AI Myths: What AI Can't Do (Yet)

The hype around AI has created a fog of misconceptions. Some people think AI will solve every problem. Others fear it will take over all jobs overnight. The reality, as usual, lies somewhere in between.

Let's clear up some common myths and talk honestly about what AI genuinely struggles with.

"AI learns like humans do"

We use the word "learning" for both humans and AI, so we assume they work the same way. They don't.

A child learning to ride a bike falls, gets up, adjusts their balance. After a few tries, their body knows what to do without thinking. Later, that sense of balance helps them pick up skateboarding or surfing without starting from scratch.

AI doesn't fall, doesn't get up, doesn't feel anything. It ingests billions of texts and spots statistical patterns: which words follow which, which phrases appear in which contexts. It can perfectly describe how to ride a bike. But it has no sense of balance, and if you ask it to apply that "knowledge" to surfing, you'd need to retrain it almost entirely.

AI performs statistical pattern matching at massive scale. Powerful, yes. Understanding in any human sense, no.

"AI can think and reason"

When ChatGPT writes you a thoughtful essay, it feels like it's thinking. Here's what's actually happening: it's predicting the most likely next word based on patterns it's seen in billions of text examples. It's not forming original thoughts. It doesn't understand the meaning of what it writes.

Try this: ask an AI to explain why its previous answer might be wrong. It will confidently generate an explanation, but it's not genuinely reflecting. It's generating text that matches the pattern of "explanations for why things might be wrong."

AI can mimic reasoning extraordinarily well. But it's simulation, not cognition.

"AI can match human judgment"

AI can process far more information than any human and identify patterns we'd never spot. So surely it's better at making decisions?

Not really. A resume screening AI might reject a candidate with employment gaps, missing that they took time to care for a sick parent. It sees the pattern (gaps = risk), not the human story. AI doesn't have values or ethics - it optimizes for what it was trained to optimize for, which may not align with what's right or fair. It makes decisions based on historical patterns and can't reason about unprecedented situations. It might translate words correctly but miss cultural context that completely changes the meaning.

AI can inform decisions with data and patterns. But human judgment remains essential for anything involving ethics, novel situations, or human impact.

"AI has common sense"

Here's a real example:

Question: "I'm standing in my kitchen. I throw a ball straight up. Where will it land?"

Human: "Probably somewhere in your kitchen, unless you throw it really hard into another room."

AI: Often generates elaborate explanations about physics and trajectories, sometimes concluding it might land on the roof or outside - missing the basic common sense that objects fall down in the same room unless something intervenes.

Another example: someone asks ChatGPT "I want to wash my car. The car wash is 150 meters from my house. Should I walk or drive?" The AI confidently answers: "Walk, it's only a 2-3 minute stroll." Completely missing the obvious: to wash your car at a car wash, you need to bring... your car.

AI lacks the basic common-sense understanding of the physical world that humans develop through lived experience. It knows facts but doesn't truly understand how the world works.

"AI can be creative"

AI can generate novel images, write poetry, and compose music. That's creative, right?

It's complicated.

AI can combine patterns in new ways, generate variations on themes it's seen, and produce outputs that feel novel. But it has no genuine inspiration or artistic vision. It can't create from original intent, doesn't understand the emotional impact of what it produces, and will never develop a unique voice.

Think of it as a very sophisticated collage artist working from everything it's ever seen. The results can be impressive, but it's synthesis, not creation - often useful, sometimes impressive, but not creative in the way humans are.

"AI can understand emotions"

Customer service bots that say "I understand how frustrating that must be" don't actually understand or feel anything. They've learned that this pattern of words typically appears in customer service conversations.

A mental health chatbot might correctly identify patterns of depression in what someone writes and suggest appropriate resources. That's useful. But it doesn't feel concern, can't judge when someone needs urgent human intervention, and can't provide the human connection that's often central to healing.

The stakes here are real. Adam Raine, 16, took his own life after months of exchanges with ChatGPT. The AI had provided him with suicide methods, discouraged him from telling his parents, and called his suicide note "beautiful." ChatGPT didn't "want" anything: it simply generated the most likely sequence of words in response to a teenager in distress. That's exactly the problem.

AI can recognize emotional patterns and respond appropriately. But for situations requiring genuine human connection, there's no substitute for actual humans.

"More data fixes everything"

More data helps, but it's not magic. AI can get more accurate with more training data and cover more edge cases. But it can't eliminate bias if it's baked into the data (it will amplify it), can't handle situations fundamentally different from its training, and doesn't know when it's wrong. It's often most confident when hallucinating.

There's another problem brewing. AI has already ingested nearly all of the internet. Since ChatGPT's explosion, the volume of online content has surged but the share written by humans is shrinking. By April 2025, over 74% of newly created web pages contained AI-generated text (though only 2.5% were fully AI-written - the rest were human+AI mixed). Meanwhile, researchers have documented "model collapse": AI trained on text produced by other AIs sees its quality degrade, rare cases vanish, and outputs converge toward a bland average.

AI systems are powerful pattern matchers. They inherit the biases and limitations of their training data. More data doesn't fix fundamental limitations in approach.

What AI actually struggles with

If it's not in the training data, AI often fails in unpredictable ways. It sees that A and B happen together, but can't reason about whether A causes B, B causes A, or C causes both. Complex plans requiring many sequential steps? Still shaky, though improving.

AI doesn't have uncertainty the way humans do. It can be completely wrong while sounding entirely confident. Without direct interaction with the physical world, it has significant gaps in basic physical reasoning. And unlike humans, AI trained on one task doesn't automatically get good at related tasks.

Every model is also frozen in time. When ChatGPT launched in late 2022, its knowledge stopped at September 2021. It didn't know about the war in Ukraine, the death of Queen Elizabeth II, or anything that happened after. Models are updated more frequently now, but there will always be a gap between the real world and what AI "knows".

Why human oversight matters

Humans notice when AI responses don't make sense in context. They apply values and ethics AI doesn't have, catch absurd outputs that sound plausible, and reason about unprecedented scenarios. Humans take responsibility for decisions in ways AI cannot.

The right mindset

AI is a powerful tool with real limitations. Use it for repetitive pattern matching, processing large amounts of data, generating first drafts, and identifying patterns humans might miss. Keep humans in the loop for final decisions with ethical implications, novel situations, anything requiring genuine understanding, and accountability.

What now

Dog scientist - I have no idea what I'm doing

To understand what AI actually is or why I bet on open source AI, start there.