Chapter 10: AI and Work — Automation, Augmentation, and the Future of Jobs

"The factory of the future will have only two employees: a man and a dog. The man will be there to feed the dog. The dog will be there to keep the man from touching the equipment." — Warren Bennis (attributed), management scholar

What you'll learn: How to think clearly about AI's impact on work — not with panic or denial, but with frameworks that help you evaluate claims, understand who's most affected, and plan your own career in a world where AI is a coworker, not just a tool.

Why it matters: If you work for a living — or plan to — AI will change how you do your job. Maybe it already has. The question isn't whether AI will affect work. It's how, for whom, and what you can do about it. Getting these answers wrong has real consequences: bad predictions lead to bad policies, which leave real people behind.

🔗 Connection: In Chapter 1, we established that AI is a spectrum of techniques, not a single technology. In Chapter 3, you learned how machines learn from patterns in data. Those foundations matter here because the type of AI — and the type of learning it requires — determines which tasks it can and can't perform. A system that classifies images (Chapter 6) threatens different jobs than one that generates text (Chapter 5).

10.1 Automation Anxiety: A Very Old Fear

In 1830, a group of English textile workers called the Luddites smashed mechanical looms in factories across the Midlands. They weren't anti-technology in some abstract sense — they were skilled craftspeople watching their livelihoods evaporate as machines did in minutes what had taken them hours. The new looms didn't need a weaver's years of training. They needed a machine operator who could be hired for a fraction of the cost.

The Luddites lost. The machines stayed. And yet — here's the part that rarely makes the headlines — the total number of jobs in England didn't decline. It grew. The textile industry expanded so dramatically that it employed more people than before, just in different roles. The jobs shifted from skilled artisan weaving to factory operation, machine maintenance, logistics, sales, and management.

This pattern has repeated with eerie consistency:

📜 Historical Context: Waves of Automation Anxiety

Era	Technology	Fear	What Actually Happened
1810s–1830s	Power looms	Mass unemployment of weavers	Textile employment grew; wages eventually rose
1920s–1930s	Assembly lines	"Technological unemployment" (Keynes's term)	Manufacturing boomed; service sector expanded
1960s–1970s	Mainframe computers	Office workers obsolete	New categories of information work emerged
1990s–2000s	Internet & software	"End of work" predictions	Entirely new industries created (e-commerce, digital media)
2010s–present	AI & robotics	Robots take all jobs	Still unfolding...

Each wave followed a similar emotional arc: initial fear, genuine disruption for specific groups, followed by broader economic adaptation. So should we simply relax? Is AI just the latest Luddite panic?

Not exactly. And understanding why requires moving beyond the simple "technology creates more jobs than it destroys" narrative. That narrative is historically true in aggregate, but it hides crucial details. The weavers who lost their jobs didn't become factory managers — their children and grandchildren did, after decades of social upheaval. The "aggregate" story glosses over a lot of individual suffering.

Here's what makes AI different from previous automation waves — and what makes it similar:

What's similar: AI automates tasks, not entire jobs. Just as a power loom automated the task of weaving but didn't eliminate the textile industry, AI automates specific tasks within jobs — scheduling, data entry, pattern recognition, first-draft writing — without necessarily eliminating the jobs themselves.

What's potentially different: Previous automation waves primarily affected routine physical tasks (factory work) or routine cognitive tasks (data processing). AI, particularly generative AI, is reaching into non-routine cognitive territory — creative work, analysis, professional judgment — that was previously considered automation-proof. A machine that can write legal briefs, generate marketing copy, and analyze medical images is qualitatively different from a machine that welds car frames.

🔄 Check Your Understanding: Name one historical technology that caused genuine short-term job losses but led to long-term employment growth. What made the transition painful for the workers initially displaced?

The honest answer to "Will AI take my job?" is: probably not your whole job. But it will almost certainly change how you do your job. And for some workers — particularly those whose roles involve a high proportion of automatable tasks — the changes will be profound.

