Dichotomies of Generative AI: Accelerating outputs while mitigating bias
Editor’s note: We’re continuing to share our team’s Dichotomies series, a collection of stories that explores the potential impact of emerging technologies through speculative fiction. We’re thrilled to share that our team’s work on the Dichotomies report will be featured on the main stage of South by Southwest 2024! Deloitte Consulting’s Chief Futurist Mike Bechtel will be a Featured Speaker for the 2050 Track at the annual conference in Austin.
This week, we’re looking at generative AI. Since this technology exploded onto the scene more than a year ago, businesses have been racing to figure out how it can fit in their operations and what it will mean for their industry. These two narratives look at how, when implemented mindfully, generative AI could enable breakthrough discoveries that benefit broad swaths of society. However, on the flip side, when implemented with a mind only for cutting costs, generative AI could amplify biases and harm people and the businesses that serve them. Here’s how these scenarios could play out:
The Allure
Aarti’s car drives itself with ease into a sharp loop off the highway. She drums her fingers incessantly on top of the steering wheel — a bad habit she picked up from her father. As the car parks at her office, Aarti’s ears perk up to the voice of Anderson Cooper, rendered by generative AI at her request. Her Monday podcast, tailored to her interests in biotech news, plays a snippet on the latest healthcare scandal. She cringes as she leaves the car.
Aarti strides into her lab and greets the tired faces of her researchers. For the past few weeks, they’ve been assigned to a drug development project that could ease the symptoms of dementia, and their board wants results as soon as possible. Aarti and her colleague Roger study the latest outputs of their proprietary generative AI program: a dozen viable, high-fidelity protein structures, replete with percentages to indicate likelihood of side effects.
As Aarti guides Roger on which structures to feed into their quantum molecule simulator to forecast viability, she receives a call on her tablet from the CEO of their company SaluTech.
“Ooh, Aarti’s in trouble,” Roger jokes.
“Shush,” she replies.
A hologram of their CEO Lindsay, looking distraught, appears on Aarti’s tablet. She immediately shares a video with Aarti: a press release from the WHO alerting the world to a novel zoonotic virus identified in Zurich.
Aarti’s eyes widen. “Lindsay, I want to —“ “I know, that’s why I called. Shelve the current project and give me five viable vaccine options to move towards clinical testing by the end of the week.” Lindsay cuts the call short. Roger and the other researchers stare.
“Let’s get started!” Aarti declares, and the lab springs into a frenzy. Someone shouts out that the WHO has already sequenced the virus, so others begin feeding the info into their AI to produce vaccine candidates.
While the team scrambles, Aarti sits still at her desk, drumming her fingers nervously across the marble surface. More than a decade ago, her father passed away from COVID-19 before a vaccine was available. Even with only one spike protein to address, drug development took an entire year. This newest virus could have hundreds of mutating spikes or require an entirely novel method of inoculation. Whatever it was, Aarti was bent on advancing the field. With the speed of AI simulations, she knew her team could stop the next pandemic before it affected millions of families like hers.
Aarti snaps out of her reverie and employs an AI marketing assistant to draft a press release, prompting it to talk about her past and her company’s desire to create the first vaccine. Remembering the scandal she heard about on her morning podcast—about the marketing issues of the Gen AI startup Deliveri, Aarti makes sure to send the article to SaluTech’s PR manager for review. She also provides permission to generate a video using her face and voice, so SaluTech’s audience could connect to the emotions of her father’s passing. Then, she rejoins her team — she’s eager to dive into the details.
The Concern
“Only one today,” he mutters to himself as he sits down with his morning coffee. Xavier, the marketing lead of Gen-AI startup Deliveri, is trying to cut back on his caffeine intake after dozens of alerts from his smart watch about caffeine fueling his anxiety and insomnia. He opens a laptop for his Daily Download, a personalized report generated each morning with his daily agenda and relevant industry news.
Expecting a leisurely read, he instead snaps to attention as his AI assistant Cara alerts him of being late to an urgent meeting. “How did I miss that?” he wonders. His alarm increases as he reads the article within the invite.
Xavier can’t believe the contents of the article. Before he can even process, Cara alerts him that the company’s founder, Ajay, is waiting in his virtual meeting. Xavier knew better than to make his boss wait.
Ajay is yelling at the team as Xavier joins. “What do you mean you can’t retrain the program? Isn’t that what I pay you engineers to do?”
“Well…” Laurence, the head engineer, hesitates to find the right words for Ajay’s temperament. “You asked us to use AI-as-a-service to cut costs, and the bias is baked into the vendor’s AI training data. It’s going to take time to rectify.”
“Ugh! Xavier, let’s see if you can prove more useful today. Use BrandBoost for a marketing campaign that shows how inclusive we are. Send it out before lunch.”
Before Xavier can object, Ajay ends the meeting. Ajay had laid off over half the staff and increased reliance on generative AI vendors, which meant Xavier was the entire marketing department. Xavier rakes his hands through his hair as if to bring some ideas out of his head to life. He opens BrandBoost, an AI program used to build multi-modal marketing campaigns. With the fear of Ajay’s deadline looming, he rushes to enter various prompts to produce press releases and video advertisements and uploads them without review.
As the afternoon passes, Xavier asks Cara to assess engagement with the posts he previously distributed. “Not positive,” she declares. Xavier’s eyes widen at the flurry of comments pointing out that the ads only include White mothers and infants, not the Black mothers who’ve been impacted by Deliveri.
“Book an urgent meeting with Ajay in the next available time slot,” Xavier instructs Cara.
He looks at his coffee mug from this morning, pondering how much more caffeine he’ll need to get through what he knows will be a horrible evening.
Takeaways
Human and machine, better together
As generative AI becomes more accessible, reliable, and robust, more workers can expect to partner with these tools in their daily work, as detailed in Deloitte's recent Benefits and Limitations of Generative AI report. Lower-order tasks such as preliminary research or drafting, content generation, and summarization can be delegated to machines, while humans focus on higher-order tasks. For instance, Aarti relies on AI to generate options for protein structures but applies her own expertise to determine the best options. Going forward, organizations should be looking to hire people with uniquely human skills like ingenuity, adaptiveness, and problem-solving, while the machines do what they do best.
From black box to glass box
Widespread adoption of AI across industries could turn algorithms into high-level decision-makers. While this may greatly lower costs and increase productivity, trust will be the differentiating factor between successful adoption and disastrous outcomes. As detailed in Deloitte's Tech Trends 2023, deploying frameworks to make AI more responsible and transparent, as we would expect a human colleague to be, can ensure that organizations maximize value and mitigate risk. Otherwise, Xavier’s trouble with an opaque and unreliable AI could become all too common.
Move fast, but don’t break things
Generative AI technology could eventually lead to breakthroughs for seemingly intractable problems, like dementia or the next global pandemic. The computational power of AI can exponentially speed the completion of tasks that are typically inefficient or time-consuming for humans, like trial-and-error experimentation. Yet, as Xavier finds out, the speed of generative AI often needs to be tempered by human reviewers, as detailed in Deloitte's Proactive Risk Management in Generative AI. Organizations can develop a generative AI strategy by pinpointing the areas with the highest potential for efficiency gains, and where checks and balances may be required.
- Deloitte NExT Team, Office of The Chief Technology Officer
- Abria Perry, Field Notes From The Future Acting Editor