At a time when businesses increasingly rely upon artificial intelligence (AI) to accelerate processes and generate insights, business leaders are striving to properly balance their human and AI resources. This balancing act comes down to one question: Should we trust increasingly powerful AI to perform traditional business functions with humans providing oversight, or equip people with AI tools to accomplish business tasks and goals?
As I noted in my last Fast Company Executive Board byline, I believe strongly in the power of AI to elevate almost everyone—if we’re smart and careful in adopting the technology. But “smart” doesn’t mean using AI for everything, or in the same way every time. It’s increasingly clear that over-reliance on AI can alienate human workers, devaluing their uniquely human skills and negatively impacting the quality of their work. Moreover, while trusted data enables trusted AI, data alone isn’t enough to make AI trustworthy—building trust requires the right level of human involvement, which can change from task to task.
So, to increase internal and public trust in AI, I believe business leaders will need to choose between two approaches: AI solutions with human oversight, and human solutions with AI assistance. Moving forward, it will be critically important to identify the right approach for each specific business challenge. Knowing which to use, and when to use it, will separate successful executives from the growing pack of AI adopters.
HUMAN IN THE LOOP
Over the past year, “human in the loop” became popular shorthand in AI circles for “human oversight of AI labor.” And in some cases, AI-first workflows make sense. Trusted AI systems are now generating life-impacting medical diagnoses, controlling autonomous and semi-autonomous vehicles, and performing business tasks such as creating marketing content. Whether the outputs are human life-impacting or critical to human perception of a brand, human judgments should naturally serve at a minimum as intermediate and final quality checks on AI processes.
Tasked with examining data and making decisions, AI can handle a lot of aggregating, sifting, and heavy lifting, as well as completing some tasks in ways that approximate human responses. But if AI makes incorrect assumptions, hallucinates an answer, or otherwise drops the ball, a human needs to notice and fix the issue before other people are impacted. This only happens when accountable humans are present, thoughtful, and engaged throughout the AI lifecycle.
AI IN THE LOOP
Most businesses believe that AI is the future, but the path to that future isn’t clear or easy. Few organizations are prepared to capitalize on AI’s massive opportunities, and trust remains a serious concern: U.S. trust in AI companies has dropped from 50% to 35% over the past five years. Regaining public trust in AI will require businesses to refocus on people, transparency, and value creation.
Especially during these early fragile days for AI, I believe businesses should design around the idea of “AI in the loop”—a rebalancing of the division of labor between AI and humans, using AI to supplement rather than replace human capabilities.
Compared with human-in-the-loop solutions, which hand off big tasks to AIs with human supervision, AI in the loop keeps tasks in human hands while giving them AI tools to complete their work. Put another way, this strategy uses AI to support people, rather than using people to support AI.
Even for businesses that aren’t concerned about AI output problems, AI-in-the-loop solutions address a looming internal issue: Handing important tasks to AI risks turning humans into minor actors when they are critically important to differentiating and growing businesses. Flipping a collection of front-line workers and creatives into AI supervisors can make people feel alienated from their employers, downplay the value of their human skills, and decrease their work quality. Monitoring AIs almost certainly wasn’t what they signed up for.
WHEN (AND WHEN NOT TO) USE THE RIGHT “IN THE LOOP” SOLUTIONS
While AI-in-the-loop solutions make sense for almost any business, they’re arguably best suited to situations that directly impact customer perception and satisfaction.
In the medical realm, while a human-in-the-loop solution would have a doctor review AI-gathered patient complaints and approve or change an AI-provided remedy or response, an AI-in-the-loop solution would equip a doctor with an AI note-taking system, freeing the doctor to have a discussion with the patient and use their judgment to consider solutions. AI has an important role to play in each case, but in one, the doctor and patient merely become inputs into an AI-generated solution—the type of “care” that leads people to feel dissatisfied, like they’re seen as numbers rather than human beings.
The same concepts apply in retail. While a human-in-the-loop solution might use AI to screen customer inquiries and suggest items to purchase based on human-approved patterns, an AI-in-the-loop solution would use AI to log and monitor calls, enabling salespeople to focus on human interactions and completing sales. This doesn’t preclude the AI from producing valuable insights or suggestions; instead, they go to the salespeople, who can leverage them to improve transactions and customer satisfaction.
NOT A ONE SIZE FITS ALL SOLUTION
AI tools are already driving a lot of business change, and there’s no question that they will improve efficiency, particularly as they mature. But at the same time, we should use AI tools to improve the worker experience, as that will lead to superior and more trustworthy AI-powered results for employees and customers alike.
Introducing this incredibly important new technology into our lives requires more than a one-size-fits-all approach. Switching between human-in-the-loop and AI-in-the-loop solutions will be key for the next generation of business leaders, enabling them to make the most of powerful AI technologies while earning their employees’ and customers’ trust.