Artificial Intelligence in a Biased World

Dataforce Presentation


DataForce Panel: Artificial Intelligence in a Biased World

Location: European Court of Auditors (ECA)

Date: March 30, 2022

Presenters: Alexandros Poulis, Senior Director, AI, DataForce

Session Description:

Bias is a disproportionate weight in favor of or against something. We, humans, are all biased in one way or another. Human bias can be harmless, but it can also be extremely damaging to our society. Bias in AI, however, is terrifying. Imagine racially biased self-driving cars, gender-biased drug developments profit-biased insurance plans. In each of those three cases it only takes one, centrally deployed algorithm to affect hundreds of millions of people anywhere in the world.

Not long ago we witnessed concrete cases where computer vision discriminated against people of color, a recruitment algorithm discriminated against female applicants and an online conversational AI turned into a hateful, racist bot within minutes of each public release. But AI is, in some sense, a mirror of humanity – just orders of magnitude more powerful. AI learns from data, and data has been mostly created by people.

In this session we’ll discuss how we can affect AI’s “upbringing” by focusing on good data. We’ll look at real-life examples from multiple industries and we’ll review some of the most powerful tools to identify and limit bias in AI. Some of those tools are for data scientists and some can be used by every single person.