My professional role as faculty member in an institution of higher learning is increasingly focused on AI. I continually encounter it used incorrectly in student work. While the grammar and structure of many articles I peer review is improved, the quality of content and proportion of inaccurate information has increased tremendously. AI was a primary focus of two recent online conferences I attended - and I’m making notes for some future posts inspired by those sessions. Maybe in part due to feeling like impact use is ubiquitous, I have unfortunately found myself more often using “AI” than my former preferred, more precise description “artificial intelligence and other large language models (AI and other LLMs).” 
 
This bothers me for a couple of reasons. One is this suggests “AI” is an all powerful, multi-functional, phenomenon/entity, sort of the way “the internet” and “Google” have also come to be described in more recent years. I’ll accept “the internet” is broad and multi-functional but “Google” is just one of multiple search engines, and what “Google says,” is always in response to what was asked, and often the asker is a person. “I conducted a Google search and these are the results” is precise in a way that “Google says” is not. The tendency toward anthropomorphic identification may have its roots in Alexa and Siri – computer programs that their owners designed and marketed to be perceived as personal assistants.
 
As an aside, the fact that I do not regularly hear from my female colleagues some complaints about use of female voices – reinforcing the secretary or “girl Friday” stereotype – is something I still find disturbing. Customization aside, why is this almost always the default? Way back in the early AOL days, “You’ve got mail” was, I seem to recall, a male voice. Was this a problem? Once custom settings because available, I set mine up to use the late soul singer Barry White, which made opening my mail a happier experience.

But back to AI. The second reason I’m concerned about overuse of the generic identifier “A”" is that this can, and does refer to a huge range of things a LLM does, whether giving cooking directions, providing physical therapy advice, making “art” or “music” or “writing a story,” (my use of quotation marks is intentional), or acting as your long or short-term romantic partner.  

One of the things I learned in my recreational therapy coursework was how important it is to be able to perform task and activity analyses. Activity analysis was described in my coursework as the process of identifying the domains involved in a process or practice (or activity). Task analysis was described as a stepwise description of the things that need to be done to accomplish the practice, activity, or tasks within.
This is similar to the content/process approach to lesson planning, which is described nicely for recreational therapy contexts in Therapeutic recreation program planning by Sylvester, Voelkl, and Ellis (Sagamore publishing, ISBN: 9781892132208). This is an old book and I’m not getting anything for promoting it. It was one of my favorite texts, however, and I still refer back to the content/process directions when I’m working on programs.

My recent encounters with AI-created work have reinforced the importance of task analysis and revealed some potentially serious limitations with use of AI in scholarly writing and similar tasks which  require precise and detailed thinking.

I have had a primarily negative view of letters to the editor, commentary and opinion papers when published in scholarly journals.
Many of the letters to the editor I have seen are attempts to sneak in a brief research report that is not of the quality, or not written in a detailed enough way, to make it through peer review. The exceptions to this include some explicit criticisms (and author responses): these I often find informative and enlightening, although interestingly, these items sometimes appear as articles rather than letters.
My early exposures to commentary and/or opinion papers suggested these suffered from bias and lack of credible or scholarly support. I started to change my view during the time period when I served as Editor of the Ohio Journal of Public Health. The founding editorial board (which I was not part of) developed guidance for commentaries, as an article type, mandating these be solidly grounded in scholarly sources.
​
​

I started this blog to address two things: first was a class assignment to keep a research journal of some sort and the second was to chronicle my explorations with technology. And both of these aims focused on qualitative inquiry.
​
I don’t look for new technologies with the same energy I did back in the early days of this blog but now and then I encounter something new. Procreate is not a new app for me but I had my first ever training session last week and I’m sorry I waited so long to do this.

There are lots of photo editing programs with varying levels of complexity and price. Procreate charges a reasonable, one time fee for (what seems to be) perpetual use, although that could change at some point in the future. Mostly I’ve just wanted to be able to edit photos, and not necessarily in a realistic way. For example, I sketched over the grass in an outdoor pic of a bicycle to make it look like it was done with colored pencil.
​
I took several pictures of walls and tile a long time ago, with a plan for developing some image when I compared qualitative inquiry to mortar or grout. My effort above is very basic and was done in about 3 minutes. I didn’t use any of the fine tuning features available - this was done by adding a couple of layers, painting in the bluish color in one and erasing and adding wording in another. In addition to speed and simplicity, other advantages are that Procreate goes on my Ipad and works well with the Apple pencil. This means I can draw or sketch, or edit photos or files to create the images I have in my head. 

It is close to 15 years since I first read “Classifying the findings in qualitative studies” by Margarete Sandelowski and Julie Barroso (Qualitative Health Research, 2003; https://doi.org/10.1177/1049732303253488) and it remains one of the papers that had the most profound influence on me as a practitioner, instructor, and author of, and about qualitative research.

I go back to the typology which features in this paper over and over while engaging in a variety of active (doing research) and less active (reading or reviewing research) endeavors. For me, the critical shift comes between topical survey - a summary of what subjects were discussed, and conceptual / thematic description - an effort by analysts to describe how participants make meaning of those subjects in the contexts of their lives. This might alternately be described as the distinction between what something is, and what it means.

