A few weeks ago, while writing an automation script for a friend’s project, I had just decided to use systemd timer instead of a cron job. It got me thinking: What’s the state of our trust in artificial intelligence (AI) technologies? Even in my own workflow, I’d find myself questioning, “How accurate is this result?” while simultaneously asking, “How much longer would my work take without this tool?” This dilemma is what pushed me to write this post. The fact that AI’s usage rate has climbed to 84% despite a drop in trust in it as a technology to 29% is a serious contradiction that warrants reflection. This situation highlights our approach to technology and the role of pragmatism.
Today, news headlines about AI typically promise either great excitement or deep concern. On one hand, we’re told AI will simplify our lives and boost productivity; on the other, scenarios of job loss, ethical dilemmas, and even loss of control are discussed. This oscillation between two extremes makes it difficult for the average user to form a clear picture. Consequently, while many of us don’t find AI fully trustworthy, it’s hardly possible to distance ourselves from these technologies that have become a part of our lives.
Why is Our Trust in AI Low But Our Usage High?
The primary reason behind this situation is that AI is still in its early stages and cannot fully deliver on all its promises. Especially the recent “hallucination” cases, where AI generates completely fabricated or false information, play a leading role in this erosion of trust. Imagine a scenario where, while working on a production ERP, an AI-based planning module generates incorrect supply chain data. This can lead to serious operational disruptions, far beyond a simple error.
Let me give you a real-world example: Last month, while creating content drafts for my own blog, I noticed that an AI tool I was using generated a completely fabricated CVE (Common Vulnerabilities and Exposures) number. There was no such thing as “CVE-2026-31431,” but the tool presented it as if it were real. Situations like these cause users to have a suspicion towards AI, thinking, “Could this information also be wrong?” This suspicion directly reflects in trust ratings. Yet, the interesting part is that despite this distrust, we continue to use AI tools.
Another dimension of this contradiction is the speed at which AI is integrating into our workflows. In many fields, AI tools offer the potential to automate tasks, reduce repetitive work, or perform analyses that were previously impossible. This potential saves users time and increases productivity. Therefore, even if we are not entirely sure about the accuracy of the information AI generates, we continue to use it for the speed and convenience it offers. It’s a kind of “it saves me time, I’ll handle the rest” approach.
The Reality of “Hallucination”: AI’s Biggest Weakness
AI “hallucinating,” meaning generating non-existent, fabricated information, is one of the most fundamental problems of current AI models. This is a common occurrence, especially in language models (LLMs). As models try to mimic patterns in their training data, they sometimes misinterpret these patterns or combine them with incomplete information, producing inconsistent and incorrect outputs. This situation carries significant risks, especially in technical fields.
For instance, imagine you are a system administrator trying to optimize a systemd unit file on a Linux server and you ask AI for help. AI might suggest a cgroup setting that looks valid but could actually cause errors on the system. Indeed, in a previous project, while adjusting memory limits for a systemd service, an AI-suggested memory.high setting caused services to be unexpectedly oom-killed. While not a direct hallucination, this indicated that AI didn’t fully grasp the context and offered a flawed optimization.
This “hallucination” issue arises not only in technical fields but also in general information generation, translation, or summarization. An AI-generated summary distorting the main idea of the original text or making a wrong emphasis can lead to the user being misinformed. This, in turn, damages trust in AI and necessitates users to always verify information.
graph TD; A["User: I need information"] --> B["AI Model (Training Data + Algorithm)"]; B --> C{"Generate Output"}; C --> D{"Is Information Real? (Accuracy Check)"}; D -- No --> E["Hallucination! False Information"]; D -- Yes --> F["Correct Information"]; E --> G["Loss of Trust"]; F --> H["Increase in Trust"]; G --> I["User Needs Verification (More Time)"]; I --> J["Pragmatic User: Continues to Use for Time Savings"]; J --> B; G --> K["User: I don't trust AI, but I have to"]; K --> B; As seen in this diagram, the accuracy of the information generated by AI is a critical threshold. When the accuracy check fails, hallucination occurs, leading to a loss of trust. However, pragmatic users continue to use it for benefits like time savings.
