Is It Possible To Remove Bias From AI? Former Google AI Researcher Timnit Gebru Weighs In

Is It Possible To Remove Bias From AI? Former Google AI Researcher Timnit Gebru Weighs In

    The idea of removing bias from computer science is a complicated one, and according to AI ethics researcher Timnit Gebru the answer may not be particularly clear-cut.

    How do you design a search engine that provides equal data on men and women in positions of power, when almost 90% of CEOs are men?

    AI is a complicated enough business as it stands, but there are conversations making waves at present regarding the ethical side of digital technologies.

    The conversation topic at play – that of removing bias from AI and search engines, and introducing fairness – is made difficult by the fact that computer scientists have a different definition to both than what most of us would be familiar with.

    While most people consider bias to mean “prejudiced against a certain group or characteristic”, in computer science it’s actually more akin to programs that are consistently incorrect in one direction or another.

    As an example, if a weather program were to consistently overestimate the likelihood of rain, it could be described as statistically biased.

    The issue with this clash in terminology is that for tech companies, there is no obligation to clarify which concepts of bias and fairness they are prioritising, and can easily lead consumers down pitfalls accordingly.

    According to AI ethics researcher Timnit Gebru, the lack of restrictions on this front for tech companies means they arguably aren’t being held responsible enough for how they communicate the issues.

    “There are industries that are held accountable,” said Gebru. “Before you go to market, you have to prove to us that you don’t do X, Y, Z. There’s no such thing for these [tech] companies. So they can just put it out there.”

    Until tech companies can figure out a way to make clear the separate terms of bias and fairness, there simply doesn’t appear to be a clear answer for how to solve this ethical dilemma, and AI researchers will likely have to continue on as they currently are.

    For a deeper look into this topic, check out Vox US’s take on the situation: AI bias: Why fair artificial intelligence is so hard to make – Vox

    Please login with linkedin to comment

    AI AI ethics Google

    Latest News