What do you do if you want to leverage artificial intelligence and machine learning in public administration?

In my experience, most organizations have not yet solved their data problem. If you haven't organized and made available your data, you need to solve that first before you have any hope of using AI and ML to any useful degree. At NVIDIA GTC, Lowe's presented a session that talked about the importantance of data and giving their developers easy API access. Their devs are spending their times innovating instead of wrangling Salesforce queries. If you had a data problem before the AI revolution, you still have a data problem now. Solve that first. Walk before you run. It'll be worth it, trust.

- Identify key areas for AI/ML integration in public administration - Engage stakeholders to understand their needs and concerns - Assess current processes, data availability, and regulatory constraints - Prioritize areas with the highest potential impact on efficiency and citizen services - Conduct thorough research on AI/ML technologies and best practices - Develop a clear strategy for AI/ML implementation, including budgeting and timelines - Ensure transparency and accountability in AI-driven decision-making processes - Address ethical considerations and potential biases in AI algorithms - Provide training and support for staff to adapt to AI/ML technologies

To leverage AI and ML in public administration, begin by integrating big data analytics to inform policy-making, enabling evidence-based decisions and predictive modeling for future planning. Utilize chatbots powered by AI for public inquiries, providing a first line of response to efficiently handle common questions and requests, freeing up human resources for more complex tasks. This dual approach enhances operational efficiency, improves public engagement, and drives more informed, data-driven policy development, ensuring a modern, responsive, and effective public administration system.

Choosing the right model or algorithm is very important. Data Scientists and Machine Learning Engineers alike learn intuition on which are the best for the right kind of problem. There's a lot of test and learn. You have to be ok with pivotting. In the case of LLMs, nothing beats the reasoning capabilities of GPT-4 right now, but its slow and expensive. It might be right to use GPT-4 for a plan-making or am ambiguity checking agent. Using Llama or Mixtral for those won't have nearly as good results. Claude 3 might lecture you about good vs evil instead of doing what you asked it to do. Try different models, learn intuition, and jam on.

Implementing evalutions or "evals" is super important in your AI and ML applications. You need to set a standard for what success looks like and test against it. If you ever have to switch models, say GPT-4 to GPT-4 Turbo or even a 1.0 to 1.1, you have to re-evaluate every prompt against the new model, even for tiny model weight changes. Your prompt from 2 months ago might now work the same as today if the model you used changed in any way. Writing good evals lets you test your prompts and models are still working as intended and let you try out new models with peace of mind. Unfortunately, evals can be tough and expensive. But worth it for the stability.

To harness AI and ML in public administration, envision a digital ecosystem where smart algorithms predict infrastructure needs, preemptively optimizing resource allocation. Imagine AI-driven platforms that dynamically match citizens' skills with job openings, boosting employment rates. Deploy ML-enhanced systems to monitor environmental data, proposing actions to sustainably manage natural resources. Incorporate AI-powered assistants in educational platforms, offering personalized learning paths for students. This comprehensive integration not only streamlines operations but also fosters a proactive, adaptive public service environment, leading to a more engaged and thriving community.

While you are working through points 1-6 (not after the fact) ensure that communications counsel is at the table to help you navigate everything from avoiding bias to ethical implementation. Comms can ask the hard questions along the way that others may struggle to formulate or be timid to ask. And if you're building an internal LLM, an absent comms voice at the table could result in headlines you regret. End result: transparency and trust.