Pragmatism: Tools We Use Even If We Don’t Trust Them
So, why are we so heavily drawn to a technology we’re not sure we trust? The answer is simple: Pragmatism. In today’s world, time is one of the most valuable resources, and AI tools help us use this time more efficiently. Based on my many years of experience in the tech world, I can say that a tool doesn’t need to be perfect; it just needs to be good enough and get the job done.
Let me give another example: While doing SEO optimization for my blog, I use AI-powered tools for processes like keyword research and content draft generation. Of course, I don’t accept everything these tools produce without question. I check the relevance of the generated keywords and rewrite the content drafts, blending them with my own knowledge. However, without these tools, I can’t even imagine how much longer these processes would take. Perhaps the time I spend on a blog post would double.
This situation isn’t limited to blogging. Similar pragmatism is seen in corporate software development processes. For instance, using AI-assisted tools for code completion or debugging in production ERP development can speed up development. Of course, additional tests are needed to ensure the code suggested by AI is secure and efficient. However, this extra effort might take less time than writing our code from scratch.
This pragmatic approach also accelerates the adaptation process to technology. People prefer to use the aspects of a technology that make their jobs easier, without feeling the need to learn all its details or fully trust it. This, in turn, facilitates the adoption of rapidly evolving technologies like AI.
The Future of AI: Building Trust or Optimizing Usage?
When discussing the future of AI, two main scenarios emerge: First, AI models becoming more accurate, transparent, and trustworthy; second, users learning to use AI more intelligently, compensating for its flaws. Looking at current trends, I believe the second scenario is more likely.
Without fundamental changes to the core architectures of current AI models, it seems difficult to completely eliminate the “hallucination” problem. Therefore, it’s becoming essential for users to learn to work more effectively with AI. This means critically evaluating information from AI, always performing additional verification, and viewing AI solely as a “tool.” In my own projects, I use AI like an assistant; it provides ideas, generates drafts, but I always make the final decision and verify the accuracy of the information.
At this point, skills like prompt engineering are becoming even more important. Asking the right questions allows us to get the information we want from AI more clearly. Furthermore, techniques like Retrieval-Augmented Generation (RAG) improve accuracy by allowing AI to access more up-to-date and reliable information sources. When I tested RAG implementations in my own systems, I found the AI’s responses to be more contextual and less erroneous.
I believe that in the future, AI will have a longer “lifespan,” and we will learn to “befriend” it better. This doesn’t mean AI will completely change us, but rather that we will guide AI better. Perhaps in the future, our trust in AI won’t reach a full 100%, but its usage rate will increase even further. Because AI, when used correctly, becomes an indispensable assistant.
Why Do We “Not Trust” But “Use”?
In conclusion, the reason behind the high usage of AI despite low trust is the pragmatism in our approach to technology. Although AI is not yet perfect, the efficiency gains, time savings, and new capabilities it offers make it indispensable in many areas. This situation is a complex reflection of humanity’s relationship with technology. We always have some degree of skepticism towards new technologies, but when the benefits outweigh it, we set aside these doubts and continue to use them.
Based on my own experiences, I find that learning how to work better with a technology, rather than expecting it to be perfectly trustworthy, is often a more efficient path. AI falls into this category. We know its tendency to “hallucinate,” but as we learn how to verify this information or how to guide AI more accurately, we can work with it more effectively. It’s akin to smartly using a flawed but powerful tool.
Ultimately, for trust in AI to increase further, models need to become more transparent and accurate. However, in this process, we users can also maximize the potential offered by technology by learning to use AI more consciously and intelligently. The high usage despite the lack of trust is an indicator of how central AI’s role will be in the future. Understanding this technology and learning to work in harmony with it will be critically important for all of us in the coming years.