Section 1.2 — What AI Can and Cannot Do

An Impressive Assistant—But Not an Infallible One

Imagine asking an AI assistant to help you prepare for an important job interview.

Within seconds, it can suggest likely interview questions, help you organize your experience, and conduct a practice interview. You might feel much better prepared than you would have without it.

Now imagine asking the same assistant whether the job offer you received is legitimate. It confidently assures you that everything looks fine, but overlooks signs of a scam.

The same technology that helped you prepare could mislead you if you relied on it without checking.

This illustrates an essential distinction:

Being capable is not the same as being reliable.

AI can perform an extraordinary range of tasks, sometimes with impressive results. Yet its performance varies across tasks, situations, and systems. Understanding both its capabilities and limitations is essential to using it wisely.

What AI Can Do Well

Modern AI systems can assist with many activities. Three especially useful categories are understanding, creating, and analyzing.

Understand: Working with Language and Information

AI can explain unfamiliar ideas, summarize documents, translate languages, and organize information.

For example, it can turn a lengthy report into a short overview, explain a difficult concept in simpler language, or help someone compare different perspectives on an issue.

These capabilities can make information more accessible and save time. They are particularly useful when we want to explore a subject, identify important questions, or prepare for further study.

Example: Understanding a Complicated Document

Suppose you receive a lengthy insurance policy that is difficult to understand.

You might ask AI:

“Explain the main benefits, exclusions, and limitations of this policy in plain English. What questions should I ask before making a decision?”

AI may help you identify important points and prepare useful questions. But its summary should not replace checking the original policy, especially when coverage or financial consequences are significant.

Create: Developing Ideas and Producing Content

AI can help brainstorm ideas, draft messages, generate images, write computer code, suggest designs, and develop preliminary plans.

A writer might ask for alternative titles. A teacher might request additional examples. A community organization might use AI to explore ideas for an event.

AI can also help people overcome the difficulty of starting a project by producing an initial draft that they can evaluate and improve.

However, producing content quickly does not guarantee that it is original, accurate, appropriate, or valuable.

AI can generate possibilities; people still need to evaluate which possibilities are worth pursuing.

Analyze: Finding Patterns and Comparing Options

Some AI systems can identify patterns in large datasets, classify information, make predictions, or support complex analysis.

AI is used in applications such as detecting suspicious financial transactions, analyzing medical images, forecasting demand, and supporting scientific research.

Conversational AI assistants can also help compare alternatives, identify possible trade-offs, and organize information relevant to a decision.

However, different AI systems have different capabilities. A model that performs well in one application may not be suitable for another.

The right question is not simply “Can AI do this?” but “Can this particular AI system do this task well enough for my purpose?”

Where AI Has Limitations

AI’s limitations are not always obvious. A response may sound persuasive even when it contains mistakes or overlooks important information.

Three limitations are especially important.

Accuracy: AI Can Be Wrong

AI can produce incorrect facts, flawed calculations, misleading explanations, or references that do not exist.

Some mistakes are obvious. Others are difficult to detect because the response appears polished and confident.

Warning: Confidence Is Not Evidence

A fluent, confident AI response is not necessarily correct. For important information, check the answer against reliable sources, original documents, or other appropriate evidence.

We will examine these errors and practical methods for verifying AI responses in Section 1.4.

Context: AI May Not Know What Matters

AI can offer useful general guidance while missing important details about a person’s circumstances.

Consider the question: “Should I accept a new job that pays more but requires longer working hours?”

AI can help compare compensation, career opportunities, commuting time, and possible consequences. But it may not fully understand your family responsibilities, personal values, health, financial obligations, or long-term goals.

Even when you provide extensive information, the quality of its advice depends on the accuracy and completeness of that information.

AI can help structure a decision without necessarily knowing which choice is best for you.

Judgment: AI Cannot Determine Your Values for You

Some decisions involve more than facts and calculations.

Should you pursue a demanding career opportunity at the expense of family time? How should you balance financial security with personal fulfillment? What responsibilities do you have toward other people?

