Responsible Experimentation with Generative AI

Posted on November 8, 2023

A laptop computer sits on a desk. On the screen is a webpage with the headline "Introducting ChatGPT."

Today’s Morning Buzz is brought to you by Kayla Barber-Perrotta, Budget & Performance Manager at the City of Brighton, CO. Connect with Kayla on LinkedIn.

What I am Listening To: 1989 (Taylor’s Version)

What I’m Watching: Inventing Anna

What I’m Working On: Developing a Halloween decorating plan for our competition at work. 

I used ChatGPT for the first time recently. I know I am rather late to the bandwagon, and much of that was due to my own biases. The idea of using generative AI just seemed like it was cheating or like it was some type of loophole around real work. It wasn’t until a colleague presented it as a tool to support work, much like Google or Microsoft Word, that my curiosity got the best of me. 

 My first foray was a simple personal one. I am planning a trip to Italy and spent hours researching and meticulously drawing up a schedule for two weeks. I wondered how well AI could replicate that work, so I asked it to plan a two-week Italy trip using only public transit. Watching text scroll across the screen and produce in seconds a very similar itinerary to what I had spent weeks creating was fascinating and I was hooked. I couldn’t contain my excitement and quickly began asking others around me about whether they had tried ChatGPT themselves. Below are some of those uses:

  1. As a Virtual Editor: One of the most common starting points for my peers was in the realm of editing. This included inputting memos, reports, or even emails and asking ChatGPT to make corrections including fixing grammar, spelling, or even tone. I’m sure we all have written an email while upset. Ideally, we sit on that email, revisit, and rework once calmer, but there isn’t always time. Enter AI. Simply copy and paste the email into the chat and ask it to make the email more professional or to streamline it. If you are wordy like me, they can also be great for removing all the filler and getting straight to the point.
  2. Combat Writer’s Block: Ever had an impending deadline for a report and all you can do is rewrite the same three lines over and over again or spin out thoughts into increasingly pointless nonsense? We have all been there. Sometimes it is just hard to figure out where to start. AI can help with that. You can start from scratch such as asking it to develop a sample schedule for a two-week schedule for a pre-school summer camp or a draft generative AI policy for a local government entity. You can also enter a bulleted list of key points for your affordable housing strategy and ask it to develop a draft report. The key here is to use these drafts as a starting point to tailor to your voice, your community, and your needs.
  3. Find Themes: Another great use for ChatGPT is quickly finding themes. My team had a great time experimenting with this feature as some of our first forays into the use of AI in the workplace. Remember, AI is built on information that already exists and generative AI is taking the next step by taking that information and reforming it into something new. In this way, it excels at finding themes. We experimented with everything from writing job descriptions to developing sample job postings for the city. It was a fun way to synthesize information and compare our own work with what was happening elsewhere in the industry and gave us inspiration for reorganizing and rephrasing our own information. It was also great for summarizing our own data from various open-ended survey questions we’ve extended to employees and residents.

Building a Culture of Responsible Experimentation

I am lucky enough to be a part of an organization that welcomes the new and innovative. Our City Manager, City Attorney, and our Information Technology Department have all been very welcoming of employees experimenting with generative AI, but as with any emerging technology or approach, there are certain parameters that must be set in place to ensure safe experimentation. These include:

  • Recognizing that whatever generative AI produces is only as good as the data it is trained on. This is key when looking at your own community needs as the data may not reflect the unique identity of your residents. You must always review what is produced and course correct where needed. This is particularly poignant in the space of diversity, equity, and inclusion where research has shown many generative AIs including the popular ChatGPT have been found to convey stereotypes and misinformation as fact.
  • Understanding that producing something through AI does not release you from responsibility for that product. If anything, the use of generative AI means employees must be more accountable than ever for their work. In traditional research and writing, it can be assumed an employee is filtering as they go. They review sources, reference where required, and put their own meaning to the ideas they convey. Generative AI removes this from the drafting phase, and employees must now add that back in somewhere else. This means meticulously reviewing the work product for correctness and meaning, and if they don’t understand where something came from, researching it. The complexity of doing this when you cannot readily ask ChatGPT for sources means that employees will need to weigh when and where the benefits of generative AI usage outweigh the risks.
  • Finally, the use of AI requires transparency. This is still an emerging technology with a rapidly changing landscape, and as mentioned, significant pitfalls if used improperly. To ensure residents can continue to trust the information we as government employees produce it is important to disclose how and when generative AI is used in the development of a program, policy, or report, and how these pitfalls were mitigated. In the case of this article, I was having writer’s block. I entered several points around generative AI I wanted to convey and asked ChatGPT to provide me with an outline. I also used it to edit my final product before submitting it to ELGL for publication.
Close window