Hey there, tech enthusiasts! Let’s unravel one of the most talked-about questions: What can AI actually do? And what is it terrible at? Trust me, the line is blurrier than you might think. If you’re diving into one AI project a year, it’s like sampling a dish once a year. You’re not really developing a palate for what’s good or not-so-good. So, let’s speed up that learning curve with some real-world examples.
Concrete Examples to Hone Your Intuition
Self-Driving Cars: A Tale of Two Scenarios
What AI Nails
Let’s kick off with self-driving cars. Your car sees what’s in front of it—maybe through a camera or other radars and sensors—and then decides, “Hey, there’s a car over there!” AI has gotten pretty darn good at this. Companies have figured out how to gobble up tons of data and churn out algorithms that make our cars smarter every day.
What AI Flubs
But can AI tell if the person gesturing at your car wants you to stop or is just saying hi? Nope, not really. You see, human gestures are incredibly diverse. The sheer volume of different ways a person could wave, point, or gesture is a data-collecting nightmare. Even if you could, making an AI system comprehend human intentions with 100% accuracy is still a pipe dream.
AI in Medical Diagnosis: The X-Ray Saga
A+ for Effort
Another cool example is diagnosing pneumonia through chest X-rays. Feed AI an X-ray, and it can tell you whether the patient has pneumonia or not. Yep, it’s that capable.
But Don’t Get Your Hopes Up
Ask that same AI to diagnose pneumonia from 10 images of a medical textbook chapter explaining the condition, and it’ll go haywire. Humans can look at a small set of images and understand paragraphs from a medical textbook. But for AI, that’s not a straightforward A-to-B problem it can solve.
Strengths and Weaknesses: Breaking It Down
Where AI Shines
- Simple Concepts: If you can think it up in less than a second, AI can probably learn it.
- Abundant Data: The more data, the merrier.
Where AI Stumbles
- Complex Concepts: AI struggles with complex problems, especially if there’s limited data.
- New Types of Data: If the AI hasn’t seen it before, good luck getting accurate results.
One Example to Rule Them All: The Fragile AI in Medical Diagnosis
Suppose you’ve built an AI that’s fantastic at diagnosing pneumonia from high-quality X-rays. You think you’ve hit the jackpot, right? Now, move this AI to a different hospital where the X-ray images aren’t as high-quality or standardized. Your AI could misdiagnose left and right. It’s not that adaptable.
Humans can adjust to these nuances easily, but AI? Not so much. That’s one area where AI is significantly weaker than us humans.
Don’t Worry If You’re Still Confused
Look, it’s normal to find the limitations of AI a bit fuzzy. Heck, even experts can’t instantly tell whether something is doable. It takes some time, a bit of trial and error, and a bunch of technical digging to get there.
