Page 36 - IAT
P. 36
The Human Algorithm:
Why Diversity is AI’s Missing Code
judicial decisions to healthcare access—the cost
of homogeneous development teams becomes
exponentially higher. Every biased algorithm
doesn’t just make a mistake; it perpetuates and
amplifies societal inequities at scale. A biased
hiring AI doesn’t just reject one candidate; it
shapes thousands of careers. A skewed medical
AI doesn’t just misdiagnose one patient; it
impacts countless lives.
These aren’t hypothetical scenarios—they’re
Ashutosh Upadhyay already unfolding around us with alarming
consequences. Let us take the example of what
happened with Amazon’s AI recruitment tool,
AI Automation Consultant & Marketer which turned into a cautionary tale of gender
ashutosh@thealgohype.com
bias in tech. Trained on a decade of male-
dominated hiring data, the system taught
One morning in 2015, a Google engineer’s inbox itself that being male was a mark of success. It
pinged with a message that would shake the tech began automatically downgrading resumes that
world: their AI was labelling Black people as “gorillas.” included terms like “women’s” or mentioned
This wasn’t just a technical glitch but a stark warning all-women colleges, effectively slamming doors
that we were building artificial intelligence with real before deserving women candidates could even
human blindspots. It was the moment that forced the reach them.
tech world to confront an uncomfortable truth: the
most sophisticated code in the world is worthless if it
doesn’t understand the humans it serves.
The Hidden Code in AI Development
Imagine building a universal translator while only
speaking one language. Sounds absurd, right? Yet
that’s precisely what happens when homogeneous
teams build AI systems to serve a diverse world.
The code might be flawless, but the context is fatally
flawed. We’ve seen this play out time and again—
healthcare AIs missing heart attacks in women
because they present symptoms differently than
men, facial recognition failing in different lighting
conditions because it was tested primarily on lighter
skin tones and voice assistants creating digital
barriers for entire communities because they struggle Even more troubling is the healthcare
with accents. algorithm that affected millions of Americans.
What seemed like an objective measure—
The Stakes Have Never Been Higher using past healthcare costs to determine
medical needs—revealed a dangerous blind
As AI systems increasingly shape our world—from spot. Because Black patients historically
32 | INSIGHT EXCHANGE INTERNAL AUDIT TODAY
