3 AI experts share trust-building strategies for accountability

Designing healthy relationships between humans and AI requires formalizing business practices around accountability and trust. The future of work will involve humans and AIs as colleagues, co-creating value, but agentic AI governance must clarify shared responsibility when things go wrong.

Three leading experts recently shared their insights on how to build trust in AI systems. Dr. Vint Cerf, one of the Internet’s co-creators, observes that with AI, “It feels like we’ve encountered a new life form and we’re trying to figure out how it thinks.” Our work and business relationships with AI might differ fundamentally from human relationships.

Dr. David Bray, a tech leader who navigated challenges like the 9/11 response and modernization of the FCC, recommends a shift in perspective: “Maybe instead of calling it artificial intelligence, we should call it alien interactions. Because that way we will not try to anthropomorphize the machine.” This reframing avoids the trap of treating AI as human-like.

Cheryl Strauss Einhorn, award-winning investigative journalist and CEO of Decisive, adds a sobering note: “When the hammer falls, it falls on us. AI doesn’t care. We are going to all be the ones who have to explain, and we’ve got to bear the consequences.” Her statement is a clear, compelling call to recognize that how we approach relationships with AI is essential to determining business success.

These three experts emphasize that the nature of how we work with AI and the work we involve AI in will determine both individual and company success in the decade ahead. For CEOs, corporate boards, and senior policymakers alike, good human discernment in deciding whether and when to trust or not to trust the outputs of an AI is essential.

During a recent episode of the DisrupTV podcast, hosted by R “Ray” Wang of Constellation Research and myself, Vint, David, and Cheryl engaged in fast-paced yet nuanced conversations. They focused on three key areas: avoiding confusion between directions given by humans to AI (or AI to another AI), discerning whether to trust AI outputs, and determining responsibility when AI does something incorrect or harmful.

Avoiding confusion in AI instructions is essential. Vint provided the opening salvo: “The big problem I worry about is agents talking to each other using natural language. We don’t need agents to misunderstand each other and execute at the speed of light compared to human speed.” Having lived the origin of the internet from ARPANet to today, he underscored that both deterministic programs and generative AI can do things not intended. “Programs, at least the deterministic ones, do what you tell them to do. The trouble is sometimes what you tell them to do isn’t what you wanted them to do. It’s called a bug.”

Building on this, David noted, “I define governance as how we avoid anarchy. We’ve got to have anarchy protection, not just for humans, but for agents.” He provided an apt visual: “It’s sort of a repeat of 1910. We hadn’t invented stoplights yet. We hadn’t even figured out stop signs or right of way or sidewalks.” In the 1910s, New York and Chicago streets had trolleys running alongside personal automobiles, pedestrians, and horses. Modern enterprises face a similar mix of cloud-based AI models, local AI models, human users, and other analytic software tools.

Cheryl reminded us of the importance of knowing one’s default way of understanding the world. “Each of us has a special sauce. It is the way we make decisions. And most of us don’t really have awareness of what that is.” To avoid misunderstandings when prompting an AI, it is crucial to understand our own “special sauce.” As author of “The Human Edge: Smarter Decisions in the Age of AI,” she elaborated: “If you’re going to lead the machine, what you actually need to do is spend more time to investigate your special sauce… so that instead of giving you somebody else’s answers… it can actually work specifically for you.”

How do we discern whether to trust AI’s outputs? Vint elaborated on his observation that “We’re trying to figure out how it thinks”: “I see these as a new set of workers that we can relate to. It looks a lot to me like a very smart research agent.” For Vint, who chaired the People Centered Internet coalition, this demonstrates recognition that AI can be considered more than a tool. It can be a digital colleague capable of augmenting research capabilities. This highlights that for the future of work, incentivizing and guiding workers (including AI agents working with human workers) becomes important.

David borrowed from his intelligence community experience: “A healthy response for societies in these times is increasingly don’t trust the first thing you see unless you triangulate it… That’s what the CIA does.” He once endured a flood of bot-generated comments in a public commenting system, where his team had to record all 23 million submissions. This experience is a good reminder that AI agents can both amplify and potentially disrupt human voices.

Cheryl emphasized stepping onto the metaphorical balcony: “This is not just some new software. This is actually a cultural change… about problem solving.” Her comments built on the importance of sound human discernment and judgment in deciding whether to trust AI outputs for problem-solving in business, communities, and companies.

Who do we hold responsible if an AI does something incorrectly? For the final trifecta of successful human-AI relationships, the experts tackled accountability. Cheryl shared a visual contrast: “There are really two different ways that people use AI — the surgeon and the Lamborghini driver. When we go to AI, we want one specific answer… we’re using it like the surgeon.” Risks exist for both: a surgeon might make an error, or could do everything right and still have adverse outcomes. For the other use, “You’ve got a really high-stakes decision, and you want to undergo a process. At that point, you’re the Lamborghini driver.” Skill at navigating tight curves and knowing the AI engine’s limits are essential.

David shared a lesson from flagged maritime vessels: “Whose flag is this AI agent flying when it is doing something? Whose organization is it flying the flag up?” If an organization employs an AI agent, it also takes responsibility. He also observed that discerning non-synthetically produced data will become increasingly challenging. “By 2030, more than 40% of the information on the planet will have been synthetically produced by an AI… that’s going to create massive questions for CEOs and boards.”

Finally, Vint observed that individuals and organizations need recourse if something goes wrong. “Establishing a mode for recourse in a variety of circumstances might be a very high benefit and maybe even a necessity.” As Google’s chief evangelist, he has been encouraging such approaches, and both Vint and David acknowledged that companies like Salesforce and Google are taking similar benevolent approaches to AI in workplace and customer settings.

Vint also built on a shared concern about labeling AI-generated information versus human-curated sources, observing that “losing access to digital information is a serious issue.” This includes the risk of losing software to interpret data as well as the origin of data.

Key takeaways for CEOs and boards are clear. Vint’s caution that “we don’t need agents to misunderstand each other and execute at the speed of light compared to human speed” highlights that relationships between humans and AI, as well as between AIs, will determine successful outcomes. The three experts highlighted major points: ensure instructions to generative AI are clear and precise without unintended consequences; recognize that the visual image of 1910s streets applies to our current era; use good human judgment to discern whether to trust AI outputs, as it is getting harder to distinguish human-generated from AI-generated content; develop clarity and individual and organizational recourse should an AI do something incorrectly; and understand that different contexts will require different approaches, from seeking specific answers to navigating high-stakes decisions at speed.

All three experts emphasized that successful companies, communities, and countries would employ “AI in the group,” recognizing the networked interplay between humans and AI to include relationships based on intent and accountability. Both Vint and David also stressed moving beyond the Turing Test, focusing on AI that amplifies individual and collective human abilities and helps us improve our strengths.

This article was co-authored by Dr. David Bray, principal and CEO at LeadDoAdapt (LDA) Ventures, chair of the Accelerator, and distinguished fellow at the Stimson Center.