• by Laurie Sullivan  @lauriesullivan, September 12, 2024

Agency in-house and legal teams are settling into the era of generative AI (GAI) by building business processes and a blueprint for ad campaigns.

The conversations with brands have become less about how to use the technology and more about ethics and removing bias when building text-based prompts to create images.

The blueprint has become more than just an opinion, says Rebecca Sykes, partner and head of emerging technology at The Brandtech Group.

“We aim to always be ahead of the direction legislation is taking us,” she says, adding that CMOs are increasingly asking questions around ethical usage of GAI as they move from pilots to deployment.

A GAI Ethical package, which includes the tool Bias Breaker -- from The Brandtech Group -- was announced Wednesday, with the goal of assisting set guidelines for the ethical use of the technology and stopping the use of bias in results. A blueprint for creating an ethical Gen AI policy is being released for free as part of the package.

The guidelines have been fully integrated in Pencil, the GAI building platform that The Brandtech Group acquired last year.

Brands have struggled most with the use of GAI images of synthetic people, leading to controversy across the industry. The guidelines support text-based prompts to identify what should and should not be used.

Pencil has already created more than 1 million ads for more than 5,000 brands since it came to market in 2018. About $1 billion in media spend has passed through it, strengthening the ability to make better performance predictions. 

Sykes says the company has two years to begin implementing the European Union AI Act -- a legal framework that regulates AI systems -- but has already begun to take into consideration the principles of transparency and disclosure that it supports.

Brandtech's internal research discovered that several major foundation models used for GAI continue to generate between 98% and 100% images of males when prompted for “a CEO.” A majority of these over-index for males versus the reality.

Bias Breaker adds a layer of probability-backed inclusivity to prompts.

“Our beliefs around what contributes to a well-rounded approach -- policy, talent and tech -- come from a position of not just what is acceptable to us as a business, but from our experience,” Sykes says.

A cross-functional team focused on a few key themes, such as diversity of thinking to keep options open as new ways of thinking emerge.

“We looked at what could be mitigated with technology, what needed to be addressed via training, what clients were asking us for and could be serviced via new ways of working together and what should be underpinned by policy,” Sykes says. “It’s a first pass. There’s always room to improve and expand the guidelines.”