We Take Our Whiskey Neat: Key Takeaways of AI Projects in Government

This is a conversation between Jim Johnson (Jim), and Hans Mulder (Hans) on the state of Artificial Intelligence Projects (AIP).

Hans: So, Jim did you look through the CHAOS Database for AIPs and see the results.

Jim: The CHAOS Database of The Standish Group has 50,000 completed projects from 2010 to 2020. There are only a handful that have AI components. In addition, I looked at the some of the source and saw AIPs that were not complete, therefore they were not coded or submitted to the CHAOS Database. However, the few I did find were in two major segments, general customer call-in centers and health care. The call-in centers were completed, but mostly challenged, but in use. None of the healthcare were complete. I concluded that there was not enough observations to create a finding. What did you find, Hans?

Hans: As promised, I found on the internet by ChatGPT 4o, 100 AI-projects in Government (excluding the well-known AI-projects dr. Watson of IBM and Tesla car from the private sector/industry). My findings are 1) Most AI-projects are in the experimental stage, so it is hard to adjudicate success, challenged or failure up to this moment: 2) Many of AI-project just started last of this year; 3) The AI-projects are actually part of a program or portfolio of projects and not stand alone projects: When we apply the Standish adjudication process on AI-projects we see that the kind of project approach is more (Infinite) Flow rather than Agile or Waterfall because AI-projects have different characteristics such as the productivity and swiftness of delivery of a team of 7 human actors is different from a team powered by AI-systems and only 3 or 4 human actors. You bought a Tesla, what was your experience?

Jim: I took my Testa Model 3 in 2018 with the full self-driving experience. It was considered Beta at this time. However, I immediately tired in out. I live on Cape Cod, and satellite communication can be somewhat challenging. I found that off-Cape it worked reasonable well, but on-Cape strange things happen. A few months ago, Tesla released the production version, and it now work fine. Of course, we developed our own AIP for looking at the results of database for prediction analysis. It was compiled by our two developers using JAVA and MySQL over a couple years, but not full-time. We did not have a budget, timeframe or scope it could be classified as either challenged or successful. So, Hans what else did you learn.

Hans: I feed these observations and the 100 AI-projects in Government back into ChatGPT and it pointed out the integration of AI within government sectors is an exciting frontier, showcasing a spectrum of initiatives aimed at enhancing public services and operational efficiencies. An analysis of 100 AI projects currently underway reveals several intriguing insights about the state and nature of AI adoption in the public sector. A significant observation from these projects is that most are still in the experimental phase. This stage of development means it's too early to definitively classify many of these projects as successes, challenges, or failures. The experimental status underlines a broader trend of governments testing the waters with AI technologies, learning from pilot projects, and gradually scaling successful initiatives. This cautious approach is understandable given the public sector's need to ensure ethical considerations, data privacy, and public trust.

Jim: When did these projects start?

Hans: Many AI projects within governments have only started in the last year. This recent surge indicates a growing recognition of AI's potential to revolutionize public administration and service delivery. The fresh inception of these projects also suggests that governments are still in the early stages of understanding and harnessing AI's capabilities, which may involve significant learning curves and adaptation periods.

Jim: Where these Integration part of larger programs?

Hans: yes, another notable trend is that these AI projects are often not standalone efforts but are embedded within broader programs or portfolios. This integration reflects a strategic approach to AI adoption, where AI initiatives are part of a more extensive digital transformation agenda. Embedding AI projects in broader programs can enhance coherence, ensure alignment with organizational goals, and facilitate better resource allocation.

Jim: What was their Project Management Approaches

Hans: Interestingly, when we apply the Standish adjudication process to these AI projects, we observe that their management style leans more towards a "Flow" approach rather than traditional Agile or Waterfall methodologies. This shift is largely due to the unique characteristics of AI projects, where the productivity and speed of delivery from a team augmented by AI systems differ significantly from those involving only human actors. Teams working on AI projects often consist of fewer human actors, with AI systems taking on roles that require high-speed data processing and analysis, thereby necessitating a flexible and continuous flow of work rather than rigid project phases.

Jim: Hans: what is your conclusion

Hans: I think the journey of integrating AI into government functions is just beginning, with many projects still in their nascent stages. The insights from these 100 AI projects underscore the importance of flexibility, strategic integration, and a willingness to experiment and learn. As these projects mature, they will provide invaluable lessons for scaling AI in ways that genuinely enhance public sector effectiveness and citizen services.

“We Take Our Whiskey Neat” is a series of on watered down conversations between Hans Mulder and Jim Johnson on Artificial Intelligence.

---