Five stages of AI in EMS that every leader should understand

This year, I was sitting with two of my colleagues at an American Association of Critical Care Physicians conference, and the conversation turned to vendors developing unmanned ambulances.

At first we laughed. Then we paused. Because we all realized something at the same time.

This is no longer science fiction.

|Watch now: The Future of the FDNY: AI, BWC, Compensation Equal to Secretary Lillian Bonsignore

Artificial intelligence is already shaping dispatch systems, documentation, scheduling, and clinical decision support. And as with any new technology EMS has ever faced, from radios to cardiac monitors to RSI to electronic PCR, we have two choices.

We can understand it…or we can be dragged along by it.

Leading your agency responsibly requires a simple way to think about AI maturity. One of the most useful frameworks is the five stages of AI development. This helps leaders understand what AI can do now, what will happen next, and how to use AI safely.

Let’s explain it in EMS language.

Stage 1: Rule-based AI (protocol automation)

This is where most EMS systems are already installed.

Rule-based AI follows programmed logic. It doesn’t learn. That’s not logical. Just do it.

Examples of EMS include:

• CAD triage script
• Drug interaction warning
• Protocol checklist
• Edit billing
• Phone triage decision tree

Consider the cardiac arrest checklist in CAD. It reminds the crew of the steps but does not adapt.

• Strength: Rules-based AI is reliable and auditable.
• Weaknesses: Unable to adapt to the complexity of the real world.

This is the same type of automated EMS that was employed decades ago using protocol cards and dispatch algorithms.

Stage 2: AI learning (predictive analytics)

This is what today’s progressive EMS systems should be.

Machine learning systems analyze large datasets to identify patterns.

Examples include:

• Predicting call volume by time and location
• Anticipate staffing needs
• Identify people who make frequent, high-risk phone calls
• Billing error detection
• predict Hospital unloading delays

According to research Predictive deployment, when implemented properly, can improve response times by 5-15%.

• Strength: Learning AI allows you to deploy units better, staff smarter, and plan budgets more accurately.
• Weaknesses: Works only within one task. It doesn’t “think”. Recognize patterns.

Stage 3: Context-Aware AI (Assistance AI)

This is the stage we are in now.

Generative and context-aware AI systems can process language, reports, and complex information to assist humans.

EMS example:

• Drafting PCR narratives from monitor data
• QA review summary
• Creating a training simulation
• Provide a description of the protocol
• Medical Director Case Review Support

Imagine a new EMT asking, “Why do you give nitro for pulmonary edema?” The AI ​​system instantly pulls out physiology, protocol guidance, and takeaways.

That is support AI. It’s like having a knowledgeable instructor. Although helpful, it still requires the supervision of a senior physician.

• Strength: Context-aware AI improves education and workflow.
• Weaknesses: Requires human supervision.

Stage 4: Autonomous AI (Operational Decision System)

This is where the conversation about unmanned ambulances starts to make sense.

Autonomous AI can plan and execute multi-step decisions with minimal oversight.

In the future, the following EMS applications are possible.

• Real-time system implementation coordination
• Automatic schedule and overtime management
• Supply chain optimization
• Automate inter-facility transfer coordination
• Clinical triage routing

Imagine an AI system monitoring call demand, weather, staffing levels, hospital conditions and automatically redeploying units.

This could significantly improve response reliability in rural systems.

But serious questions arise, such as:

• Who is responsible for decisions?
• How do you audit errors?
• What if the AI ​​is wrong?

EMS has always operated under the direction of a physician. AI needs to operate under governance.

• Strength: Autonomous AI has the potential to improve system efficiency, deployment, and resource management at a scale that humans cannot achieve in real time.
• Weaknesses: Liability, safety, and oversight issues can arise when systems make operational decisions independently.

Stage 5: Artificial General Intelligence (AGI)

We have not yet reached this stage.

AGI will be able to reason across multiple domains, just like human clinicians and executives.
Theoretically it is possible to:

• Diagnose complex cases
• EMS system design
• write protocol
• Agency management
• teach leadership

But today we are not there. And responsible leaders need to distinguish between hype and reality.

• Strength: AGI may eventually integrate clinical reasoning, operational planning, and leadership support into a single system.
• Weaknesses: No one fully understands the risks, governance challenges, or unintended consequences of systems operating with such levels of intelligence.

Why AI innovation matters to EMS leaders

All major innovations in EMS have been met with resistance.

• heart monitor
• electronic PCR
• RSI
• telemedicine

Still, systems that adapted early have improved patient care and workforce stability. The same goes for AI. The issue is not whether AI will enter EMS. That’s already the case. The real question is whether EMS leaders will take responsibility for guiding EMS.

Practical steps to bring AI to your EMS today

1. Start with low-risk applications

Use AI when mistakes are easy to spot.

• Documentation support
• QA overview
• training simulation
• Schedule optimization

These save time without jeopardizing patient care.

2. Build governance before deployment

AI must operate in the following environments:

• Medical Director Supervision
• QA audit trail
• Clear responsibility policy
• Data privacy protection

If you wouldn’t trust a new doctor without supervision, don’t trust an AI without supervision.

3. Train your employees

Fear is born from misunderstanding.

Teach your crew:

• What AI can do
• What you can’t do
• How to check the output

This builds trust and prevents abuse.

4. Protect clinical judgment

AI should support healthcare workers, not replace them.

EMS is still:

• scene recognition
• patient empathy
• ethical decision
• clinical reasoning

No algorithm can replace a moment of human patient care.

5. Think long term

Unmanned ambulances may or may not become a reality. But AI-assisted EMS systems definitely will. Leaders who prepare now will protect their patients, staff, and agencies.

final thoughts

The NAEMSP conversation stuck with me not because an unmanned ambulance was just around the corner, but because it reminded me of something important.

Technology does not define EMS. Leadership does that.

Once you understand AI, you can shape it to improve response times, reduce burnout, enhance education, and provide better patient care.

If we ignore it, someone else will shape it for us.

And EMS has never been good at letting outsiders decide how we practice.

The future is coming.

Let’s approach this issue with the same knowledge, ethics, and professionalism that has propelled this field forward for 50 years.