DATA STORIES | GENAI | KNIME ANALYTICS PLATFORM
Developing a KNIME Workflow with ChatGPT-4o
My experience with ChatGPT-4o in developing a workflow in KNIME to collect data from Google Analytics 4
Throughout my career, I have used various tools and technologies to facilitate and enhance the process of data collection and analysis. I have worked with ODI (Oracle Data Integrator) and PDI (Pentaho Data Integration) from Hitachi Vantara, and recently I had the opportunity to use KNIME. The project involved collecting data from Google Analytics 4 and using it for analysis. KNIME is an open source platform with a plethora of features, and despite being very intuitive, I used OpenAI ChatGPT-4o to help me configure the nodes to speed up the process.
Developing a Workflow in KNIME Using GPT-4o
KNIME is a highly flexible data analysis platform that allows the integration of various data sources for advanced analysis. It offers a range of data manipulation features, facilitating data collection, transformation, and analysis. When I started developing my workflow to collect data from Google Analytics 4, everything went as expected. Using an interactive mode connection, where my email and password were sufficient, I was able to configure the workflow without significant difficulties.
However, this type of connection did not work to process the workflow in a non-graphical environment, and the Google API access keys were necessary.
The connection to the service itself was simple; the problems began when the GA4 property code was not passed to the next node. That’s when I decided to turn to ChatGPT-4o for information on what might be wrong.
The tool was very efficient in explaining how to configure the nodes, offering clear and precise guidance. However, some unexpected difficulties arose when dealing with the APIs to be enabled in Google.
Seeking a Solution with a “Know-It-All”
When I described my situation to GPT-4o, the initial response was to recommend enabling the Google Analytics Data API, which I had already done. Although this recommendation was technically correct, it did not work in practice. GPT-4o even provided a Python code to test the connection (unsolicited), but there was a problem: the code required the manual insertion of the Google Analytics property ID, whereas in KNIME, this ID should be obtained automatically, and it had not understood this.
GPT-4o insisted that the Google Analytics Data API was sufficient but did not mention the need also to enable the Google Analytics Admin API. I did not know this was necessary, especially since the interactive mode connection had worked without this service enabled. This crucial detail only came to light when I reformulated my question, explaining that the property identifier must be obtained dynamically at the time of connection. It was then that GPT-4o suggested enabling both services, allowing the connection to be successfully established.
This insistence by GPT-4o on a solution that did not fully meet my needs became somewhat annoying. To make matters worse, it tried to guess my next question and provided detailed explanations about configurations irrelevant to my specific problem. Even when informed of this, it continued to provide these lengthy explanations, which increased frustration. Imagine talking to someone who gives you several useless explanations at the wrong time.
To find the correct solution, I had to adjust my approach. By reformulating the question and emphasizing that I did not have the property identifier and that it needed to be obtained dynamically, GPT-4o finally suggested enabling the Google Analytics Admin API in addition to the Google Analytics Data API. This suggestion resolved the issue, and the connection was successfully established.
In the old Google and Bing style, the word “dynamically” was the keyword to find the appropriate solution.
In the End, a Touch of Human Creativity
This experience demonstrated the importance of the human role in problem-solving, even when using advanced tools like GPT-4o. Artificial intelligence has its limitations and may not be able to identify context changes or specific nuances of a problem. It was necessary for me, as the user, to identify the flaw in the path suggested by GPT-4o and reformulate the approach to reach the correct solution.
Furthermore, the experience was also marked by moments of frustration with GPT-4o’s insistence on providing irrelevant information. Despite its processing power and extensive knowledge, it could not automatically adjust its responses to meet my specific needs precisely. Sometimes, I had to reprimand it, asking it to respond only to the formulated questions. This worked temporarily, but soon returned to insisting on lengthy, useless explanations.
In the end, after resolving the issues and getting the process running, the situation became quite funny.
I reflected on the evolution of technology and the ongoing need for human intervention to guide artificial intelligence in the right direction. I remembered the 1980s and 1990s when the only recourse was to turn to extensive books to find solutions.
Conclusion
My experience with ChatGPT-4o in developing a workflow in KNIME to collect data from Google Analytics 4 showed that AI is extremely useful in many aspects. It simplifies the development process and saves time searching for information. However, it also revealed significant limitations, highlighting the importance of human creativity and judgment in problem-solving.
GPT-4o, despite its advanced capabilities, still depends on human intervention to identify alternative paths and propose effective solutions. It behaved like a “know-it-all” in many moments, trying to guess my questions and providing detailed explanations on topics I had not asked about. This underscored the need for a critical and adaptive approach when using artificial intelligence as technical support.
Ultimately, the interaction between humans and AI can be highly productive, provided there is a clear understanding of the limitations and capabilities of both parties. Human creativity and judgment remain essential to navigating challenges and finding innovative solutions. As I concluded in a previous article, GPT-4o’s support is valuable, but human intervention remains indispensable to achieve effective results.
Note. ChatGPT-4o supported the text from a draft I wrote and then had various parts rewritten. Naturally, the images were generated by ChatGPT-4o. The cover image was generated using Photoshop AI from a square image generated by ChatGPT-4o.
