Learning Objectives

After completing this lecture, you will be able to:

• Understand the key concepts of the ai decision framework: when to use and when not to in a government context

• Connect the ai decision framework: when to use and when not to to your agency's AI initiatives

• Identify next steps for applying these concepts in your role

Key Topics Covered

• Decision tree for evaluating whether AI is appropriate for a task
• Risk factors, data sensitivity, citizen impact, and approval requirements
• Government context for the ai decision framework: when to use and when not to
• Practical applications and next steps

Why This Matters for Government

Government agencies face unique challenges when it comes to AI adoption. This lecture addresses these challenges head-on by providing all government employees with the knowledge and frameworks needed to navigate AI in the public sector responsibly and effectively.

As part of the L1 (AI Aware) curriculum, this lecture builds on the foundational principle that every AI system in government ultimately serves citizens. Whether you are working with AI tools daily or setting strategy for your agency, understanding the ai decision framework: when to use and when not to is essential for responsible, effective government AI adoption.

Lecture URL: https://skill.re/learn/govt/the-ai-decision-framework-when-to-use-and-when-not-to.php

======================================================================

TRANSCRIPT: The AI Decision Framework: When to Use and When Not To

======================================================================

What you will learn: Decision tree for evaluating whether AI is appropriate for a task. Risk factors, data sensitivity, citizen impact, performance requirements, timing constraints.

Here's a scenario you might encounter: Your agency has a process that's slow, frustrating, or expensive. Someone says, "We should use AI to fix this." Everyone nods. The idea feels modern, efficient, technological.

But "we should use AI" is not automatically the right answer. Sometimes AI is the perfect tool. Sometimes it's a terrible idea. Sometimes it's okay but not worth the implementation cost. The question is: how do you decide?

That's what we're going to do in this lecture. We're going to build a decision framework that helps you evaluate whether AI is actually appropriate for a particular task. This framework considers risk, data quality, citizen impact, and practical constraints. By the end, you'll be able to walk into a meeting and make an informed recommendation about whether to pursue an AI initiative.

WHY THIS MATTERS FOR GOVERNMENT

Government agencies are under pressure to modernize. Budgets are constrained. Backlogs are deep. Someone asks: "Can't AI help?"

The problem is: not always. And sometimes the answer is "yes, but the costs and risks outweigh the benefits."

This matters because:

• Bad AI decisions waste resources. Building, maintaining, and ultimately replacing a poorly-chosen AI system costs time and money that could be spent elsewhere.

• Bad AI decisions damage trust. If an AI system makes wrong decisions that affect citizens, and people find out that the decision to use AI was not well-considered, public trust in government AI (and government generally) declines.

• Bad AI decisions can cause harm. An AI system that systematically disadvantages certain groups, or that makes errors in high-stakes situations, can cause real harm to real people.

• Good AI decisions are invisible. When you choose AI wisely—when you use it where it's appropriate and avoid it where it's not—the benefits compound quietly. Decisions improve. Processes become more efficient. Citizens get better service.

Having a decision framework helps you make good AI decisions.

THE AI DECISION FRAMEWORK

Here's a framework to help you think through this. Walk through each section. If you hit a "no" or "high risk" that you can't mitigate, the answer is probably "don't use AI" or "reconsider this approach."

Question 1: Is the Problem Well-Defined and Data-Driven?

AI systems need a clear problem to solve and data to learn from. If your process is ambiguous or judgment-heavy, AI won't help.

Examples of well-defined problems:

• "Identify which benefit applications have high risk of fraud" (clear outcome, data available)

• "Predict which neighborhoods have highest infrastructure maintenance need" (clear outcome, data available)

• "Summarize a long policy document into key points" (clear outcome, can be evaluated)

Examples of poorly-defined problems:

• "Improve service delivery" (too vague)

• "Make hiring better" (too broad, deeply subjective)

• "Streamline everything" (not specific enough)

Also consider: Do you have quality data? AI requires data to learn from. If you have only a few hundred examples, or if your data is full of errors, AI won't work well.

Question 2: What's the Citizen Impact? Is It High-Stakes or Low-Stakes?

This is critical. High-stakes decisions require more caution.