AI can help identify alternatives, clarify trade-offs, and challenge assumptions. It may even identify considerations that you have overlooked.

But technical capability alone does not give an AI system the authority to determine what should matter most in your life.

AI can support human judgment without replacing human responsibility.

Capability, Reliability, and Authority

To understand when AI should be trusted, it helps to distinguish three questions.

Question	Meaning
Capability	Can the AI system perform this task?
Reliability	How consistently and accurately can it perform the task?
Authority	Should it be entrusted to make or carry out the decision?

These are related but different questions.

For example, an AI system might be capable of analyzing medical information but insufficiently reliable for a particular diagnosis. Even a well-tested medical AI system may need to operate under professional supervision, with appropriate safeguards and accountability.

Similarly, AI might compare retirement strategies accurately without having the authority to decide how much financial risk you should accept.

This distinction helps us avoid two extremes: blindly trusting AI because it is impressive, or dismissing it entirely because it sometimes makes mistakes.

Good AI use requires matching its capabilities, reliability, and authority to the task.

Choosing the Right Role for AI

Rather than treating AI as either completely trustworthy or completely unreliable, consider what role it should play in a particular activity.

AI as an Assistant

AI can help with drafting, summarizing, organizing, brainstorming, and other routine tasks.

For example, you might ask AI to suggest a weekly meal plan and then adjust it according to your preferences and needs.

You remain responsible for reviewing the result.

AI as a Collaborator

AI can help you explore alternatives, identify weaknesses, and improve your work through repeated discussion.

For example, you might ask AI to critique a presentation outline, identify unclear explanations, and suggest improvements.

You and the AI system contribute to the process, but you retain responsibility for the final product.

AI as an Adviser

AI can organize information and suggest considerations relevant to important decisions.

For example, before making a retirement decision, you might ask AI to explain financial concepts, compare possible scenarios, and prepare questions for a financial professional.

However, recommendations involving substantial financial, medical, legal, or personal consequences require careful verification and appropriate expertise.

These roles are flexible. The same AI system might serve as an assistant for one task and a collaborator or adviser for another.

How Much Should You Trust AI?

A practical way to decide how much oversight is needed is to consider the potential consequences of an error.

Practical Guide: Match AI Oversight to Risk

Risk	Example	Recommended Approach
Low	Brainstorming vacation ideas	Explore freely; review for usefulness.
Moderate	Preparing a work report	Check facts, calculations, and sources.
High	Making medical, legal, or financial decisions	Verify independently; involve appropriate professionals.

These are illustrative categories. Actual risk depends on the circumstances, the AI system, and how its output will be used.

The higher the potential consequences of a mistake, the stronger the need for verification, independent judgment, and appropriate human oversight.

This principle applies even when AI appears highly capable.

Using AI with Good Judgment

AI can help us understand information, generate ideas, analyze problems, and explore decisions. Its usefulness is real, but so are its limitations.

We do not need to distrust every AI response, nor should we accept every response without question.

Instead, we can develop a simple habit:

Understand what AI is being asked to do, consider what could go wrong, and decide what level of verification is appropriate.

As AI systems become more capable, this habit will remain valuable. Specific technologies will change, but the need to distinguish capability, reliability, and authority will continue.

In Section 1.3, we will explore how AI is already being used in everyday life, including applications that people may not immediately recognize as AI.

Key Takeaways

AI can help us understand, create, and analyze. Its strengths vary by system, task, and context.
Capability, reliability, and authority are different. An AI system may be able to perform a task without being sufficiently reliable or appropriately authorized to make the final decision.
Match AI oversight to the consequences of error. Use AI flexibly for low-risk tasks and apply stronger verification and human judgment when the stakes are higher.

Reflect and Apply

Exercise: Before You Trust AI

Choose one task in your work, education, or personal life for which you might use AI.

Capability: What can AI help me accomplish?
Risk: What could go wrong if its response is incorrect?
Oversight: What should I verify or decide for myself?

Optional: Decide whether AI should act primarily as an assistant, collaborator, or adviser for this task.

The goal is to develop the habit of using AI deliberately rather than automatically.