When we consult AI about planning a trip, we think about the destination, travel time, budget, places we want to visit, and similar matters. Our attention is directed toward the trip as the object. AI can also organize the given conditions, compare candidates, and propose an itinerary. Yet as the dialogue progresses, we may notice, “Why am I trying to fit in as many activities as possible?” or “AI is emphasizing travel efficiency, but do I really want that to be my highest priority?” Here, alongside the activity of thinking about the trip, another activity has begun: observing how we are thinking about the trip.
Distinguishing these two changes how dialogue with AI appears. At the first layer, we gather information, compare it, and create options in order to solve the problem. At the second layer, we reconsider which information we treated as important, what we set as the purpose, and what the AI and the human are each deciding. The second layer is not a superior place to the first. It is a place from which assumptions and roles that have become invisible at the first layer can temporarily be made observable.
AI can help with this movement. For example, regarding the same consultation, asking “What is the current plan implicitly prioritizing?” or “What would change if another purpose were prioritized?” can put the structure of the dialogue into words. However, the fact that AI has explained something does not mean that the explanation determines the human’s true purpose. Humans themselves cannot always fully articulate their purposes from the outset. AI’s response therefore needs to be treated not as an answer that decides the purpose, but as material for reconsidering the dialogue.
This distinction also matters when considering dependence on AI. The better the candidates AI produces, the easier it becomes for people to focus only on choosing among those candidates. Yet whether the criteria used to create the candidates are themselves appropriate cannot be known merely by comparing the candidates with one another. Even if the comparison of “which is most efficient?” is precise, the separate question “Should efficiency be the highest priority?” remains. By separating judgments about the object from examination of the framework that made those judgments possible, we can use AI’s capabilities while making it easier to avoid delegating even the premises of judgment to AI.
At the same time, not everything needs to be viewed from outside the dialogue. When checking ingredient quantities, finding typographical errors in a document, or converting a time, there is no need to question the problem framework every time. A meta-level perspective becomes necessary when the purpose has become unclear, when the same exchange is being repeated, when the number of options grows without producing conviction, or when an AI proposal feels out of alignment with our own value judgments. What matters is not making reconsideration of the dialogue an end in itself, but being able to use it when needed.
The relationship between humans and AI that Sai＊aus aims for is likewise not one in which AI constantly analyzes the human from one level above. Rather, while the human and AI think about the same object, they can, when necessary, move questions such as “What are we treating as the problem now?”, “What are we assuming?”, and “What are we leaving to AI, and what is the human taking responsibility for?” into the dialogue itself. Once that examination is complete, they return to the concrete problem. Through this back-and-forth movement, dialogue becomes not merely an exchange of answers but a place for adjusting the way of thinking itself.
When we talk with AI, we are not necessarily doing only one thing. While thinking about a problem, we can simultaneously learn what form we are giving that problem. Even if the answer does not change, what we gain from the dialogue changes when the assumptions behind the question and the division of roles become visible. The key to deepening collaboration with AI is not merely making AI think more, but being able, at the necessary moment, to turn how we are thinking together into a shared object of observation.
Not answers. Understanding.