The Dimensions of Trust: What AI Can and Cannot Own
A colleague, Eric, shared his perspectives on trust after reading Building a Durable Business in the AI Era: that in domains where AI can deliver overwhelmingly superior performance, a transfer of trust can happen, and Generative AI can build brand power. An example he shared is in the area of auditing, where AI could one day outperform humans in both accuracy and efficiency while also eliminating the risk of fraud. When that happens, being “Audited by Claude” could become a legitimate standard trusted by companies.
Incidentally, Anthropic has recently unveiled ten financially focused agents, including agents that “can build a pitchbook, audit statements or draft credit memos”.
In this case, will relying solely on long-standing customer relationships be sufficient to hold on to the trust placed by clients? This made me think more carefully about something I touched on in the previous essay: that AI can replicate certain aspects of trust, but not all of them.
Through my experience with generative AI, I believe it can own some dimensions of trust completely and even surpass humans, such as in reliability and competence. But other dimensions have a limit that better models alone cannot overcome. Trust, as I see it, has six dimensions:
- Reliability is the belief that the other party will show up and perform consistently over time.
- Competence is the belief that they can do the job well.
- Integrity is the belief that they are honest and act in your interest.
- Compassion is the belief that they genuinely care about your wellbeing, not just the transaction.
- Relational trust is built through shared experience, history, and human connection over time.
- Accountability is the belief that if something goes wrong, someone will take responsibility.
What AI owns
Some of the most mentioned winning qualities of AI are that they perform well regardless of time and situation, never wavering in attitude, making it an absolutely reliable and competent digital confidante, colleague, or advisor.
Over time, better models and repeated interactions build trust through reliability and competence. Areas where AI shines are jobs like auditing and diagnostics, where AI will not tire and make mistakes after many hours of processing large amounts of data, or credit scoring, where decisions are not influenced by emotions or personal bias. This is where AI trust compounds — and in some domains, will surpass what humans can deliver.
What AI can’t own alone
Over the past two years, many people have built the habits of using their favorite LLM chatbots or agents professionally and personally. The trust built through reliability, competence, and perceived integrity and compassion has made AI a trusted confidant during difficult times and uncertainty. But in reality, AI is still only a tool.
As an employee, if you use AI for work and act on its output, you are still responsible for the outcome — positive or negative. While the productivity gains are unmatched, there are also instances that remind us of the need to supervise and review AI’s output before acting on it. The same goes for companies that use AI in their workflows. We have seen this play out in recent months, in instances where the sense of reliability and competence breaks: a gaming CEO who asked ChatGPT how to avoid paying a $250 million bonus found it didn’t work, and Deloitte issued a partial refund for an error-ridden Australian government report that used AI.
There is also an element of believing that AI will never lie and will always act as intended. After all, it has no self-interest or ulterior motives. But the absence of self-interest doesn’t automatically mean it will always act in your interest. AI agents have wiped out production databases on at least two separate occasions, and apologized afterward.
. . .
AI today cannot take accountability for its actions. At the end of the day, doctors are responsible for their diagnoses, lawyers for their arguments, and financial advisors for their advice to clients — at least in today’s society. Beyond accountability, there are dimensions of trust that are harder to articulate but no less important. Compassion matters: it’s not just about the accuracy of a diagnosis or advice, but the sense that someone is genuinely invested in the outcome. As is relational trust, where a financial advisor who has managed your portfolio through two market crashes carries context and history that no new AI tool inherits.
AI can earn trust through performance, and in some domains, will surpass humans entirely. But performance alone is not a foundation. The dimensions AI cannot own are where durable businesses are built. Use AI to enhance your solution, not as the solution itself.