I’m trying to use the KNIME AI Assistant. It is evident that this extension is in its early stages. During my usage, I have some ideas for your reference. One point I thought of this morning is context.
In the “build” workflow construction feature, our ultimate expectation is to construct the corresponding workflow by stating our requirements. However, this step is very challenging in practice. Similar to text programming, AI provides us with a rough framework, but there are often small errors that require us to make modifications. In other words, if the task given to the AI is clear and specific enough, it should be able to assist in completing the task quickly.
In a large analysis task, if we can assign a small and specific task to AI, then it is possible to enter the practical stage in the future. How can we assign a small and specific task to AI? It requires AI to understand the context. In KNIME, the context should be the node we currently select!
Imagine this scenario: we select the current node and then make our requirements known to AI. AI will then construct the corresponding part of the workflow after this node. Imagine that we select a portion of the workflow, and AI can optimize this part of the workflow.
Similarly, in the Q/A function, we can ask AI to help us gain insights into the data. For example, if we select a node and ask a question, the content ultimately fed to the AI API would be a combination of the question and a portion of the data, like this:
give me some feature engineering ideas about the data:
rowID, col1, col2, col3 1, 2, 3, 4 5, 6, 7, 8
This is what I want to emphasize about context. Of course, it is evident that using AI in KNIME’s production environment still has a long way to go. However, I believe that integrating context is a crucial step.