High-stakes AI decisions:

• Determining eligibility for vital services (food, housing, healthcare, income support)

• Criminal justice applications (who to investigate, predict dangerousness, bail decisions)

• Employment decisions

• Immigration decisions

• Anything that affects freedom or access to essentials

Low-stakes AI decisions:

• Suggesting which form someone might need

• Summarizing documents

• Categorizing incoming inquiries

• Finding best times for service appointments

For high-stakes decisions, you need:

• Higher accuracy requirements

• More extensive testing and validation

• More human oversight

• More transparency and explainability

• Clear fallback and appeal processes

For low-stakes decisions, you can move faster with lower accuracy requirements.

Question 3: Data Sensitivity - What Type of Data Will You Use?

Consider what data the AI system needs:

If it needs classified information: Stop. Don't do it (unless you have a classified AI system, which is rare).

If it needs PII or PHI: Proceed with extreme caution. You need legal authority, security review, privacy office approval, strong data handling, and security measures.

If it needs CUI: Moderate caution. Legal authority, security review, data minimization.

If it needs only unclassified data: Lower caution, but still maintain good security.

If it needs no personal data at all: Lowest caution.

The more sensitive the data required, the higher the bar for proceeding.

Question 4: Performance Requirements - What's the Acceptable Error Rate?

If your AI system has to be 99.9% accurate to be useful, you might not be able to achieve that. If you need 85% accuracy and the system can achieve 90%, great.

Think about:

• What's the cost of a false positive (system incorrectly flags something)?

• What's the cost of a false negative (system misses something)?

• Can humans catch errors the AI makes?

For benefits determination: If the AI denies someone a benefit they should receive, they go without food or housing. That's a severe cost for a false positive. You need high accuracy.

For summarizing documents: If the AI leaves out one detail, a human can catch it. The cost is lower. You can tolerate lower accuracy.

Question 5: Can You Explain Why the System Makes Decisions?

This matters for high-stakes decisions. If the system denies someone a benefit, can you tell them why? "The AI decided you don't qualify" is not acceptable. "The AI identified these factors as indicating you're outside the eligibility range" is better.

Some AI systems are more explainable than others. Decision trees and linear models are transparent. Deep neural networks can be opaque.

For high-stakes decisions, you should prefer more explainable systems (even if they're less accurate) over black-box systems (even if they're more accurate).

Question 6: What's Your Mitigation for the Risks You've Identified?

By now, you've probably identified some risks (data sensitivity, accuracy requirements, explainability concerns, etc.). Can you mitigate them?

Examples of mitigation:

• Implement strong security for sensitive data

• Require human review before high-stakes decisions

• Test for bias and fairness

• Provide transparency and explanation to affected people

• Build in fallback and appeal processes

• Monitor performance over time

If you can't mitigate the risks, don't proceed.

Question 7: Is There an Existing Solution That's Better?

Maybe you don't need AI. Maybe a traditional software solution is cheaper and more appropriate. Or maybe the right answer is to hire more people, or to redesign the process entirely.

Before you commit to AI, ask: What are the alternatives? What are their pros and cons compared to AI?

Sometimes AI wins that comparison. Sometimes it doesn't.

PRACTICAL EXAMPLES OF THE FRAMEWORK

Example 1: Should We Use AI for Hiring?

Problem: Our hiring process is slow. We get 1,000 applications per position. Reviewing them all takes months.

Apply the framework:

Q1: Is the problem well-defined and data-driven?

Answer: Somewhat. We have years of hiring data. But "good candidate" is subjective. Hiring success depends on many factors we don't measure well.

Risk: Moderate

Q2: High-stakes or low-stakes?

Answer: High-stakes. This determines access to government employment.

Risk: High

Q3: Data sensitivity?

Answer: We'd use educational background, work history, maybe demographic information.

Risk: Moderate

Q4: Accuracy requirements?

Answer: Very high. We can't afford to systematically exclude qualified candidates.

Risk: High

Q5: Explainability?

Answer: We need to be able to tell candidates why they were not selected.

Risk: High

Q6: Can we mitigate?

Mitigation: Require human review of every candidate the AI recommends. Test AI for bias before deployment. Provide explanation to candidates.

But: The AI accuracy requirements and human review requirements together might make this not much faster than manual review.

Q7: Alternatives?