Let's build the framework you need to think about this clearly.

10.2 Tasks, Not Jobs: The Automation Framework

The single most useful idea in the entire AI-and-work conversation is this: AI automates tasks, not jobs.

This insight comes from economists Daron Acemoglu and Pascual Restrepo, and it transforms how we think about automation. Instead of asking "Will AI replace lawyers?" — a question that invites either panic or dismissal — we ask: "Which tasks that lawyers currently perform could AI do equally well or better?"

💡 Intuition: The Task-Based Framework

Think of any job as a bundle of tasks. A nurse's bundle might include: - Taking vital signs (temperature, blood pressure, heart rate) - Administering medication - Monitoring patients for changes in condition - Comforting anxious patients and families - Educating patients about their care plan - Coordinating with doctors and specialists - Documenting everything in medical records

AI can already assist with some of these tasks — automated vital sign monitoring, medication dosage checking, documentation via voice-to-text. But comforting a frightened patient? Reading the room when a family member is about to break down? Making a judgment call about whether a subtle change in a patient's appearance warrants waking a doctor at 3 a.m.? Those tasks require empathy, physical presence, contextual judgment, and the kind of holistic understanding that current AI doesn't have.

The job of "nurse" isn't going away. But the mix of tasks that nurses spend their time on is already shifting.

This task-based framework is our first new concept, and it's the foundation for everything else in this chapter. Let's formalize it.

The Task Decomposition Method (your new technique for this chapter):

1. List the tasks. Break any job into its component tasks. Be specific — not "communication" but "explaining treatment options to patients in plain language."
2. Classify each task. For each task, assess whether it is: - Routine and structured (follows clear rules, uses standardized data) — more automatable - Non-routine but pattern-based (requires judgment, but judgment that follows learnable patterns) — partially automatable with current AI - Non-routine and deeply contextual (requires empathy, physical dexterity in unstructured environments, creative problem-solving, or ethical judgment) — less automatable, at least for now
3. Assess the proportion. What percentage of the job's time goes to each category? A job where 80% of time is spent on routine, structured tasks is more vulnerable than one where 80% requires contextual judgment.
4. Consider the interaction effects. Sometimes automating one task changes the value of the remaining tasks. If AI handles all the routine documentation, the remaining human tasks (clinical judgment, patient communication) might become more valuable, not less.

Let's apply this to a real example. Consider a financial analyst at an investment firm:

By this analysis, roughly 60% of a financial analyst's tasks are highly or moderately automatable. Does that mean 60% of financial analysts will lose their jobs? No. It means the role of financial analyst will evolve. Analysts who can leverage AI to do the routine work faster — and then spend more time on the high-value contextual tasks — will be more productive. Firms might need fewer analysts to do the same volume of work, but the remaining analysts will likely be doing more interesting, higher-judgment work.

This is the difference between automation and augmentation — our second key concept.

Automation means AI replaces a human task entirely. The human no longer does it.

Augmentation means AI enhances a human's ability to do a task. The human still does it, but faster, better, or with fewer errors.

The same AI system can automate and augment, depending on the task. A radiology AI might automate the task of initial image screening (flagging images that look normal) while augmenting the radiologist's ability to detect subtle anomalies (highlighting regions of interest that the human might miss). We'll see exactly this scenario in Case Study 2.

⚠️ Common Pitfall: People often assume automation and augmentation are opposites — that a technology either replaces you or helps you. In reality, most AI deployments do both simultaneously, just for different tasks within the same job. The question isn't "replacement or augmentation?" but "which tasks are replaced, and which are augmented?"

🔄 Check Your Understanding: Using the Task Decomposition Method, break down a job you're familiar with (a part-time job you've held, a family member's profession, or your intended career) into at least five specific tasks. Classify each task as routine/structured, non-routine/pattern-based, or non-routine/deeply contextual.

