‘AI has no moral obligation’: Industry calls for DE&I to be part of AI training

“AI is not human. It has no moral obligation in any capacity. We have the obligation.”

That was a reminder from Karina Wilsher, partner and global CEO at Anomaly, during a session at the AdColor conference on Friday about how to responsibly approach generative AI with diversity, equity and inclusion (DE&I) at the forefront. 

Held at the J.W. Marriott in Los Angeles, the session also featured Ariba Jahan, founder of Up Next in Tech, and Chandreyi Saha Davis, global VP of brand marketing, trips brand at Expedia Group, who discussed how to best train and use AI in a way that elevates diverse voices. 

While the panelists agreed that AI is a powerful tool that can be used to enhance creativity, streamline processes and get to ideas faster, they also acknowledged that from a diversity perspective, it has a lot of challenges. 

“Prejudice is already baked into the data” that feeds AI, Wilsher pointed out, as it learns from human inputs and the “sexist, racist and homophobic content” that lives online. 

She shared that Anomaly ran an experiment that revealed that prompts for visual representations of women often generated hypersexualized and stereotypical images. Meanwhile, people with darker skin tones weren’t accurately represented at all. 

Expedia is already putting DE&I at the core of a new ChatGPT plug-in that uses conversational AI to help customers plan their trips. Davis noted that it builds out the product, its training the AI on how different kinds of people travel to avoid biases in its results. 

“We're helping the AI get smarter about bias that may exist,” she said. 

Often, however,  the scientific vocabulary surrounding AI can make the technology feel inaccessible to diverse communities who are often excluded from the tech industry, Jahan said. But the more diverse audiences shy away from the technology, “the more we perpetuate this continuation of stories being told for us and not with us,” she said.

Jahan urged diverse leaders beyond the scientific community to help influence AI models by calling tech companies out and urging them to reexamine the data they feed to training models, as well as their policies on how that data is collected and handled. 

As calls grow louder to regulate and implement standards around AI, legislative bodies will continue to take action around data standards and usage. 

But when it comes to developing AI models and creating processes for how it's used at their companies, diverse groups must proactively take part in those discussions if they want to be involved. 

“Oftentimes, we come into the conversations having to retroactively address issues. [But] we have the power to fix it right now,” Jahan said. “Let's make sure that future technologies are not built without us.” 

Expedia for instance, is already putting DE&I at the core of a new ChatGPT plug-in that uses conversational AI to help customers find information and plan their trips. Davis noted that as Expedia builds out the product, it simultaneously trains the AI on how different kinds of people travel to avoid biases in its results. 

“We're helping the AI get smarter about bias that may exist,” she said.

“It's emergent technology,” she continued. “If we don't confront these conversations to discuss the biases, where does that land us?”