Our 17th Year – Prompts Gonna Prompt
Listen to this blog post:
Without much risk (because if you’re paying attention, AI slop is real), I recently turned to “Microsoft Copilot” to create an image that would support a cross examination in an ABA False Claims Act Mock Trial Institute presentation. I had very little time for the task, and accuracy was not as important as creating a “gist” of an idea (or thumbnail). The whole exercise filled a need, but also served to inform for future, more consequential and practical use of AI in trial communications.
First, I was heartened by the process for a number of reasons. The attorney had asked for a chart to show the relatively “small” percentage of patients in order to minimize the allegations of neglect in the mock exercise fictitious fact pattern. I shared a quick example while we were both traveling (and tending to real clients). He responded favorably because yes, this is exactly what he was thinking. So, in the words of the Staples media campaign, I thought “that was easy.”
It was easy, that is until I took a closer look at the image. I’ve been surprised that AI has such a difficult time with accuracy when numbers are involved. Sure enough, the AI generated image contained 80 icons instead of my prompt for 100 highlighting 6. It turned out to be 4.8% of the total icons, instead of 6%. Additionally, the consultant in me wondered if the six highlighted figures would achieve the exact opposite of what the attorney had set out to communicate. I suspected they would make a small percentage appear more significant, not less. So, we prompted for a larger sample (since the fact pattern was fictitious):
“Please generate a visual chart of 1000 icons indicating male and female people. Then color 60 of them a different color to highlight 6% of this population.”
Copilot generated:
Reasoning completed in 9 steps
I’ve built a clean “pictogram chart” with 1,000 alternating male/female icons and colored 60 of them in a distinct accent to represent 6%. You can use it as a standalone image or drop it right into your deck.
I immediately questioned the male/female shapes. Don’t they seem unrecognizable and wouldn’t they cause confusion to the viewer? I also questioned whether they effectively conveyed “a small, almost inconsequential number of patients.”
Even with the increased volume of icons, the original goal was left unmet.
This is a tiny example of the challenges we all need to address when AI comes into play, whether we appreciate it or not. With all the recent coverage of attorneys being sanctioned for the use of AI in their briefing (one example here: https://storage.courtlistener.com/recap/gov.uscourts.ca6.152857/gov.uscourts.ca6.152857.50.2.pdf), attorneys are well advised to review with extra care, any use of AI, especially those that are used to save time.
#happytrails #happytrials #dontgototrialwithoutus #podiumperfect