10.3 Who's Most Affected? Inequality in AI's Labor Impact

Here's where the conversation gets uncomfortable. AI's impact on work is not evenly distributed. It never was — and it likely never will be. Understanding who bears the costs of technological change is just as important as understanding the technology itself.

Previous waves of automation disproportionately affected blue-collar workers — manufacturing, agriculture, routine manual labor. Economists call this skill-biased technological change: technology increases demand for skilled workers while reducing demand for unskilled workers, widening the wage gap.

AI complicates this pattern in two important ways.

First, AI reaches into white-collar territory. For the first time, automation is affecting tasks that require college degrees. Legal research, medical image analysis, financial modeling, software testing, marketing copywriting, translation — these are educated-professional tasks that AI can now perform at a functional level. This doesn't mean lawyers and doctors are about to be replaced (remember: tasks, not jobs), but it does mean that the comfortable assumption that "just get a college degree and you'll be fine" is no longer sufficient.

Second, AI can simultaneously threaten and augment the same profession. Consider journalism. AI can generate routine news articles (earnings reports, sports scores, weather summaries) with minimal human involvement — several major news organizations already use this approach. That threatens entry-level reporters whose first jobs often involved exactly those routine articles. But AI can also augment investigative journalists by analyzing thousands of documents for patterns, identifying potential sources, and speeding up fact-checking. Same profession, different impacts depending on what part of journalism you do.

📊 Real-World Application: Who's Most Exposed?

Research from multiple institutions has attempted to map AI exposure across occupations. While exact numbers vary by study and methodology, some consistent patterns emerge:

Higher exposure to AI automation: - Data entry and processing clerks - Bookkeeping and accounting clerks - Paralegals and legal assistants (routine legal research) - Customer service representatives (routine inquiries) - Translators (routine document translation) - Telemarketers - Basic financial analysis

Lower exposure to AI automation: - Skilled trades (electricians, plumbers — unstructured physical environments) - Nurses and home health aides (empathy + physical care) - Elementary school teachers (child development + social skills) - Social workers (complex interpersonal judgment) - Emergency responders (unstructured, high-stakes physical environments) - Skilled construction workers

Notice the pattern? High-exposure jobs tend to involve routine cognitive tasks in structured digital environments. Low-exposure jobs tend to require physical presence in unstructured environments, deep interpersonal skills, or both. The irony: some of the "safest" jobs from AI automation are among the lowest-paid in the economy.

This raises a profound equity question. If AI automates middle-skill cognitive work while leaving both high-skill professional work and low-skill physical work relatively intact, we could see a deepening of the labor market polarization that's been underway for decades — a hollowing out of the middle class, with jobs concentrating at the high-wage and low-wage extremes.

And the impacts aren't just about skill level. They intersect with:

Geography. AI's labor impact hits differently in a tech hub like San Francisco (where AI companies are hiring) versus a mid-sized city whose economy depends on call centers or data processing.

Race and gender. Occupational segregation means that automation of certain job categories disproportionately affects specific demographic groups. In the United States, Black workers are overrepresented in some of the job categories most exposed to automation. Women dominate administrative support roles that are highly automatable but are also strongly represented in caregiving roles that are less automatable.

Age. Older workers may find it harder to retrain for new roles. A 55-year-old bookkeeper whose job is automated by AI faces a very different set of options than a 25-year-old bookkeeper.

Global inequality. AI is predominantly developed in wealthy nations but deployed globally. When AI automates call center work, the job losses often occur in countries like India and the Philippines — where those jobs represented pathways to the middle class — while the productivity gains accrue to companies headquartered in the United States or Europe.

🔄 Check Your Understanding: Why might AI's impact on work be more unequal than previous waves of automation? Identify at least two factors that differentiate AI from earlier technologies like assembly-line robots.

🔗 Connection: Chapter 9 explored how AI systems can embed and amplify existing inequalities through biased data and design choices. The labor market impact of AI is another dimension of the same pattern: the costs and benefits of AI are distributed along lines of existing social inequality.