. . .is quickly losing its attractiveness as a place of abundant information. 
I chose this picture - taken about a decade ago when I was sharing a recycling area with less careful neighbors - with some care, as it demonstrates both abundance and crap.
​

I have to be honest and say I have been pretty blasé about research methods in the recent months although things are shifting around for me.
I’ve discovered and started working with a new method - new to me, anyway: “Ideal Type Analysis” (Stapley, O’Keeffe, & Midgley, APA published book ISBN: 978-1-4338-3453-0) that I plan to discuss one of these days in this space.
I attended the live TQR: The Qualitative Report 16th annual conference earlier this month - and this was my first time since COVID of attending an academic event.  It was refreshing to present in person and I got great feedback on my work - which I am working to develop into a manuscript for publication.
But the point of this brief post is to promote a freeware program I came upon when updating course content for a data analysis course. Taguette is freeware, with web-based or local download version and works on Mac, Win, or Linux. It also allows project sharing.

I was curious to see what results I would get when I entered an online course assignment prompt into ChatGPT. I used v 1.2024.268, free plan, for GPT-40 (until I ran out of access time) for the Mac OS.

The prompt I used directed ChatGPT to describe a qualitative research study on a topic. I specifically directed use of citations. Below is the original reference list provided by Chat (item 3 relates to the topic of the prompt). It 3 is a real paper, by the way, almost as if ChatGPT saw me coming:

    1.    Creswell, J. W. (2013). Qualitative Inquiry and Research Design: Choosing Among Five Approaches (3rd ed.). SAGE Publications.
    2.    Smith, J. A., Flowers, P., & Larkin, M. (2009). Interpretative Phenomenological Analysis: Theory, Method and Research. SAGE.
    3.    Gould, D., & Carson, S. (2008). Life skills development through sport: Current status and future directions. International Review of Sport and Exercise Psychology, 1(1), 58-78.
    4.    U.S. Department of Health & Human Services. (2018). Decision charts: 2018 Requirements. Retrieved from https://www.hhs.gov/ohrp/regulations-and-policy/decision-charts-2018/index.html.
    5.    Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology, 3(2), 77-101.
    6.    Patton, M. Q. (2002). Qualitative Research and Evaluation Methods (3rd ed.). SAGE Publications.

As an avid and experienced use of Interpretative Phenomenological Analysis, I recognized right away that this method was not appropriate (nor described in ChatGPT’s proposal). ChatGPT also did not associate the reference citations with the written text. I prompted it to do so, using what felt like clear language (“associate the citations with specific content”).  In response it modified the text with some trend information from a website and added that item to the list.  The citation in text had the little quotation mark hyperlinks ChatGPT uses, although it still neglected to associate any of the others. Then I started to get confrontational.
​  

Anyone else involved in research and/or education has probably been exposed to a lot of information about Artificial Intelligence aka AI.

Large language models, machine learning, and applications of algorithm to make sense of information are not new ideas or processes. Several summers ago I was involved in Twitter, now X, research where a team of analysts worked to train the coding models to distinguish illegal drug related-posts from posts containing other, sometimes similar language use. The big challenges then included creative use of language such as in humor and sarcasm, and errors including spelling and word use,  the latter consisting of malapropism (“for intensive purposes” rather than “for intents and purpose”), homonyms (“there” instead of “their”), and omissions, where inclusion or exclusion of negative (“not doing” something when the actual meaning is “doing” something).

I appreciate these models have come a long way in a few years although for many people the most profound change is availability of a user interface that allows use of commands expressed in plain language rather than computer code. I am of course talking about the free and not freely available products marketed as AI.

The inspiration for this current post came from what feels like a barrage of public relations and marketing communications efforts, directed at me and my colleagues, to convince us that we are falling hopelessly behind unless we not only allow but encourage students’ use of AI in classrooms.  

I’ve seen a more balanced approach in the communications regarding AI for applied research use although ironically I see more potential value in research applications than in student class assignments.

Similar to some appeals I’ve seen regarding social media, a big part of the "you must use AI" argument is that it is here and so nothing can be done about it. Consider if this was the attitude taken toward the invention of the motorized vehicle and so no one bothered to develop regulation, including traffic control. I’ve seen some old pictures of the chaotic pre-traffic light roads, where horses, pedestrians non motorized cycles and motorized vehicles moved, drove, and dodged each others. Nuclear weapons present another example where efforts to control use, following initial development and early uses, continue to be made. With respect to AI I don’t buy that just because something under that name is now widely available, we as a human race need to be prepared if not welcoming of any and all potential uses. 
​
For me, AI use falls primarily into two broad categories and there are benefits and perils with each.
​

I have been following the Guardian UK reports on the investigation into how bugs in an accounting system directly and indirectly led to charges, fines, jail time, and other life-changing outcomes for falsely charged sub postmasters. You can see a historical overview at the link below:

https://www.theguardian.com/business/2024/jan/07/what-is-the-post-office-horizon-it-scandal-all-about

What struck me about this story from the beginning was the illogic of the accusations. In fact, there is a solid and sensible reason to investigate an alternative causal chain - because the discovery of unprecedented financial shortages were entirely precipitated by instillation of a new software system. Thinking of Mill and causality (summarized here in very basic terms): cause precedes effect, effect does not occur without cause, competing causes can be eliminated* - the clear logical target of any investigation to me seems to be the software system rather than a  few thousand, often long term employees, who suddenly went rogue. Adding support to this is that there were recorded instances  of complaints or questions about system functionality.

My reason for posting about this in my research blog is it reminds me of a couple of conversations I had with others some years ago regarding DXA (bone density) scans. I have not had one of these myself for some time and I am also aware that the algorithms or predictive models have been updated over time. Early in my doctoral program (2010) there was interest in promoting calcium intake among women of Asian ancestry due to the impression that Asians were more prone to bone density concerns. I asked a couple of people if this was also evidenced by high rates of fractures and no one could offer convincing evidence that it was; my own casual research at that time also failed to reveal particularly concerning trends.

* https://www.britannica.com/topic/Mills-methods