Facilitate to create codings based on AI search - suggestions

[AndrzejWawa]

After the first experience I have a couple of suggestions:

1. there should be checkbox or drop-list, by default unselected, but allowing for select options:

• automatically code fragments found by AI,
• automatically add explanation of AI to memo for created coding.

2. Moreover, when you have list of fragments found by AI, there should be also:

• right-click options to code fragment by specific one or many codes at once, e.g. you righ-click on the option, then window with list of codes appears, and you can select one or more codes,
• the same option but with adding explanation of AI to memo for this coding or codings (when coded by more than 1 code),
• possibiliy to select more than one fragment or select all fragments, with these right-click options

[AndrzejWawa]

3. What is realted to the above, there could be option, e.g. in settings, on minimal coverage of text selection of AI. I found that AI often found relevant information but usually select only small part of text. Depending on the research needs it could be ok, or too small. So, maybe an option regarding selection coverage:

• automatically by AI (by default),
• minimal coverage in numbers of signs,
• maximal coverage in numbers of signs,
• or both.

[kaixxx]

Thank you for your suggestions. With regards to a (semi) automated coding, your first suggestion: Could you explain a litte bit more what your usecase for that would be?
I'll have to think about how to implement this. One non trival question for instance is when to stop with the coding. The AI will always try to suggest new data whenever possible (it's trained to be a helpful assistant afterall). It is not easy to find a reliable and repeatable cutoff point for the search. If you want to do a quantitative analysis later, this will be a problem since the number of codings is not a very reliable measure. If you do a qualitative analysis, I'm not sure about the benefit of a fully automated coding since you want to look at the original data anyways. Or not?

Whith regards to the length of the quote: Yes, they are often too short. I'll try to tweak the promt for GPT-4 to nudge it into selecting longer quotes. We'll see how it goes...

[AndrzejWawa]

Hmmm, ok. So, maybe only point 2 - right click-options to code one or many AI-found fragments by one or more codes at once. Also, with possibility to add AI explanations to memos for earch coding.

4. I have also idea that maybe there would be an option to ask AI to inspect only selected fragments? I think, that sometimes there is no need to analyse whole test, but e.g. only fragment we have problem with. If AI select fragments based only on the fragment or fragment in the context of the whole document?

5. And question: is it explained why AI finds certain number of fragments and not more/less, but we can ask to "find more"? If the first step of search differs from the second? (e.g. during the first only directly-related fragments are chosen, during the second indirectly related)? Maybe, there would be added some explanations in QC manual on how AI select fragments and how can be used (e.g. as an assistance, you mentioned)? For example, I wondered how many times I should ask AI to check the text again, if there were any findings?

[AndrzejWawa]

6. I have also an idea that:
   now you can conduct AI search basing on names of codes and memos. Maybe there could be possibility to take into account already coded text segments? Then, you can code some text manually to provide example for AI and later, allow AI to code basing on this example? Coded segments could be also used for machine learning.

[kaixxx]

Just a small update: I have now included the option to add the interpretation of the AI to the coding, as you suggested. Also, the longer quotes have been implemented.
However, there are still many other interesting ideas in these posts that I may revisit later. Thank you, @AndrzejWawa