10.4 New Jobs, Changed Jobs, Lost Jobs

Every conversation about AI and work eventually arrives at the same question: "But will there be new jobs?" The historical record says yes — but history also teaches us that the transition matters enormously. Let's break this into three categories.

Lost Jobs

Some jobs will genuinely disappear. Not tomorrow, and probably not entirely, but the number of humans performing certain tasks will decline significantly. This has already happened:

• Toll booth operators have been largely replaced by electronic tolling systems (not AI per se, but automated systems).
• Travel agents declined by over 60% after online booking platforms emerged.
• Bank tellers dropped significantly after ATMs and online banking.

AI is accelerating this process for a new set of roles. Companies are already reducing headcount in customer service, basic content writing, data entry, and first-level technical support by deploying AI chatbots and automation tools. These aren't predictions — they're things that are happening now.

The honest thing to say about lost jobs is this: for the individuals affected, knowing that "new jobs will eventually emerge" is cold comfort. The transition costs are real and often borne by those least able to absorb them.

Changed Jobs

Most jobs won't disappear — they'll transform. This is the augmentation story, and it's the most common outcome. Consider these examples:

Doctors aren't being replaced by AI, but AI is changing what their day looks like. AI-assisted diagnostic tools mean less time squinting at images and more time talking to patients. AI-generated clinical notes mean less documentation burden. The core of the job — clinical judgment, patient relationships, ethical decision-making — remains human. But the wrapper around that core changes significantly.

Teachers aren't being replaced by AI tutors, but they're increasingly expected to integrate AI tools into their classrooms, help students use AI responsibly, and redesign assignments that can't simply be completed by ChatGPT. Teaching is evolving from "delivering content" (which AI can do) to "facilitating learning experiences" (which AI can't).

Software developers increasingly use AI coding assistants for routine code generation, debugging, and documentation. This doesn't eliminate the need for developers — it shifts their work toward architecture, design, code review, and understanding user needs. Junior developers, who previously did much of the routine coding, face a particular challenge: the traditional entry-level tasks that taught them the craft are increasingly automated.

💡 Intuition: The "Last Mile" Problem in Job Automation

Think of automating a job like automating a delivery route. AI might handle 90% of the route flawlessly — the highway driving, the navigation, the package tracking. But the "last mile" — navigating a cluttered apartment building, dealing with a locked gate, handling a customer complaint — is disproportionately difficult and expensive to automate.

Many jobs have a "last mile" of tasks that AI struggles with: the ambiguous case, the emotional conversation, the ethical judgment call, the physical manipulation in an unstructured environment. That last mile is often where the most human value resides.

New Jobs

New technologies create new job categories that didn't exist before. Some new AI-related jobs are already emerging:

• Prompt engineers who optimize interactions with AI systems
• AI trainers who provide human feedback to improve model behavior
• AI ethicists and responsible AI specialists who evaluate and mitigate AI risks
• AI integration specialists who help organizations deploy AI effectively
• Data annotators and content moderators who prepare training data and review AI outputs

But let's be honest about the limits of this "new jobs" narrative. Many AI-related jobs are themselves precarious. Data annotators and content moderators — people who label images, flag harmful content, and evaluate AI outputs — are often poorly paid contract workers, sometimes in the Global South, with limited protections. The gig workers who train AI systems are, ironically, among the most vulnerable to displacement by the very systems they help create.

📊 Real-World Application: The International Labour Organization and various research institutions have studied AI's potential labor market impact across different sectors and countries. While estimates vary widely — some suggest AI could affect roughly 40% of jobs globally to some degree — most serious researchers emphasize that "affect" means "change," not "eliminate." The fraction of jobs that are fully automatable by current AI technology is relatively small (estimates typically range from 5–15% of current jobs). The fraction that will be significantly changed by AI is much larger.

10.5 The Gig Economy and Algorithmic Management

So far, we've discussed AI as something that replaces or changes work tasks. But there's a third impact that's equally important and often overlooked: AI doesn't just do work — it manages workers.

