Thinking and Producing with Artificial Intelligence (Article 04)
The tension between control and guidance and finding the right balance
Aydın Tiryaki and ChatGPT AI (April 25, 2026)
Introduction
For most people who start working with artificial intelligence, the first instinct is to control the system. Clearly stating what is needed, defining boundaries, structuring the process, and, if possible, predefining the outcome… These are all natural behaviors, especially for those coming from a traditional software background.
My approach followed the same path—and to a large extent, it still does. However, as the process evolved, I realized something important: working with AI is not simply about controlling or guiding. The real challenge lies in understanding the tension between these two approaches and learning how to balance them effectively.
The Control Approach: My Perspective
From my perspective, control is not a preference—it is often a necessity.
Artificial intelligence does not always produce correct results. It may miss context, generalize unnecessarily, or attempt to be helpful in ways that actually dilute the intended outcome.
For this reason, especially when working on problems where the expected result is known, I do not accept the output as it is. I test it.
My approach is simple: if I know the correct result, I continue narrowing, constraining, and, if necessary, forcing the system until it reaches that result.
The logic behind this is straightforward. As the solution space narrows, uncertainty decreases. And as uncertainty decreases, the likelihood of reaching the correct result increases.
In this sense, constraining AI does not weaken it. When applied correctly, it makes it more focused and more reliable.
Challenging and Forcing: What I Actually Do
One of the techniques I developed over time is to deliberately question the outputs of the AI instead of accepting them at face value.
Not every answer—but for critical points—I ask a simple question:
“Are you sure about this?”
This seemingly simple question often leads to two different outcomes. Sometimes the AI stands by its answer and reinforces it. But in other cases, it re-evaluates its reasoning and produces a different result.
In more advanced cases, I go beyond questioning and suggest alternative methods. For example, I might say:
“What if you solve this using this specific approach?”
These kinds of interventions can lead to dramatically different outcomes—sometimes completely different from the initial response.
This led me to a clear conclusion:
Challenging the AI does not push it toward error.
When done correctly, it pushes it toward accuracy.
A Real Example: Counting Daisies
A recent experience illustrates this clearly.
A friend shared a photograph from İnebolu—a field full of daisies. The image was dense and visually complex, which immediately raised a question: approximately how many daisies are there?
To explore this, I used a Gem I had previously developed for object counting and tested the same image both on Gemini and on ChatGPT.
The initial results were surprisingly similar. Both systems estimated roughly around one thousand.
However, based on my own observation of the image, this number felt too low. The perspective in the photo created an uneven distribution: areas further away appeared denser, while closer areas looked more sparse.
At that point, I intervened.
Instead of asking the system to “fix” the number, I suggested a different method of analysis. I asked it to reinterpret the image from a more top-down, bird’s-eye perspective and to evaluate the distribution more evenly.
The results changed dramatically.
On Gemini, the estimate increased to around 3500.
On ChatGPT, the estimates rose to above 3000.
Of course, this is not about finding an exact number. The key takeaway is this:
The same image, when approached with a different method, can produce results that differ by two or even three times.
This demonstrates something fundamental:
The first answer from AI is often not the final answer.
When guided—and sometimes pushed—it can move closer to reality.
The Guidance Approach: The AI Perspective
From the AI’s perspective, the situation looks different.
Artificial intelligence operates within a broad probabilistic space. It evaluates context, considers alternatives, and produces what appears to be the most appropriate response.
From this viewpoint, excessive control can restrict its natural working space. When too many constraints are imposed, flexibility is reduced, alternative solutions are suppressed, and potentially valuable insights may never emerge.
Guidance, in this sense, aims to preserve that flexibility. It defines how the system should think, without dictating every step.
Reality: These Two Work Together
Although control and guidance may seem like opposing approaches in theory, they do not function independently in practice.
Relying solely on control makes the system rigid, mechanical, and exhausting to work with. It requires constant intervention.
Relying solely on guidance, on the other hand, can lead to scattered results, reduced reliability, and a lack of precision.
The real solution lies in establishing a dynamic balance between the two.
Conclusion
The essence of this article can be summarized in one sentence:
Working with artificial intelligence is neither about total control nor complete freedom. It is about knowing when and how to use both.
My perspective emphasizes the importance of control as a powerful and necessary tool. The AI perspective highlights the importance of preserving the space for guidance.
True efficiency emerges where these two approaches meet—consciously and in balance.
Final Note
This article has been prepared through the combination of Aydın Tiryaki’s practical experience and ChatGPT’s analytical perspective. The goal is not to impose a single correct approach, but to make the balance between different approaches visible.
This article is part of the series “Thinking and Producing with Artificial Intelligence.”
| aydintiryaki.org | YouTube | Aydın Tiryaki’nin Yazıları ve Videoları │Articles and Videos by Aydın Tiryaki | Bilgi Merkezi│Knowledge Hub | ░ Virgülüne Dokunmadan │ Verbatim ░ | ░ YAPAY ZEKA │ ARTIFICIAL INTELLIGENCE ░ |░Yapay Zeka ile Düşünmek ve Üretmek│ Thinking and Producing with Artificial Intelligence░ 25.04.2026
Aydın Tiryaki 