuss bias in AI I guess. The key is to set up the right strategy to store useful data before machine learning and AI can come in. They are just algorithms; without data they are just academic thought.


Agreed @xiaodai . The huge first step is getting the data and storing it right. As you can imagine, we are in the throes of data strategy and design right now. As you say, you need an awful lot of data before you can start applying algorithms!


The issue with many so-called AI solutions is that they are really just hyped-up machine learning. As such, the parameters by which such machine learning occurs are constructed to reflect what it is the user thinks they need to know or learn. Genuine AI however must use unsupervised deep learning by which the software deduces data connections and insights from what it ingests. That then reduces user biases and discloses insights not requested and otherwise unobtainable.