Alternative 1: Use traditional software (resume database, keyword search) to initial-screen, then human review

Alternative 2: Hire a recruiter to do initial screening

Alternative 3: Use a simpler AI (rank candidates by keywords) rather than a complex model

Recommendation: Don't use a complex AI system for hiring. Use a simple tool to help humans with initial screening. Keep humans in control.

Example 2: Should We Use AI for Document Summarization?

Problem: We receive thousands of citizen letters per year. Staff is spending days summarizing them for leadership briefing.

Q1: Is the problem well-defined and data-driven?

Answer: Very well. Summarization is a clear task. We have many examples.

Risk: Low

Q2: High-stakes or low-stakes?

Answer: Low-stakes. A summary that's incomplete or slightly inaccurate won't cause harm.

Risk: Low

Q3: Data sensitivity?

Answer: Some letters contain personal information. But we're only using the summaries, not storing the original data in the AI system.

Risk: Low

Q4: Accuracy requirements?

Answer: Moderate. The summary needs to hit main points, but doesn't need to be perfect.

Risk: Low

Q5: Explainability?

Answer: Not critical for a summary. We can explain how the AI works (looked at key phrases, extracted main points) but don't need to explain every word choice.

Risk: Low

Q6: Can we mitigate?

Mitigation: Have a human review summaries before they go to leadership. Don't automate the summaries without human review.

Risk: Low

Q7: Alternatives?

Alternatives: Hire someone to summarize, use templates to guide summarization, don't summarize at all and send full letters to leadership.

Recommendation: AI for document summarization is a good fit here. Use an approved AI system. Have humans review summaries. Deploy.

ANTI-PATTERNS / MISUSE RISKS

Anti-Pattern 1: "AI Sounds Good, So Let's Do It"

An organization falls in love with the idea of AI and skips careful evaluation. "AI is innovative" becomes the only rationale for adopting it.

The risk: They build an AI system for a task where it's not appropriate. They waste resources. They cause harm.

Anti-Pattern 2: Ignoring the Data Sensitivity / High-Stakes Combination

An organization decides to use AI for a high-stakes decision (like benefit determination) without carefully considering the data sensitivity and accuracy requirements.

The risk: They deploy a system that doesn't work well enough, making wrong decisions that affect people's lives.

Anti-Pattern 3: Assuming AI Will Be More Accurate Than It Is

An organization assumes an AI system will be more accurate than it actually is, and designs the process assuming AI makes final decisions.

The risk: The AI makes errors. Those errors affect people. The organization faces liability and loss of trust.

Anti-Pattern 4: Not Considering Alternatives

An organization fixates on "should we use AI?" and never asks "what's the best solution to this problem?" Maybe the answer is something other than AI.

The risk: They pursue a complex, expensive AI solution when a simpler solution would be better.

PRACTICE / REFLECTION PROMPTS

• Think of a process in your agency that's slow or inefficient. Would AI be a good solution? Apply the decision framework. What's your recommendation?

• Think of an AI system you know about (in your agency or elsewhere). Evaluate it using this framework. Does it seem like a good fit for the task? If you were asked to defend the decision to use AI for that task, what would you say?

• What's the most high-stakes decision your agency makes? Should AI be involved in that decision? Why or why not?

• In your agency, who makes the decision about whether to pursue an AI solution? Is this person familiar with the factors discussed in this framework?

KEY TAKEAWAYS

• Not every problem is an AI problem. Evaluate carefully before committing to an AI approach.

• High-stakes decisions require much higher scrutiny. The more a decision affects citizens, the more caution you should exercise.

• Data sensitivity and accuracy requirements are tightly linked. The more sensitive the data, the higher your accuracy requirements, and the more you need human oversight.

• Explainability matters for high-stakes decisions. You should be able to explain why the system made a decision to the people affected by it.

• Always consider alternatives. Sometimes the best solution is not AI.

• Risk mitigation is critical. Identify the risks and make sure you can mitigate them before you proceed.

• A well-decided "no" is better than a poorly-decided "yes". If you can't clearly articulate why AI is appropriate for a task, don't do it.

TERMS / GLOSSARY ITEMS

High-Stakes Decision: A decision that significantly affects a person's rights, freedom, access to essential services, or opportunities.

False Positive: An error where the system incorrectly identifies or predicts something (incorrectly flags someone as high-risk).

False Negative: An error where the system fails to identify something (fails to identify someone who actually is high-risk).

Data Sensitivity: The degree of harm that would result from unauthorized disclosure of data.

