During a press conference on February 12, 2002, i.e. during the months leading up to the Iraq War, then US Secretary of Defense Donald Rumsfeld gave remarks in a response to questions about evidence for weapons of mass destruction.
Reports that say that something hasn’t happened are always interesting to me, because as we know, there are known knowns; there are things we know we know. We also know there are known unknowns; that is to say we know there are some things we do not know. But there are also unknown unknowns—the ones we don’t know we don’t know. And if one looks throughout the history of our country and other free countries, it is the latter category that tends to be the difficult ones.
Wikipedia article There are Unknown Unknowns, retrieved May 15, 2026.
The phrasing in those remarks was both amusing and insightful. I don’t love that this is the context in which I learned about the concept and that I will probably forever associate it with that. But, here we are.
For better or worse, we were left with what The Decision Book reasonably refers to as the Rumsfeld Matrix.
It looks kind of like the following.
This is a useful model, depending on context, with clear applications in risk management and decision making, for example. It also adds well to understanding of and reasoning with another model, the circle of competence.
Here is how that model is typically visualized.
The more expertise you have in a given area, the more you understand it. You also tend to have a clearer image of the edges of your understanding, i.e. the things you realize you do not (yet) grasp.
Everyone faces unknown unknowns. Especially in the context of the circle of competence, those unknowns should not be misunderstood as limited to unknown to everyone, but are often just unknown to you, i.e. the person or group that the circle of competence is about. Learning more can lead to answers and better understanding, thus turning unknowns into knowns, as well as clearer awareness of what you do not know.
A person who has attained expertise and is operating within that circle of competence, has fewer unknown unknowns that are of that personal nature. Or put another way, a person acting outside of their area of expertise has more things, where they do not know what they do not know, where an expert would. This is the difference between a novice (or complete outsider) and an expert.
In recent years, advanced artificial intelligence tools are increasingly enabling even relative beginners to produce impressive-looking results in a variety of domains.
If you have made a career in software engineering, then you may at this point be tempted to say (or grumble) a few things about people with little to no experience developing software, who are being convinced that they can and should vibe code potentially complex software systems by themselves. I do not think it is categorically a terrible idea – always depending on context, et cetera, of course – but the failure cases of this are easy to imagine and documented examples are not difficult to find online.
I won’t go there specifically. However, the direction is correct and by stopping earlier, we can see that the risk is actually larger and worse. Vibe coding is an application of generative AI. It is the latter that I do want to comment on. The excellent recent essay Appearing Productive in The Workplace describes it well:
Generative AI can produce work that looks expert without being expert, and the failure arrives in two shapes. The first is when novices in a field are able to produce work that resembles what their seniors produce, faster or more advanced than their judgment. The second is when people generate artifacts in disciplines they were never trained in.
This is a decoupling of output from competence, i.e. the person generating the output created something that they are in fact unable to understand in depth, since they lack the appropriate training, so would not have been able to produce the output without outsourcing the majority of effort to the AI.
Generative AI, as a field, provides increasingly powerful tools. An expert (operating in their circle of competence) can use a tool to much better effect than a beginner (or person fully outside their circle of competence). That is the value of experience, of competence in a field, rather than just of tool usage: The expert can better judge the quality of the output.
Those generative AI tools tend to not get stumped. Given a question, they will usually produce answers. You get to determine tone, style and formatting. The resulting answers, whether brief paragraphs or full “research” reports may look substantial and the reasoning register as seeming plausible.
Of course, there is an important difference between something being grammatically correct and appearing plausible – and it also being well-reasoned and factually sound. When the latter matters, you need to be able to tell and correct as needed. You have to be able to verify the generated output.
That is near impossible for the person, who is using the tool in an area fully outside their area of expertise. Worse, their (personal) unknown unknowns make it harder for them to see that as the problem it is.