Algorithmic management — our third new concept — refers to the use of AI and automated systems to assign, monitor, evaluate, and discipline workers. If you've ever driven for a rideshare company, delivered food through an app, or worked in an Amazon warehouse, you've experienced algorithmic management firsthand.

Here's what it looks like in practice:

Task assignment. An algorithm decides which driver gets which ride, which warehouse worker picks which order, which gig worker gets which task. The worker doesn't choose — the algorithm assigns.

Performance monitoring. The algorithm tracks everything: speed, efficiency, customer ratings, time between tasks, route choices, even facial expressions (in some systems). Every second of work is measured, recorded, and analyzed.

Evaluation and discipline. Based on the data, the algorithm rates workers, adjusts their access to work (better or worse shifts, more or fewer rides), and can effectively "fire" them — not through a human conversation, but through deactivation. The worker's relationship isn't with a manager; it's with a system.

📊 Real-World Application: The Rideshare Driver's Invisible Boss

Consider the experience of a rideshare driver. The app determines: - Which rides they're offered (and doesn't explain why) - What route to take (and penalizes deviations) - How much they're paid (using dynamic pricing the driver can't see in advance) - Their rating, which determines future access to rides - Whether they're "deactivated" — the gig economy's euphemism for fired

The driver has no way to appeal to a human manager, no union representative, no formal grievance process. The algorithm is the boss, and the boss doesn't explain its decisions.

This matters for our discussion because algorithmic management raises questions that go beyond "will AI take my job?" to "what kind of job will AI leave me with?"

Consider the warehouse workers at major e-commerce companies. Their jobs haven't been eliminated — they're still physically picking, packing, and shipping products. But the experience of the job has been transformed. AI systems dictate the pace, monitor bathroom breaks, track "time off task" in seconds, and generate productivity scores that determine whether a worker keeps their position. The humans are still there, but they're increasingly extensions of the machine rather than autonomous workers.

⚠️ Common Pitfall: It's tempting to frame algorithmic management as simply "more efficient" management. But efficiency for whom? Algorithmic management optimizes for the company's metrics — speed, throughput, cost reduction. It doesn't optimize for worker well-being, job satisfaction, or career development. The interests of the system and the interests of the workers are not the same.

There's also a transparency problem. When a human manager evaluates your work, you can ask them to explain their reasoning. You can appeal. You can have a conversation. When an algorithm evaluates you, the reasoning is often opaque — a "black box" of the kind we discussed in earlier chapters. Workers subject to algorithmic management frequently report that they don't understand why they received a particular rating, why they were assigned certain shifts, or why they were deactivated.

🔗 Connection: Remember the discussion of AI decision-making from Chapter 7? The same issues of transparency, accountability, and the difference between prediction and truth apply here. When an algorithm decides that a warehouse worker is "underperforming," it's making a classification based on data — but that classification carries real consequences for a real person.

The rise of algorithmic management also intersects with the decline of traditional employment relationships. In the gig economy, workers are classified as independent contractors, which means they don't receive benefits, workplace protections, or the legal rights that come with employee status — even though the algorithm controls their work as tightly as any traditional employer.

🔄 Check Your Understanding: What's the difference between AI that replaces a worker's tasks and AI that manages how a worker does their tasks? Why does the second category get less public attention than the first?

10.6 Preparing for an AI-Augmented Career

Let's get practical. If you're reading this as a student, a professional, or someone thinking about your career trajectory, you're probably wondering: "What should I actually do?"

The bad news: nobody can predict the future with certainty. Anyone who tells you they know exactly which jobs will exist in 20 years is selling something. The good news: there are strategies that have been resilient across multiple waves of technological change, and they're even more relevant now.

