#DebugDiversity: What We Can Learn From Tay, Microsoft’s Racist AI Chat Bot

“Diversity” has become the keyword for addressing the residual impacts of racism. People get deeply offended by the “r” word — particularly when something they say or do is called out as racist. It conjures images of Klansmen and violent, indignant “Whites” depicted in Civil Rights era documentaries like “Eyes On The Prize.” But racism is not the same as being racist.

racism
noun — rac·ism \ˈrā-ˌsi-zəm also -ˌshi-\1: a belief that race is the primary determinant of human traits and capacities and thatracial differences produce an inherent superiority of a particular race 2: racial prejudice or discrimination
rac·ist play\-sist also -shist\ noun or adjective
rac·ist
noun — rāsəst/1. a person who believes that a particular race is superior to another. synonyms:racial bigot, racialist, xenophobe, chauvinist, supremacist
adjective 1. having or showing the belief that a particular race is superior to another. E.g.: “we are investigating complaints about racist abuse at the club”

[Update: “Race” itself is arbitrarily assessed with definition ranging from “a group of people sharing the same culture, history, language, etc” to “a population within a species that is distinct in some way.” Thus, racism can encompass agism, sexism, and similar belief systems related to identity, geographic / national origins, etc. There can also be racism within a group / among subsets; e.g. “internalized racism.”]

Note: One can have racist sentiments without being (intentionally) racist. “How is this possible?” you might ask?

We need to acknowledge racism for the corruption that it is — like a virus that enters through backdoors of consciousness and infects operating systems undetected. Racism passes through subtle messaging in media programming (aptly so called), education, politics, the legal system, urban planning, real estate, finance, HR …Even social media, like Twitter. It has been duplicated and spread throughout society using metamorphic code to evade detection. Despite widespread denial, it is ubiquitous; systematic.

The recent de-activation of Microsoft’s AI chat bot, Tay, exemplifies what is proffered herein.

Making, subscribing to, or defending racist statements doesn’t necessarily make someone (or something) a racist. It does, however, indicate that they have been compromised and should take steps to address the corruption before it does greater damage. It could happen to anybody (or any-bot-y) of any ethic group, any gender, any culture or any color. Some are more vulnerable than others. We soften it with words like “bias,” but bias is a mere output of systems corrupted by the pervasive R-Virus — a signature if you will.

While the overwhelming homogeny in tech creates a positive feedback loop,amplifying tone deaf refrains, and perpetuating racist outcomes, negative feedback is often the most useful, and this comes with with honest assessment such as this post by CODE2040 Co-Founder, Laura Weidman Powers. When racism is identified in a output, it should be looked at as an error signal. There is no need to condemn the offender prematurely, nor for those affected to react defensively in shame, guilt or denial. Rather, it is an opportunity to optimize for the future.

Players in the innovation economy claim and endorse “pattern recognition” in recruiting, hiring, funding, etc. But it is time to pay attention to growing pattern analysis. The fact is, we need to empower controllers, compensators, and feedback structures to correct systematic racism at the root, and prevent it from spreading unchecked to future generations.

We’re all one human race, with many nationalities, ethnicities, cultures, identities…but racism is real. It’s not a bad word. It’s a diagnoses. If we are going to solve the diversity dilemma, we need to be proactive in debugging racism at the root, upgrading our systems to harness the advantages of diversity, and rebooting our collective consciousness on these topics.

See more on feedback loops social constructs: