AI Policy Making: Week 3 Summary
AI Policy Making
Week 3 Summary
Midjourney Prompt: a dairy cow, wearing gumboots, tending to their vegetable garden, holding a watering can
In week two I learnt a few tricks about how to get a bit more consistency from ChatGPT and mitigate the potential for disruptive glitches from temperamental third-party plugins. You can check out my summary of Week 2 here: AI Policy Making: Week 2 Summary.
Having completed all the high-level Bootleg Plan provisions down to the Objectives level, it was time to move on to the Policies. To test ChatGPT’s suggested methodology I asked if we should develop the maps and methods before the policies. ‘Nope’, was the answer. Policies, then methods, then maps - for iteration 1 at least:
By following this sequence, you ensure that each component of the plan is directly related and supportive of the one before it, forming a cohesive and comprehensive District Plan. Each step provides a foundation for the next, allowing for a logical progression of work. This also enables any adjustments or changes to be more easily identified and addressed as the plan develops.
Starting over (…Again)
Having hit the word limit for a second time in this project, I was grateful that I had been using OneNote to keep all my prompts and ChatGPT responses, each tabulated by Chapter and Topic. Of course, blogging about it as I went turned out to be super handy. Updating the Bootleg Kaipara District Plan on my website ‘live’ as it was developed proved to be invaluable. I was able to simply point ChatGPT to the link instead of having to re-explain where we are up to in the project during times of amnesia.
For each of the Policies I requested a "Value Proposition" (VP) - effectively the reason why the policy is needed. I did this to check the focus of the policies and what they are trying to achieve. I can also use the VP to refer back to in case ChatGPT gets lost. It occasionally still goes looking on the internet for the 'official' versions of things - which would ordinarily be a good thing if you were developing the plan for real!
Despite being a new Chat instance, it seemed to continue relatively problem-free with the approach of pasting the objectives and outline from earlier chat sessions. I did notice that the numbering started to change as the process wore on, but I didn't want to risk disturbing the flow by asking about it along the way!
It also stopped giving me the VPs when switching from 'Water' to the 'Air' topic, but again, I decided that rather than interrupting the flow I could ask about them later if needed. I also felt it was probably eating into the total conversation length (character limit 'tokens') elected to get through the policies in one session rather than risk having to start again.
Prompt engineering
I've found ChatGPT answers to be much more reliable if you provide previous answers (even from much earlier in the chat) rather than just continuing along with successive topics. This is a known phenomenon due to the way the LLMs work. The 'Tree of Thought' (ToT) prompting technique is the way to go for complex tasks and analysis as I'm doing with this project. You can read more about it here: Tree of Thoughts (ToT) | Prompt Engineering Guide (promptingguide.ai)
Section 32 Evaluation ‘on the fly’
By Day 17 we had finished with all the Policies. I was really digging some of the policy suggestions - particularly how to deal with climate change, economic, and community resilience. It's not something that you see a lot of in the current district plans, and I was super interested to see what Methods ChatGPT had in mind to give effect to them.
Responding to my first prompt about what methods could be employed to achieve the first policy on the list, ChatGPT 'helpfully' defined what methods were in the RMA context, giving examples. I then pasted s32 into the chat and asked which methods were recommended having regard to s32.
Again, the first answer was an explanation of how to do a s32 analysis and suggested that I seek the advice of a planning professional … "Thanks for that ChatGPT. I know what it is - now I want you to do it for the policies you are suggesting."
For each Policy ChatGPT provided a paragraph to explain:
Relevance
Effectiveness
Efficiency
Clarity
Risk of acting or not acting
Coherence with other objectives and policies
It also recognised that while looking okay on its own, it would need to be assessed again against specific local context, implementation methods, and interaction with other policies. Exactly. Nice answer!
Going through a preliminary s32 for each policy through Days 17 & 18 was useful quality control for the policies, and gave me material I could refer ChatGPT back to when moving on to methods (or if it gets lost). We were smashing out a LOT of words, which I was sure glad I didn't have to sit down and write myself! By the end of Day 19, this was getting fairly tedious and the evaluation at a stand-alone policy level was not adding a lot of value at this point - particularly since we would need to do it again once the methods were in place.
Show Us the Rules! (Methods)
Day 22 - Fast Forward to looking at Rules. Much excite! But the RMA states that the District Plan may contain rules ("if any") so I was careful not to be presumptuous, and elected to use the language "methods".
Let's now start to develop some methods to give effect to the Policies we have developed. Starting with the Natural Environment Chapter, District Level Policy 1 relating to land. What rules (if any) are required to achieve this: "Policy 1: Encourage sustainable land use practices that minimize soil degradation and maintain soil health."
The methods proposed were fairly familiar, although there was not a lot of detail. But that is to be expected, as I asked for the methods, and the summaries contained just enough detail to see where they were headed:
1. Soil Conservation Rules: Develop rules that outline specific requirements and practices for soil conservation, such as erosion control measures, appropriate land management techniques, and sustainable farming practices. These rules can promote responsible land use and minimize soil degradation.
2. Nutrient Management Rules: Establish rules to regulate the application of fertilizers and other soil amendments to prevent nutrient runoff and soil pollution. These rules can include guidelines on nutrient application rates, timing, and methods, ensuring that soil health is maintained while minimizing environmental impacts.
3. Erosion Control Rules: Implement rules that address soil erosion by specifying erosion control measures, such as contour plowing, terracing, or the establishment of vegetative buffers. These rules can help prevent soil erosion, which is a major cause of soil degradation.
4. Land Disturbance Rules: Introduce rules to regulate land disturbance activities, such as construction, excavation, and earthworks, to minimize soil compaction and disturbance. These rules can include requirements for site preparation, sediment control, and soil stabilization to protect soil health.
5. Soil Monitoring and Assessment Rules: Establish rules for soil monitoring and assessment to ensure ongoing monitoring of soil health and quality. These rules can include requirements for soil testing, analysis, and reporting, allowing for the identification of potential issues and the implementation of remedial measures.
It is important to note that the specific rules and their details would need to be developed based on the local context, scientific knowledge, and stakeholder input. The rules should be practical, enforceable, and support the objective of encouraging sustainable land use practices while minimizing soil degradation and maintaining soil health.
So I decided to dig a little deeper:
Should we go into further detail on each of those, or should we develop a similar list for every objective on this sub-topic? We may be able to design a suite of rules that achieve multiple policies I expect.
The approach suggested by ChatGPT was consistent with the way I would generally tackle the task. Rather than going into too much detail on each rule, we would first develop a more comprehensive suite of rules that collectively address the objectives for each topic:
By adopting a holistic approach, we can identify common themes or overlapping requirements among the objectives and develop rules that effectively address those shared aspects. This can help ensure consistency, efficiency, and clarity in the implementation of policies while minimizing redundancy.
Well said!
Chapter Topics
For each of the Chapter Topics, ChatGPT provided an outline of the rules required to achieve the outcomes sought through the objectives and policies. Once we have done this for all topics, we can then get into more specific detail such as RMA classifications and district plan standards.
Natural Environment
On the topic of 'Land', many of the suggested regulatory methods were ones I expected to see, such as soil disturbance, erosion control, and preventing soil contamination. Some interesting additional ones were also proposed, such as 'sustainable farming practices' (such as crop rotation for soil health) and soil testing and monitoring regimes.
Similarly, on the topic of 'Water' the regulatory methods also included both expected and unexpected provisions:
1. District Level:
- Rule 1: Establish water allocation limits and permits to ensure sustainable extraction and use of water resources.
- Rule 2: Implement water conservation measures for all water users, including agriculture, residential, and industrial sectors.
2. Zone Level:
- Rule 3: Require water-efficient practices and technologies in agricultural, residential, and industrial zones.
