At Stockholm Tech Show in May, Dr. David Barnes delivered the opening keynote. Barnes is a Brigadier General (Ret.) and former Chief AI Ethics Officer of the US Army, where he developed the Army's Responsible AI Strategy. Today he advises leadership teams on organizational readiness for AI, one step upstream from implementation itself.
He put his finger on the human factor needed when AI begins to function as a decision machine. Human oversight is part of the answer. But oversight carries only as far as safety reaches, because an employee speaks up first when it is safe to do so. That is why psychological safety belongs with AI, and that is the thread I follow further here.
Silence becomes a decision
His point landed at once: when people dare to speak against the machine, it weighs more than how good the system is. In its absence, the staff goes along. Quietly.
The question I carry is what happens in the room where everyone is silent.
Then the recommendation is realized on the silence. It passes because it meets agreement, and in an unsafe environment silence looks exactly like agreement. The system reads absent resistance as approval, and the proposal becomes a decision that a single person rarely stood behind.
In an unsafe environment, silence looks exactly like agreement.
Daring and being able
Interrupting rests on two things. That you dare, and that you can.
| Dare | Able |
|---|---|
Psychological safety. Speaking against the machine requires that it is safe to do so. This is the dimension Barnes raises from the stage. |
Cognitive capacity. Challenging an answer requires that the person holds their own picture of the problem. That is the level beneath, and where I want to look further. |
The ability to challenge an AI answer assumes that the person holds their own representation of the problem. When we let the system formulate the reasoning for us, the internal model thins out. It feels like efficiency, and for that reason it stays invisible. The support feels good until it disappears. Objecting requires something in hand.
Here the two threads meet. Safety and substance carry each other. The person who holds their own picture of the problem has more to stand on when they speak against it, and the person who feels safe dares to bring it out. Weaken one and the other erodes.
Barnes's test is sharp: ask an employee to explain why the AI is wrong in a specific case. It puts a finger on the gap. What I want to place beside it is the question of what keeps that ability alive over time, so that the test stays answerable a year from now.
Stopping and revising
My reflection holds a third step, and it is the hardest. Stopping is one thing. Revising is another. Stopping takes the courage to say no. Revising takes someone holding an alternative. Another reading, another path, a basis to correct against. An organization can have the safety to interrupt and still lack the ability to correct, because the substance to correct against has thinned out.
Governance as infrastructure
Dr. David Barnes during the opening keynote at Stockholm Tech Show. On the slide: Principles are not governance. Policies are not control.
Barnes gave the same idea words from the stage. One of his slides read Principles are not governance. Policies are not control, and at the bottom the line Prompts are not controls. Permissions are. It points to the difference between governance as permission and governance as infrastructure. Permission is granted in a meeting. Capacity is maintained in how the work is done every day. Control sits in permissions, and a permission carries only when the person who holds it both knows and dares to use it.
The question I take with me
If people are to both know and dare, what in the way we work keeps that ability alive, so that someone can stop the system when it is wrong and revise it once it is stopped?
Thank you for a keynote that gave me sharper words, Dr. Barnes. This is a thread I will gladly continue.
Katri Lindgren is founder and CEO of Erigo and works on cognitive integrity in the age of AI.
