Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

Does your organization have trust issues with AI?

Mistrust is well placed, but a purpose-built model can address that trust.

[Photo: Getty Images]

Profile photo of Todd Slind

Vice President of Technology · TRC Companies

Every year, PwC does a global survey of thousands of CEOs about the most urgent issues they focus on as leaders. It’s no surprise that this year’s survey showed executives are paying particular attention to AI adoption challenges. But what may be unexpected is that one of the biggest obstacles to AI momentum isn’t related to technical issues, but trust.  

Trust is a word that comes up again and again in PwC’s report. In fact, they even promoted the report with this provocative headline: “An AI trust gap may be holding CEOs back.” The survey data revealed that few executives currently have personal confidence in embedding AI into key processes. Only a third of executives reported a high degree of trust and confidence in AI. The other two-thirds of executives? Their trust level was moderate at best. 

Why the lack of trust? To me, it’s all about confidence in AI’s ability to provide effective decision support. As a senior executive making high-stakes decisions, you need confidence in the information and insights you factor into the deliberative process. But trust is hard to build when so many AI-generated results leave you shaking your head, either because they are obviously inaccurate, vague, or difficult to verify. 

That last point is a really important one. Even if the guidance an executive receives from an AI-driven tool may be correct and valuable, there is very little backup about what it was based on. The lack of traceability and verifiability does not inspire confidence, particularly compared to other traditional sources of decision-making support such as thick reports from internal business analysts and external experts. In contrast, AI too often feels like a black box giving you no-context guidance.  

THE DEEP ROOTS OF AI TRUST ISSUES 

Why are business leaders seeing AI results that are inaccurate, vague and/or lacking verifiability? AI hallucination is part of the answer (a topic I discussed in one of my prior Fast Company columns.) But these trust issues have deeper roots. I believe it’s because of the way that large language models (LLMs) have been built over the past few decades. The goal for these projects—both in their infancy in the academic community and in the way large tech companies have taken them to the next level—was never about business decision support. The goal was to achieve artificial general intelligence (AGI) that replicates the human brain’s full range of abilities.  

Given that broad goal, LLMs have been trained for decades with the broadest data sets imaginable. That breadth of data inputs makes sense when AGI is the objective, but it is a drawback when those LLMs are asked to focus narrowly on something like the specific business decisions for a company navigating a specific market landscape. That mismatch between the broad LLM training and the specific needs of corporate decision-making support is at the heart of why so many AI outputs make business leaders skeptical rather than confident.  

PURPOSE-BUILT MODELS 

Purpose-built AI models are critical to get outputs that build more trust with business leaders. Rather than using AI-driven apps that plug into LLMs, organizations can use customized AI models trained with industry-specific and company-specific information, further refining them through input, coaching, and feedback loops from experts with deep knowledge. Purpose-built AI models address each underlying issue that causes these trust issues. They deliver outputs that are highly specific, have far fewer inaccuracy instances, and more depth, insight, and actionability. And purpose-built AI models provide more traceability and verifiability about the foundation for those outputs.  

My team and I have worked with a number of organizations that developed these purpose-built models. Companies that tackle this themselves need to be willing to invest in the necessary work to monitor and retraining their models. This training is likely to become core capabilities of IT departments of any appreciable size if AI truly is here to stay in corporate computing. 

INDUSTRY INITIATIVES 

However, not every company is in a position to devote the resources to go it alone on this kind of project. That is why I think we will see a strong trend toward industry initiatives where organizations band together to build AI models and tools that spread the cost, risk, and benefits across multiple organizations. I see many advantages to this shared approach, including stronger models built using industry-wide datasets and the expert input from across an industry rather than just an individual organization.  

A great example of this is an AI-focused consortium led by EPRI that was recently launched in the power/utilities sector, bringing together utilities, power companies, scientists, and other industry experts. Members collaborate on a series of AI initiatives, including development of a purpose-built AI model and best practices that will provide better corporate decision support. Together, the participating companies will be able to arrive at solutions that provide a “future proof” answer to their needs while also reducing risk, saving time and money, and bringing AI-drive decision support tools to fruition much faster. 

This kind of collective initiative reduces the burden on each organization, giving them access to AI foundation models that are trained on specific, but general industry context. That leaves only the unique aspects of the organization and its service area to be trained into the model to help ensure it produces outputs that organizations can verify and trust. 

Trust issues are never solved overnight, including in the world of AI. But this approach is a critical step toward delivering outputs that business leaders can truly have confidence in.  

Todd Slind is the vice president of technology at TRC.