- Rule 4: Establish buffer zones and setback requirements to protect wetlands and riparian areas from land-use activities.
3. Area Level:
- Rule 5: Set water quality standards and monitoring requirements for freshwater bodies to ensure compliance with established standards.
- Rule 6: Develop and enforce regulations for floodplain management in areas prone to flooding, including land use restrictions and infrastructure requirements.
I noted some of the provisions I would typically expect to see in a Regional Plan and asked ChatGPT if there are any limitations on a District Council from having rules such as water allocation. ChatGPT recognised that regional councils are typically the regulatory authorities for water-related activities, but noted that district councils can contribute to water management through initiatives such as water conservation programs, education and promotion of water-efficient practices, and incorporating water-sensitive design principles in land use planning.
On the Topic of 'Air' there was a much greater emphasis on emissions than is typically seen in district plans. The regulations suggested were:
District Level:
1. Air Quality Monitoring and Reporting.
2. Emission Standards for Industrial Activities and Transport.
3. Promotion of Practices to Reduce Greenhouse Gas Emissions.
4. Land Use Planning for Air Quality Protection.
Zone Level:
1. Best Practices for Air Pollution Control in Industrial Zones.
2. Promotion of Sustainable Transport Practices in Urban Zones.
3. Promotion of Clean Energy Sources in All Zoning Categories.
4. Regulation of Open Burning of Waste.
Area Level:
1. Air Quality Standards for Areas of Natural Beauty or Recreational Value.
2. Air Quality Management in Areas with High Traffic Volumes or Industrial Activity.
3. Protection of Air Quality in Areas of Significance to Māori.
On the Topic of 'Biodiversity', some of the more interesting regulatory methods were:
Community Involvement in Biodiversity Conservation
Promotion of Sustainable Farming Practices in Agricultural and Rural Zones
Urban Biodiversity Initiatives in Urban Zones
Habitat Management and Restoration Plans
Climate Change Adaptation and Resilience was also a common theme in the regulatory methods, as was community stewardship and involvement.
BUILT ENVIRONMENT CHAPTER
I was looking forward to seeing what provisions ChatGPT came up with for the Built Environment Chapter - and I was not disappointed. On the Topic of Infrastructure, the District Level regulations sought to encourage infrastructure development that supports sustainable growth, balancing the economic, social, environmental and cultural values. Infrastructure resilience and innovation (such as the use of renewable energy sources, low-impact design, and smart technologies:
District Level Regulations:
1. Balanced Infrastructure Development.
2. Infrastructure Standards and Resilience.
3. Innovative and Sustainable Solutions.
Zone Level Regulations:
1. Coordinated Land Use and Infrastructure Planning.
2. Mitigation of Adverse Impacts.
3. Integration of Infrastructure and Land Use.
Area Level Regulations:
1. Area-Specific Infrastructure Planning.
2. Community Engagement in Infrastructure Decision-making.
3. Resilience of Infrastructure to Hazards and Climate Change.
The Transportation Topic was a mixture of the expected and unexpected regulatory methods. Comprehensive, safe, accessible, efficient and sustainable were all familiar terms, as was the integration of transport with land use planning. The more interesting methods were:
Multimodal Transportation Options.
Customised Transportation Planning.
Community Engagement in Decision-making.
Resilient Transportation Infrastructure.
The Waste Topic also had a focus on reducing, reusing, and recycling, but with specific reference to the implementation of circular economy principles. Community participation and partnership initiatives feature prominently, as social buy-in would be necessary to give effect to methods such as waste segregation at source.
All in all, I have been inspired by some of the less traditional regulatory methods that ChatGPT has come up with, and am excited to see further details and the mechanics of the rules themselves. But first, we have got to unpack the methods for a few more Chapters. Next up are Cultural Heritage, Hazards and Risks, and the most anticipated for me … the Economic Development Chapter.
Let's go!