After studying this material, students should be able to:

1. Define what robots are and how they differ from other automated systems
2. Explain the robot exposure index and its four-tier classification system
3. Analyze which jobs face high vs. low robot exposure and why
4. Evaluate the economic barriers (especially cost) limiting robot adoption
5. Compare robot exposure vs. LLM exposure across job categories
6. Interpret what current robot capabilities suggest about future labor market disruption

---

Step-by-Step Teaching Guide

Step 1: What Is a Robot? (Defining the Scope)

Before measuring anything, we need a precise definition.

A robot, as used in this research, is:

An autonomous physical machine that senses and acts

What counts as a robot:

• Car wash systems that scan cars and adjust sprayers ✅
• Warehouse autonomous mobile robots ✅
• Self-driving vehicles ✅

What does NOT count:

• Teleoperated surgical machines fully controlled by a surgeon ❌
• Software/AI alone (like LLMs) ❌

Why this distinction matters: The key word is autonomous. The machine must make decisions based on its environment without a human directly controlling each action.

---

Step 2: The Four-Tier Exposure System

The researchers needed a way to measure how capable robots are at any given job task. They created a rubric based on environmental control — how structured or predictable the environment needs to be.

E0 → E1 → E2 → E3
(No capability) ----→ (Most capable)

Why environmental control is the key variable:

Robots struggle with unpredictability. A factory robot knows exactly where every part will be. A nurse's aide does not know what a patient will do next.

Think of it this way: the more a robot needs the world to be arranged for it, the lower its exposure tier.

---

Step 3: How Exposure Was Actually Measured

The researchers used O*NET (a database of ~900 occupations and ~19,000 job tasks) combined with Claude (an AI) to rate each task.

The process, step by step:

Step A: Identify which tasks are physical

• "Dig trenches" → Physical ✅
• "Compute totals on report forms using a calculator" → Not physical ❌

Step B: For each physical task, generate concrete examples of how it's performed

• Example: "Dig trenches" breaks into:
  • 25% = cutting linear trenches in open ground
  • 20% = careful digging by hand around buried pipes

Step C: Rate each example using web search for real robot deployments

• Linear trenching → E3 (autonomous excavator exists)
• Careful digging around pipes → E0 (no robot can do this)

Step D: Set overall task exposure by majority rule

• More examples resemble careful digging → Overall: E0

Key rule:

Only demonstrated robot capabilities count. Robots must perform tasks as well as humans in terms of reliability, error rates, and speed.

---

Step 4: What the Data Shows — The Big Picture

Now let's look at what this system reveals about the US economy.

Breakdown of all job tasks by working time:

All job tasks (100% of working time)
│
├── Cognitive & Interpersonal tasks: 54%
│   (Not measured for robot exposure)
│
└── Physical tasks: 46%
    ├── E0 (robots can't do): ~12% of ALL tasks
    ├── E1 (purpose-built only): ~23% of ALL tasks
    ├── E2 (structured workplace): ~10% of ALL tasks
    └── E3 (unstructured): ~1% of ALL tasks

Key finding: Robots can perform 74% of physical tasks (= 34% of ALL working hours) in some setting.

Most exposed occupations:

The top 10 most exposed jobs are almost all vehicle operators, because driving is highly exposed at E3 (autonomous vehicles exist on real roads).

Least exposed work:

• Nursing (requires delicate human interaction + physical contact)
• General repair (unpredictable environments + fine dexterity)
• Personal care services (interpersonal + physical contact)

---

Step 5: Who Is Affected? Demographics of Exposed Workers

This is crucial for understanding the social implications.

Workers in the most exposed jobs (top 20%) compared to unexposed workers:

Important contrast: This is the opposite pattern from LLM exposure, which tends to affect higher-paid, more educated, white-collar workers.

---

Step 6: Robots vs. LLMs — Complementary Threats

A major insight is that robots and LLMs expose different parts of the economy.

LLM exposure alone vs. LLM + Robot exposure:

Key insight: Robots do work where LLMs cannot (physical tasks), and LLMs do work where robots cannot (cognitive/language tasks). Together, they expose 80% of all working time.

What remains unexposed (~20%)?

Tasks that are:

1. Highly interpersonal (human-to-human contact that people value)
2. Require delicate physical manipulation robots can't yet do
3. Regulated (legally must be done by humans)

---

Step 7: The Cost Barrier — Why Capability ≠ Adoption

This is perhaps the most important practical insight.

The gap between "can do" and "cost-effective":

That's a massive gap. Why?

Example: Packers and Packagers

• Robot system cost: $2 million+ to purchase and install
• Replaces: ~14 workers
• Annual robot cost per worker replaced: ~$45,000
• Annual human worker cost: ~$49,000
• Result: Robots are cost-competitive (barely — $2,500 cheaper/year)

Example: Welders

• Robot cost: ~5× more expensive than human welders
• Reason: Welders do many tasks (positioning parts, climbing ladders, quality checking) that require multiple expensive robots

Example: Dishwashers/Janitors

• Paid $25,000–$30,000 less than welders
• Robot analogs still several times more expensive
• Cleaning is hard to standardize; robots are slower

The 40-Year Problem:

Robot prices have declined ~3% per year since the 1990s

To reach cost-competitiveness for 10% of human work:
→ Requires ~70% cost decline
→ At 3%/year = approximately 40 YEARS

---

Step 8: Barriers Beyond Cost

Even if robots become cheaper, other barriers exist:

Manipulation is the biggest capability gap. Half of all physical tasks require robots to handle objects with human-like dexterity — something current robots cannot do reliably.

---

Step 9: Historical Validation — Does This Predict Real Disruption?

The researchers tested their method against 50 years of data (1977–today).

Finding: Jobs that were more exposed to robots in earlier decades experienced:

• Greater wage declines
• Greater employment declines

in subsequent decades — even after controlling for industry trends.

This validates the core premise: Current robot exposure predicts future labor market impact.

Capability growth over time:

• In 1977, robots could NOT do 62% of physical tasks
• Today, robots cannot do only 24% of those same tasks
• Rate of progress: ~2% of previously impossible tasks become possible each year

---

Step 10: Synthesis — What This All Means

Let's bring it together with a framework for thinking about robot automation risk:

High near-term risk (exposed + approaching cost-competitiveness):

• Taxi/ride-share drivers (E3 capable, regulatory barriers remain)
• Warehouse packers (E2 capable, already cost-competitive)
• Postal mail carriers (highly exposed, ~54% cost decline needed)

Medium-term risk (exposed but expensive):

• Recycling workers (broadly exposed at E2, but robots cost more)
• Drywall tapers (partially exposed at E3, messy tasks bottleneck)

Lower near-term risk (capability gaps + interpersonal):

• Nurses (regulation + interpersonal + dexterity)
• General repair workers (unpredictable environments)
• Personal care workers (human preference for human contact)

The key equation for automation risk:

Automation Risk = f(Robot Capability × Cost-Competitiveness × Regulatory Permissibility × Human Acceptance)

All four factors must align for widespread adoption.

---

Summary Table