The various AI vendors want us to trust their autonomous agents. I have several problems I would love to use AI to solve. (Including updating all the tags on this site!) However, I cannot trust the agents yet. Even when I use interactive prompting, I'm not happy with the results. Those results can change (non-deterministic answers). Worse, when I catch the AI in a lie, it cannot explain its answers.
Why would I trust a lying, deceitful computer program? Especially in an organization where people prove their competence and trust every day?
One of my management clients, Joe, said that he trusts his various AI agents more than he trusts “his” people. I was surprised and told him so. Joe's a smart guy, so I asked him for an explanation.
He said, “I can trust that the agent will find the best possible information across a wide variety of sources.” (That wide variety of sources is a small-world network.)
I asked, “Can your people do that?”
He half-laughed. “No. They need to work alone, so I can see what each person does. Otherwise, how can I reward them?”
This is an example of resource-efficiency thinking applied to humans—and not to AI agents. Joe allows the AI agent much more autonomy than he allows the people he's supposed to lead and serve. He thinks that makes the agent more trustworthy.
Instead, I suggested we work on making “his” people more “efficient” by working in flow. That would allow them to trust each other and extend that trust to others in their small-world network.
Management Minute: Consider Who You Trust
You can increase everyone's effectiveness when you increase their autonomy. Autonomy is not about acting alone, but acting with agency. Some of that agency allows them to use their small-world networks to increase their access to other people's information. You can trust them more.
Sure, use your AI agents. (What could possibly go wrong?) But don't allow your AI agent more autonomy than you offer the people in your organization.
For more specifics, see:
- Agile and Lean Program Management for specific information about small-world networks
- Modern Management Made Easy books to see how flow efficiency thinking (and actions) support what managers want