AI's Strengths and Limitations

Introduction

🎯 Learning goals

• Know which tasks AI is exceptional at
• Understand where AI struggles or fails
• Know common AI problems: hallucinations, bias, and brittle AI
• Be able to assess when AI is the right tool and when human judgment is needed

AI is not good at everything. But in certain areas it outperforms humans — both in speed and sometimes in precision. Knowing both sides is key to using AI smartly.

There are tasks where AI performs in a way that is fundamentally impossible for a human to match — not because AI is “smarter,” but because it works differently.

When AI is unbeatable: Strengths

1. Processing enormous amounts of data quickly

A human can read perhaps 300–500 words per minute. An AI can analyze millions of words per second.

Examples:

• Analyze 10,000 customer reviews and find the three most common complaints — AI does this in seconds, it would take a human weeks
• Go through 100,000 X-rays to find anomalies — AI can screen all of them faster than humans

2. Identifying complex patterns in data

AI can find connections that humans would never discover.

Examples:

• Detecting credit card fraud through subtle patterns in transactions
• Predicting machine failures by analyzing vibrations, temperature, and sound patterns
• Identifying early signs of disease by correlating thousands of data points

3. Tasks with clear rules and repetition

When the task is well-defined and repetitive, AI is extremely efficient.

Examples: Sorting email, transcribing speech, translating languages, playing chess or Go, recognizing objects in images.

4. Working 24/7 without fatigue

An AI chatbot can handle thousands of customer conversations simultaneously, around the clock, without getting tired or impatient.

5. Consistent performance

An AI does the same thing the same way every time. A human can have a bad day, be distracted, or tired.

AI has fundamental weaknesses — areas where humans are still clearly superior and where AI risks giving misleading or outright incorrect results.

When AI fails: Limitations

1. Contextual understanding and common sense

AI lacks real understanding of the world.

Example:

• “I put the key on the table. Then I flipped the table. Where is the key?”
  • A 5-year-old knows: “On the floor”
  • AI can get confused — it doesn’t understand gravity or how the world works

2. Creativity and innovation

AI can combine existing patterns in new ways (and it can look creative), but it can’t “think outside the box” in the same way humans do.

Example: AI can generate art in the style of Picasso or Monet, but it didn’t create cubism or impressionism itself.

3. Emotional intelligence

AI can recognize emotions (analyze facial expressions or tone of voice), but it doesn’t understand what emotions are and has no empathy.

Example: A customer service AI can identify that you’re angry, but it can’t truly understand your frustration or show genuine empathy. Therapists, teachers, and healthcare workers require human connection.

4. Ethical judgments and moral accountability

AI cannot make moral decisions or take responsibility for consequences. When an AI recruitment service discriminates — who bears the responsibility? The AI cannot be held accountable.

5. Adapting to entirely new situations

AI is brittle — it works within its trained domain but breaks down in new situations.

Example: A self-driving car trained in sunny California may have problems in a snowstorm in Sweden.

Beyond the fundamental limitations, there are four specific problems that are important to know about — especially if you’re going to use AI in professional contexts.

Common AI problems you need to know about

1. Hallucinations – When AI “makes things up”

AI systems, especially language models like ChatGPT, can generate information that sounds credible but is completely fabricated.

Why does it happen? AI always tries to give an answer, even when it doesn’t know. It generates what “sounds right” based on statistical patterns, not actual knowledge.

How do you avoid it?

• Always verify facts from AI against reliable sources
• Ask AI to cite sources (but check that the sources are real!)
• Be especially skeptical of names, dates, statistics, and quotes

2. Bias – When AI learns our prejudices

AI is trained on human data, and humans have biases. If the data is skewed, the AI is skewed.

Real-world examples:

• Amazon recruitment: AI trained on historical hiring decisions favored men, because most previous hires were men
• Face recognition: Works worse on people with darker skin because the training data contained mostly lighter-skinned faces
• Criminal justice: AI that predicted recidivism discriminated against minorities

How do we address it? Review training data, test AI on different groups, maintain human oversight, and require transparency.

3. Brittle AI – When AI is vulnerable to small changes

AI can be extremely good at its trained task but completely lost with small deviations.

Examples:

• Researchers changed a few pixels in an image of a panda (imperceptible to humans) and the AI classified it as a “gibbon” with 99% confidence
• Someone put tape on a stop sign — the car drove straight through

AI is not robust. We must be aware of vulnerabilities, especially in safety-critical systems.

4. The black box – Difficult to explain AI’s decisions

Deep neural networks give an answer, but it’s almost impossible to understand exactly why.

Example: An AI says a patient has an 85% risk of heart disease. The doctor asks: “Why?” — The AI can’t explain which factors carried the most weight.

Problem: Difficult to trust, improve, or debug. Ethically problematic when AI affects people’s lives without transparency. Research is ongoing on Explainable AI (XAI).

With knowledge of both strengths and limitations, you can make informed choices about when AI is the right tool — and when human judgment is needed.

When should you use AI, and when shouldn't you?

AI is perfect for:

• Tasks with enormous amounts of data (analyzing millions of documents)
• Repetitive, well-defined tasks (categorization, sorting)
• Pattern recognition (fraud, medical diagnostics)
• Speed (real-time decisions based on rules)
• 24/7 availability (customer service bots for common questions)

Humans are better at:

• Ethical and moral judgments
• Creativity and innovation beyond existing patterns
• Emotional intelligence and empathy
• Common sense and contextual understanding
• Adapting to entirely new, unforeseen situations
• Accountability and consequence thinking

Smartest: Combine AI and humans

• Let AI do the groundwork (analyze data, sort, suggest)
• Let humans make final decisions, especially on sensitive matters
• Use AI as a tool, not a replacement

Key takeaways

Here we gather the most important insights from this section before you move on to the quiz.

• AI is unbeatable at processing enormous amounts of data, finding complex patterns, performing repetitive tasks, and working 24/7 without fatigue
• AI lacks real understanding, creativity, emotional intelligence, ethical judgment, and is brittle outside its trained domain
• Common AI problems: hallucinations (AI makes things up), bias (AI learns prejudices from data), brittleness (vulnerable to small changes), and the black box (hard to understand decisions)
• The smartest strategy is to combine AI’s strengths (speed, data processing) with human judgment (context, ethics, accountability)

Test your knowledge

4 questions · 100% correct to pass · Review your answers when done

Start quiz →