Yesterday, August 29, I joined an AIFOD online session with a rather simple question behind it: How much AI is too much AI?
Six speakers from different areas of the AI ecosystem joined the discussion, looking at what happens to human creativity, judgement and responsibility as AI becomes increasingly involved in the work we create and the decisions we make.
For me, the question was not really about how much AI we can technically use. We already know very well that AI can use and orchestrate a lot. The more interesting question is what happens when we stop using AI as a tool and start allowing it to replace the human part of the process.
Human in the loop is not enough
One point I kept coming back to during the discussion was the difference between having a human in the loop and having a human who actually remains in control. These are not necessarily the same thing.
A person clicking “approve” on an AI-generated result does not automatically mean that meaningful human oversight exists. The person needs to understand what they are approving, be able to verify it, have enough expertise to recognize when something is wrong, and have the authority to reject or change the result. And the consequences matter.
If an AI system generates a ridiculous image of person with 3 arms for an entertainment project, we can laugh about it. If the same type of failure happens in medicine, law, finance, public administration or another high-consequence environment, the discussion changes completely. That is why I increasingly prefer to think about AI in the loop, rather than simply putting humans at the end of an automated chain and calling that governance.
The question behind the question
The session also explored something broader: what actually remains human when AI is involved in creation?
There is an obvious layer: the text, image, video, decision or other output produced with AI. But there is also everything underneath it: the human knowledge used to build the systems, the information they retrieve, the decisions made about their objectives and the people who ultimately accept responsibility for the result.
This connects with some of the questions I have been exploring since the AIFOD Geneva Summit 2026, where I spoke about AI visibility and sovereignty, and with the observations I later wrote down in 3 Things I Learned at AIFOD Geneva 2026.
The conversation is moving quickly as we are no longer asking only what AI can do, but we are starting to ask what humans should still do.
And that question becomes even more interesting when AI systems are allowed to operate continuously, remember previous interactions, learn from experience and develop relationships with other agents. That is one of the reasons I have started building G2V-3 as an independent experiment, not as a game, but as a way to explore what can happen when artificial cognitive agents are placed inside a persistent environment rather than simply asked to answer individual prompts.
The AIFOD discussion was only a short online session, one hour, but I think the question behind it is going to stay with us for a very long time.
How much AI is too much AI?
Maybe the answer is not a percentage. Maybe it is the point at which humans stop being capable of questioning what the AI is doing, and eventually stop being responsible for the outcome.
That is where “too much” begins.