Explainability: The ability to understand and explain why an AI system made a particular decision.

Mitigation: Actions taken to reduce or eliminate a risk.

Well-Defined Problem: A problem with a clear goal, measurable outcomes, and available data.

Your agency receives thousands of inquiries per year. Many are routine (checking status of an application, general information about eligibility, etc.). Your agency spends significant time responding to routine inquiries.

Your leadership asks: Should we use AI to automate responses to routine inquiries?

Apply the framework:

Q1: Well-defined problem and data-driven?

Answer: Yes. You have years of inquiry data. You can classify routine vs. complex. You have example responses.

Risk: Low

Q2: High-stakes or low-stakes?

Answer: Depends. If the AI is just providing general information, low-stakes. If it's making determinations about eligibility or benefits, high-stakes.

Recommendation: Use AI only for information provision, not for determinations. For determinations, route to humans.

Risk: Moderate (if carefully scoped)

Q3: Data sensitivity?

Answer: Some inquiries contain personal information. Design the system to not store PII. Just use the inquiries to categorize and route to appropriate responses.

Risk: Low-moderate

Q4: Accuracy requirements?

Answer: Moderate. If the AI routes to the wrong response template, a human has to fix it. That's annoying but not a disaster.

Risk: Low

Q5: Explainability?

Answer: Low importance for routine information. We can explain how the system works.

Risk: Low

Q6: Mitigation?

Mitigation: Have humans review automated responses before they go to citizens. Build in escalation pathway for uncertain cases.

Risk: Low

Q7: Alternatives?

Alternatives: Hire more staff, redesign inquiry process to be self-service, improve FAQ, don't respond to routine inquiries at all.

Recommendation: AI for routine inquiry response is appropriate. Use it to augment human staff (not replace them). Have humans review responses. Monitor performance.

15 minutes.

Think about an AI initiative your agency is considering (or one you know about). Answer these questions:

• What problem is it trying to solve?

• How high-stakes is that problem?

• What data does it need?

• What accuracy does it need to achieve?

• Can failures in the AI system be caught and corrected by humans?

• What are the risks if the system is wrong?

• Can those risks be mitigated?

• What would the alternative (non-AI solution) be?

• Is AI actually better than the alternative?

• Would you recommend proceeding with this AI initiative? Why or why not?

You now have a framework for making good AI decisions. Use it. When your agency is considering an AI initiative, apply this framework. Push back if the decision seems premature or insufficiently considered. Ask the hard questions.

That's how you ensure that AI initiatives in your agency are thoughtful, well-designed, and appropriate.

Thank you.

Government AI CLUB Certification Program

Level 1: AI Aware | Government AI Policy Landscape | Lecture 2.5

A GOVT.CLUB initiative.

<- 1.2.5 Understanding AI Bias 1.2.7 Reporting AI Concerns in Your Agency ->

Start Your CLUB Certification

This lecture is part of L1: AI Aware—8 hours of comprehensive government AI training.

Explore CLUB Certification

Related Lectures

L1 1.2.1—Government AI Policy Landscape 20 min - Video + Reading

L1 1.2.2—Data Sensitivity and Classification 15 min - Video + Checklist

L1 1.2.3—PII and AI: The Bright Red Lines 15 min - Video + Scenarios

Frequently Asked Questions

What will I learn in The AI Decision Framework: When to Use and When Not To?

In this 20 min interactive tool lecture, you will Decision tree for evaluating whether AI is appropriate for a task. Risk factors, data sensitivity, citizen impact, and approval requirements

What level is The AI Decision Framework: When to Use and When Not To?

This is a Level 1 (AI Aware) lecture, part of Chapter 1.2 \u2014 Responsible AI Use. It is designed for all government employees.

How long is lecture 1.2.6?

Lecture 1.2.6 (The AI Decision Framework: When to Use and When Not To) takes 20 min. It is delivered as a interactive tool format.

Do I need prerequisites for The AI Decision Framework: When to Use and When Not To?

This lecture is part of L1 (AI Aware). Prerequisites: None.

What is the CLUB Certification?

CLUB (Community Leading Unified Benchmarks) is a maturity-based AI certification for government professionals with 5 levels (L1-L5), 215 lectures, and 25 chapters aligned with NIST AI RMF, OMB, and GAO frameworks.