🧩 Self-Assessment: Your AI Exposure Profile

Before reading the strategies below, take a few minutes to assess your own situation. For your current or intended career:

1. List your core tasks. Break your work (or intended work) into specific tasks.
2. Rate each task's AI exposure. On a scale of 1 (AI can't touch this) to 5 (AI can already do this well), rate each task.
3. Calculate your exposure proportion. What fraction of your work time is spent on high-exposure tasks (rated 4 or 5)?
4. Identify your "last mile." What are the tasks that require your uniquely human skills — empathy, creativity, physical presence, contextual judgment?
5. Assess your adaptability. How quickly could you shift your time toward lower-exposure tasks if the high-exposure tasks were automated?

This isn't a prediction of your future — it's a snapshot of where you stand and a starting point for planning.

Strategy 1: Develop Complementary Skills

The most resilient career strategy isn't to compete with AI at what AI does best — it's to develop skills that complement AI. These include:

• Complex communication. Explaining, persuading, negotiating, comforting — tasks that require reading emotional context and adapting in real time.
• Creative problem-solving. Not generating novel combinations of existing elements (AI can do that) but identifying which problems are worth solving in the first place, and framing them in useful ways.
• Ethical judgment. Making decisions that involve competing values, cultural context, and moral reasoning — areas where there isn't a "correct" answer in the training data.
• Physical dexterity in unstructured environments. Working with your hands in varied, unpredictable settings — from surgery to plumbing to emergency response.
• Cross-domain synthesis. Connecting insights from different fields, seeing patterns that span disciplines. AI trained on narrow domains struggles with this.

Strategy 2: Learn to Work With AI

The professionals who will thrive aren't those who resist AI or those who surrender to it — they're those who learn to collaborate with it effectively. This is our fourth new concept: human-AI teaming, the practice of deliberately combining human and AI capabilities to achieve outcomes that neither could achieve alone.

Effective human-AI teaming requires:

• Knowing what AI is good at — pattern recognition in large datasets, consistency, speed, tireless attention.
• Knowing what AI is bad at — contextual judgment, ethical reasoning, handling genuinely novel situations, common sense.
• Knowing where you add value — the tasks where your judgment, creativity, empathy, or contextual knowledge makes the difference between a good outcome and a bad one.
• Knowing how to verify AI outputs — because AI is confident, not correct (Chapter 8).

📊 Real-World Application: Some organizations are already redesigning workflows around human-AI teaming principles. In these setups, AI handles the first pass — screening applications, drafting initial analyses, generating options — and humans handle the review, refinement, and final decision. The key insight: neither the AI nor the human does the whole job. They do different parts, and the interface between them is deliberately designed.

Strategy 3: Build Adaptive Capacity

The specific skills that are valuable will change over time. What won't change is the ability to learn new skills quickly. Economists call this adaptive capacity, and it may be the single most important career asset in an age of AI.

Building adaptive capacity means:

• Learning how to learn. Meta-skills — knowing how to teach yourself a new domain, evaluate new tools, and transfer knowledge from one context to another.
• Maintaining a broad knowledge base. Specialists are valuable, but specialists who can also connect their expertise to other domains are more resilient.
• Staying curious. Regularly experimenting with new AI tools, following developments in your field, and asking "how could this change my work?" — not with fear, but with genuine curiosity.
• Building a professional network. Relationships, trust, and reputation remain uniquely human assets that AI can't replicate.

Strategy 4: Advocate for Just Transitions

Individual strategies are important, but they're not sufficient. The societal challenge of AI and work requires collective action, too. This includes:

• Supporting workforce retraining programs that help displaced workers transition to new roles.
• Advocating for social safety nets — unemployment insurance, portable benefits, healthcare not tied to employment — that cushion transitions.
• Demanding transparency in how algorithmic management systems work and how they affect workers.
• Participating in policy conversations about AI governance, labor rights, and the future of work.

💡 Intuition: Think of AI and work like a river changing course. Individual strategies (learning to swim, building a boat) help you navigate the immediate changes. But structural strategies (building levees, creating early warning systems, designating flood plains) protect the whole community. You need both.

🔄 Check Your Understanding: Why is "learn to code" insufficient as career advice in the age of AI? What would better career advice look like?

