Is Trust Holding Back Generative AI Adoption?

By Stefan Dulman

[email protected], https://www.dhirubhai.net/in/stefandulman/

The large-scale adoption of Generative AI (GenAI) has hit roadblocks. While corporations paint an optimistic picture, actual implementation remains limited—mostly chatbots and Copilot deployments. Explanations range from technology maturity to regulations, but one key factor often overlooked is trust—our trust in technology and how it evolves over time.

The Fragility of Trust

Trust is complex, varying across cultures and shifting over time. Historically, we've moved from enthusiastically broadcasting our existence to the universe to hesitating, fearing unknown consequences. In human interactions, trust builds slowly but can shatter instantly—a reality that also applies to AI adoption.

Once privacy and security concerns are addressed, AI discussions often focus on accuracy. But is accuracy alone the right metric? Could we instead measure trust?

A Case Study: Autocorrect

Take autocorrect—powered by AI for over a decade, yet still failing users daily. This morning, mine turned “the” into “three” and even misspelled its own name as "autocorrect." Despite consistently improving accuracy, small errors erode trust instantly. After one mistake, I instinctively double-check all messages, reinforcing my lack of trust.

What if AI tools measured user trust through behaviours like scrolling back, time spent reviewing, or typing corrections? Would optimizing for trust rather than raw accuracy lead to better adoption?

AI and Trust in Critical Applications

This problem extends to more complex AI tools. A recent issue with an AI code formatting tool—where two lines vanished in a 600-line file—shook my trust. It raised unsettling questions: What else did it miss? Can I even trust the AI-generated tests?

This lack of trust might explain why the promised 50%+ productivity boost from AI coding assistants is shrinking each year. The hesitation to rely on GenAI fully—whether for writing emails, formatting code, or automating workflows—stems not from its capability but from our instinct to safeguard trust.

The Road Ahead

Will GenAI adoption stall due to trust issues, or will we find ways to reinforce confidence—just as we eventually accepted automatic software updates? The future of AI may depend less on perfecting accuracy and more on ensuring users trust the technology enough to stop second-guessing it.