1 of 32

Leveraging Generative AI in Local Government��RAISE AI Solutions Consortium Workshop

2 of 32

Agenda

  • Background
  • Innovation
  • Data Analysis Example
  • Custom GPT Demo
  • Use Cases
  • Potential Use Cases
  • Q&A

3 of 32

Background

4 of 32

Quick Bio

  • My senior year of high school I wrote a semester term paper on Artificial Intelligence
  • My background is in Mathematics and Computer Science and I have 5 patents/co-patents. 3 with AMD on Error Detection and Correction chips and 2 with UAA on Mars Rover Image Compression in conjunction with NASA JPL
  • I’ve done academic research at UAA using Genetic Algorithms (a form of Machine Learning) – Leveraging the UAF supercomputer and GPUs
  • Chief Innovation Officer at the Municipality of Anchorage with an Innovation Team (i-team)
  • Co-captain of Code for Anchorage – a volunteer Civic Tech group

​

​

5 of 32

Why does Government have unique opportunities?

  • Data is king for LLMs – Government has open data and public policies.
  • Government deals with a lot of unstructured data.
  • Gov. people like to help people. If something works in their city they want to share it with other cities.
  • Best part of Gov. is you have to help everyone. Hardest part is you have to help everyone.
  • Gov. has a lot of technical terms and lots of complicated processes and procedures that are ripe for summarization and simpler language.
  • The Municipality of Anchorage (MOA) has internal guidelines on AI and is moving cautiously. I am personally experimenting to learn about AI capabilities and the MOA isn’t using these experiments.

​

​

6 of 32

Innovation

How quickly can you learn from failures

7 of 32

Innovation – Learning from trying things

  • We don’t like to question the Status Quo
  • People who can ask why we do we do something? – Kids, people from a different field, new people and now Generative AI.
  • If you have an idea, you want to be able to test the idea quickly. If you have to put in a request to IT and then wait 3 weeks to get the answer. Which often leads to a slightly different question you want to test…
  • You need to examine and double check the results from Generative AI but you can rapidly test lots of theories. And if something seems fruitful then you can do a more detailed analysis.

​

​

8 of 32

What I learned from Genetic Algorithms

  • Genetic Algorithms use evolution and survival of the fittest for possible solutions.
  • You have a population of possible solutions at random and a fitness function. (Lossy compress an image and decompress it and compare the images)
  • Any means necessary – we are all going to be rich.
  • Trained models often can quickly adapt to new image types.
  • Why is an evolved solution better? – Explainable AI

​

​

9 of 32

10 of 32

Google DeepMind AlphaEvolve

11 of 32

Data Analysis Example

12 of 32

13 of 32

14 of 32

15 of 32

16 of 32

17 of 32

Custom GPT Demo

18 of 32

Custom GPTs – Lessons Learned

  • Provide real world advice from the permitting center.
  • City code depends on State code which depends on Federal code.
  • Larger context windows for data that is spread out in the document.
  • Better logic.

​

19 of 32

Use Cases

20 of 32

Use Cases

    • Rewriting and simplifying info at an 8th grade level
    • Easier translation from 8th grade level language
    • Document drafting in complex formats
    • Getting started
    • Changes in legislation summaries
    • Custom GPT for legislative packages
    • Brainstorm crazy ideas and have a group critique and jump off from there
    • Natural language interface for queries and data visualization
    • Compelling graphics and videos
    • ADA descriptions of website images
    • Greely, Colorado – pothole detection and reporting

​

​

21 of 32

https://dutchcyclinglifestyle.com/

22 of 32

Potential Use Cases

23 of 32

Potential Use Cases

    • Generating automatic color commentary for Assembly Meetings
    • Better OCR of filled out paper forms by training on all previous versions of the forms – UK Planning
    • More compelling employee training
    • Custom learning for areas you want to brush up on
    • Vibe coding one off or niche tools
    • Converting legacy computer code to a new language
    • Refactoring Municipal code
    • Figuring out the minimal changes needed for a Municipal code change
    • Agents
    • Image analysis for 311 and Asset Management

​

24 of 32

Time Travelling GeoGuessr

25 of 32

26 of 32

27 of 32

311 Help

28 of 32

29 of 32

30 of 32

Vibe Coded Air Theremin

Learning what hand tracking can run in a browser w/o internet

Temp URL - https://air-theremin-brendanbabb.replit.app/

31 of 32

Questions & Answers

32 of 32

Thank you

Brendan Babb

brendan.babb@anchorageak.gov

​