David Ferrucci, the creator of IBM's Watson—the computer that famously beat human champions on Jeopardy!—is now tackling a much harder problem: giving artificial intelligence (AI) common sense.
Why Is This So Difficult?
AI systems today can do amazing things like recognize faces, answer questions, and even drive cars. They learn patterns from huge amounts of data, which makes them great at tasks like speech recognition or object detection. But they often fail at simple reasoning tasks that humans find obvious.
For example, we know plants need sunlight to grow. A smart speaker like Alexa can tell you plant facts, but it doesn't "understand" why putting a plant in a dark room is bad. Similarly, a self-driving car knows how to avoid obstacles, but it doesn't truly understand why avoiding a person is more important than avoiding a traffic cone.
This gap—between knowing facts and understanding meaning—is what Ferrucci calls the missing piece of AI: common sense.
Ferrucci's New Approach
Ferrucci founded a company called Elemental Cognition to build AI that can reason like humans. His system, CLARA (Collaborative Learning and Reading Agent), learns by reading stories and asking questions—just like a curious student.
For example, after reading about Fernando and Zoey's plants, CLARA asks: "Does it make sense that Fernando put his plant in the window because he wants it healthy?" This question seems simple to us, but for AI, it's a big leap toward understanding cause and effect.
CLARA combines two methods:
- Deep learning: finding patterns in huge amounts of data.
- Logical rules: basic facts like "plants need light" and "windows let light in."
It also crowdsources knowledge from people and uses reasoning to fill gaps. This mix helps CLARA build a kind of "common-sense library."
Why Does Common Sense Matter?
Experts say common sense is essential for AI to truly understand language, make decisions, and interact safely with the world. Without it, even advanced systems make silly mistakes—like guessing "Toronto" for a question about U.S. cities.
Ferrucci believes that teaching machines to understand meaning, not just memorize facts, will transform AI. It could lead to smarter assistants, safer self-driving cars, and better tools for learning. But it's a huge challenge. Common sense involves understanding time, cause and effect, and even social behavior—things humans learn naturally but machines struggle with.
The Road Ahead
Ferrucci's team is still working on CLARA, and other researchers are trying similar approaches. Some use deep learning combined with logic, while others explore new ways to teach machines about everyday life. Everyone agrees it's hard, but also important. As one expert said, "Language isn't just patterns—it connects to meaning and reasoning."
If AI can learn common sense, it won't just answer questions—it will truly understand them. That's the next big step in artificial intelligence.