10.7 Chapter Summary

Let's consolidate what we've learned.

Key Concepts

1. The Task-Based Framework. AI automates tasks, not jobs. To understand AI's impact on any profession, decompose the job into its component tasks and evaluate which tasks AI can perform. Most jobs are a mix of automatable and non-automatable tasks.

2. Automation vs. Augmentation. Automation replaces human tasks entirely; augmentation enhances human capability. Most AI deployments do both — automating some tasks within a job while augmenting others. The net effect depends on the proportion.

3. Algorithmic Management. AI doesn't just replace or assist workers — it increasingly manages them. Algorithmic management assigns, monitors, evaluates, and disciplines workers through automated systems, raising questions about transparency, autonomy, and worker rights.

4. Human-AI Teaming. The most effective approach to AI in the workplace combines human and AI capabilities deliberately, leveraging each party's strengths. This requires understanding what AI can and can't do, and designing workflows around the interface between them.

Key Debates

• Is this time different? Previous automation waves ultimately created more jobs than they destroyed — but transitions were painful and unequal. AI's reach into non-routine cognitive work may make this wave qualitatively different, or it may follow the historical pattern. The honest answer: we don't know yet.
• Who bears the costs? The benefits of AI automation tend to accrue to capital owners and highly skilled workers. The costs tend to fall on mid-skill workers, specific demographic groups, and workers in the Global South. Managing this inequality is a policy choice, not a technological inevitability.
• What kind of work do we want? Even when AI doesn't eliminate jobs, it can transform them in ways that make work less autonomous, less meaningful, and more surveilled. The question isn't just "will there be jobs?" but "will the jobs be good?"

What This Means for Your AI Literacy

Understanding AI and work isn't just an academic exercise — it's a survival skill. The frameworks in this chapter — task decomposition, automation vs. augmentation, algorithmic management, human-AI teaming — give you tools to evaluate claims about AI's labor impact with nuance instead of panic or dismissal. When you read a headline that says "AI will eliminate 300 million jobs," you now know to ask: "Which tasks within those jobs? Over what time frame? With what transition support? And who benefits from framing it this way?"

📐 Project Checkpoint: Analyze Your AI System's Impact on Workers

For your AI Audit Report, add a labor impact analysis for your chosen AI system:

1. Who does this work currently? Identify the workers whose tasks your AI system performs or assists with.
2. Task decomposition. Break those workers' roles into specific tasks. Which tasks does the AI system handle? Which remain human?
3. Automation or augmentation? For each affected task, is the AI replacing the human (automation) or assisting the human (augmentation)?
4. Who benefits, who is harmed? Trace the economic effects. Who gains productivity? Who might lose hours, jobs, or job quality?
5. Algorithmic management. Does your AI system manage or monitor workers? If so, how transparent is that management?
6. Recommendations. What policies or design changes could make this AI system's labor impact more equitable?

Add this analysis as a new section in your cumulative audit report.

Spaced Review

Before moving on, take a moment to revisit these concepts from earlier chapters:

🔄 Spaced Review — Chapter 3 (How Machines Learn): What's the difference between supervised and unsupervised learning? How might the type of learning approach affect which work tasks an AI system can automate?

🔄 Spaced Review — Chapter 5 (Large Language Models): LLMs predict the next token — they don't understand meaning. How does this distinction matter when evaluating claims about AI replacing knowledge workers?

🔄 Spaced Review — Chapter 7 (AI Decision-Making): We discussed feedback loops — when AI decisions shape the data that future AI decisions are based on. How might algorithmic management systems create feedback loops in the workplace?

What's Next

In Chapter 11, we'll explore a specific dimension of AI and work that raises unique questions: creativity. When AI can generate art, compose music, and write essays, who is the "author"? Can a machine be "creative"? And what happens to the human artists, musicians, and writers whose work trained the AI in the first place? These questions push us beyond economics into philosophy, law, and the nature of human expression itself.

Key Terms Introduced in This Chapter