🧠AI Foundations
Your progress0%
0 of 49 lessons
Reading10 min·Lesson 1 of 5

AI Hype and How to Spot It

AI is everywhere right now — in headlines, job ads, and WhatsApp forwards. But not every claim about AI is true. Learning to separate real capability from hype is one of the most valuable skills you can build today.

What Is AI Hype?

Hype is when people exaggerate what a technology can do — usually to sell something, attract investment, or generate excitement. AI hype is especially common right now because the technology is genuinely new and impressive. That makes it easy to stretch the truth without most people noticing.

You have probably seen claims like: "This AI will replace all customer service jobs in two years," or "Our AI app can predict which students will pass their KCSE exams with 95% accuracy." These claims may have a grain of truth — but they often leave out important details about cost, failure rates, and real-world conditions.

👁️
Key idea: Hype is not always a lie. It is often a half-truth — one impressive fact used to hide ten limitations. Your job is to ask for the rest of the story.

Five Hype Signals to Watch For

  • No numbers, or suspiciously round numbers. "Our AI is 10x more accurate" — compared to what? Measured how? "90% accuracy" sounds great until you learn the baseline was 50%.
  • Lab results presented as real-world results. An AI that works perfectly on a dataset of 1,000 hospital records in the US may fail badly on records from Kenyatta National Hospital, which look completely different.
  • The word "always" or "never." Real AI systems have edge cases. Any claim of perfection should raise your suspicion immediately.
  • Missing failure stories. Every honest AI deployment has failures. If a company only shares success stories, ask what they are not telling you.
  • Replacing humans vs. assisting humans. Headlines love "AI replaces doctors/lawyers/teachers." In practice, most useful AI tools assist skilled people — they do not replace them.

A Kenyan Example: AgriTech AI Claims

Imagine a startup announces: "Our AI can tell any Kenyan farmer whether to plant maize or beans — just send us a photo of your soil." Before you trust it, ask:

  • Was the AI trained on Kenyan soil data, or data from another country?
  • How was accuracy measured — in a controlled trial, or in real farms across different counties?
  • What happens when the photo is blurry, or taken at the wrong time of day?
  • Who is responsible if the AI gives bad advice and the farmer loses their crop?
💡
Practical tip: When you see an impressive AI demo, always ask: "Would this still work in Mombasa? In Kisumu? On a KES 2,000 phone with patchy data?" If the demo was done in a Silicon Valley office on a fast laptop, the answer is often no.

How to Evaluate AI Claims Confidently

  1. "What exactly does it do?" — Get specific. "Analyzes data" means nothing. "Reads a loan application and flags missing documents" is something you can evaluate.
  2. "How was it tested?" — Look for independent evaluations, not just the company's own demos. A test done by the people selling the product is almost always too optimistic.
  3. "Who is harmed if it is wrong?" — An AI that sometimes recommends the wrong Spotify playlist is low-stakes. An AI that recommends who gets a KCB bank loan or who is flagged by the police is very high-stakes. The stakes change how much scrutiny is needed.
🧠
Remember: Healthy skepticism is not cynicism. You are not saying AI is bad — you are saying AI must earn your trust with evidence, just like any other tool or person you rely on.

Where Hype Comes From

Understanding why hype exists helps you spot it faster. AI hype usually comes from three places: investors who need a growing story to justify funding, journalists who need an exciting headline, and companies that need to look cutting-edge to win contracts. None of these groups are necessarily dishonest — they just have incentives that push them toward optimism.

The people who tend to be most honest about AI limitations are the engineers building it day-to-day. If you can find an engineer or researcher's blog post, a technical paper, or a post-mortem of a failed AI project, you will get a much more accurate picture than any press release.