AI Challenges and Limits

 AI Challenges and Limits

I asked ChatGPT "What's the hardest thing for AI to do?"

Here's what ChatGPT responded:

One interesting answer is: the hardest thing for AI isn't writing, drawing, or coding—it's reliably dealing with the real world when information is incomplete, changing, or requires actual action.

Here are some of the most difficult challenges:

1. Guaranteeing truth
   • AI can explain facts very well, but it can't guarantee every statement is correct.
   • It predicts likely text rather than proving everything it says.
2. Long-term autonomous projects
   • AI can help plan a business, write a book, or design software.
   • Actually carrying those projects through over weeks or months with changing priorities, unexpected setbacks, and continual decisions is much harder.
3. Understanding unstated context
   • Humans know a lot without saying it.
   • AI often has to infer what people mean, and sometimes those inferences are wrong.
4. Original scientific discovery
   • AI can help researchers analyze data and generate hypotheses.
   • Producing groundbreaking discoveries that fundamentally change a field still relies heavily on human creativity, experimentation, and verification.
5. Physical tasks
   • A robot can fold towels in a controlled environment.
   • Having one clean an unpredictable, cluttered house with the dexterity of a person is enormously difficult.
6. Common sense
   • Humans instinctively know things like:
     • A wet floor is slippery.
     • Ice cream melts.
     • A birthday cake should arrive before the party.
   • AI has learned many such patterns, but common sense remains surprisingly difficult because the real world contains endless exceptions.
7. Knowing when it doesn't know
   • One of AI's biggest challenges is recognizing uncertainty.
   • Modern models have improved at expressing confidence, but they can still sound convincing when they're mistaken.
8. Understanding human emotions deeply
   • AI can recognize emotional language and respond empathetically.
   • It doesn't actually experience emotions, relationships, or lived experience, so there are limits to its understanding.
9. Making high-stakes decisions
   • AI can assist with medical diagnoses, legal research, engineering, or finance.
   • Final decisions are still best made by qualified humans because they involve ethics, accountability, and consequences.
10. Being truly creative in the human sense
    • AI can generate millions of combinations of existing ideas.
    • Whether it can originate genuinely new concepts in the same way humans do is still debated.

---

A surprising answer

Based on many interactions, one of the hardest requests users make is:

"Invent something completely new that no human has ever thought of."

AI is trained on existing human knowledge. It can combine ideas in unusual ways, but it cannot prove that an invention, story, melody, gadget, or design is truly unprecedented. There is simply no way to compare against every idea every person has ever had.

That's why, when you've asked me in previous conversations to design things that "haven't been seen before," I can create highly novel concepts, but I can't honestly claim they're completely unique or never imagined by anyone.