Why Diversity Matters in AI

"Perception is the foundation of human experience, but few of us understand why we see what we do, much less how." — Beau Lotto

I always knew that our experiences very much color our perception of the world, but a recent Google talk by Deviate author, neuroscientist, and entrepreneur Beau Lotto, jolted me to think about the future as we develop AI that will soon become as ubiquitous as the air we breathe.




It turns out only 10% of what we perceive as we move throughout the world is directly attributed to our eyesight, and the other 90% is conjured up by the brain, relying heavily on the context of our other senses and previous experiences to guide what we make of what we see.

Context is Everything

What we see is the meaning of data, not the data itself, and what we use to assign meaning to data is history: our past perceptions, our culture, and our evolutionary ancestors. Data is meaningless without the context that creates meaning.

A recent example from popular culture would be the Yanni or Laurel debate, which did you hear? It turns out that it is contextual depending on which you expected to hear and what circumstances you are in when you hear it.

When I first heard this, it immediately reminded me of the McGurk Effect in which the sound you hear changes depending on what you are looking at (Fa, Fa, Fa, Ba, Ba, Ba).

This effect is the heart of the problems machine learning and other AI algorithms are solving, learning to assign meaning to data. Adaptive systems, like our brains, only take logical small steps to the next most likely possible outcomes. Different experiences create larger search spaces and more possibilities for small steps in multiple directions to result in giant leaps of innovation.

Beau makes a compelling point about the diversity of groups. He states if he could choose between two groups: one average lower IQ from diverse backgrounds and another which has a higher average IQ from the same background, he would choose the first group. Why?

Bias Recognition

We are rarely the ones that can recognize and call out our own biases, so the first group is more likely to reveal their assumptions to each other and deal with them, or at least be aware of them.

A More Extensive Search Space

This group has a higher dimensional search space because they have more diverse assumptions and experiences. Complex systems theory teaches us that the best solutions are more likely to exist in a more complex search space.

We are at a critical point in time to make sure that the future AI we build will draw from the more complex search space of experiences, history, and culture of the diverse population of the earth. I am proud to be involved with several groups that are working to advance this mission including Women in Technology International, Humans for AI, and Hip Hop Leaders International.

I encourage all technology professionals and those interested in how technology is going to shape our future to get involved in some way because diversity matters, and unfortunately, the status quo is less than ideal in this regard.

We should all be striving to expand our search spaces, and as the world becomes more connected, we can share experiences and learn from each other. It is up to us to make sure the technologies we design today lead to better solutions for those that will inherit our experiences to shape their perceptions for generations to come.

Nikki Hallgrimsdottir is a co-founder and chief evangelist of Algomus, where she works with manufacturers, distributors, and retailers to leverage Artificial Intelligence and Data Analytics. Algomus has built Algo—an AI agent that assists business users with Data Analytics and other related tasks through Natural Language Processing, Machine Learning, and Business Process Automation.

Prior to founding Algomus, Nikki worked for a decade with companies of all sizes to utilize automation technology in both manufacturing and product design. She knows enough to be dangerous in the areas of Big Data, Analytics, Artificial Intelligence, IoT Hardware, Machine Vision, Supply Chain, Industrial Automation, and Multiphysics Simulation, to name a few.

Opinions expressed by the author are not necessarily those of WITI.