Machine learning

- Imagine a computer that didn't need to be programmed. The system could learn the same way you do just by observing the world. You've seen that earlier AI systems used a symbolic approach. The idea was that if the system recognized symbols then it would start to seem intelligent. One of the key challenges was that programmers worked with experts to create the system. That's why they were called expert systems. Later, computer scientists gave up on this approach because it created too many combinations. They decided that you couldn't just program intelligence into a system but maybe you could program a system to become intelligent through observation. It wouldn't feel, hear or see or taste like a human. Instead, it would learn by sensing data. In 1959, a computer scientist named Arthur Samuel created a checkers program that could learn by playing against itself. It played both sides of the board and taught itself strategy through observation. The more the machine played the more it saw patterns on how to win. Computer scientists didn't program the machine to play checkers. It learned through its own experience. Arthur Samuel's called this idea machine learning. This was different from symbolic systems. No human program the moves and counter moves. Instead, the system was designed to learn and improve on its own. The system would quickly learn new checker strategies and after a short period of time it consistently beat its programmer. Machine learning was a breakthrough discovery. There was only one downside. These systems could play games, but in the 1950s there wasn't that much digital data. Remember that machine learning uses data as its five senses. So without data, it could only find the simplest patterns but that all changed. At the beginning of the 1990s, the explosion of the internet suddenly had everyday people creating huge amounts of data. The 1990s was a time of explosive growth for machine learning systems. The new data was like water that poured over the dry fields of artificial intelligence. At this point, machine learning systems had the fuel they needed to become more intelligent. So if you wanted to teach the system on how to identify a cat, you had access to millions of cat pictures online. Computer scientists began to create newer machine learning algorithms. There were even some researchers that started to create systems that were designed to mimic the human brain. One of the big advantages of learning through data is that machines could continue to grow with more data. If the machine finds new patterns it can adapt to the new information. But it's important to keep in mind that you still run into some of the same challenges. The machine learning system is still just identifying patterns. Still, in the last few years machine learning has been the fastest growing area in AI. As the amount of data increases This area has even shown more promise. Organizations are constantly collecting vast amounts of new data. So now the big challenge has become figuring out what to do with all this information. In a sense, you have artificial intelligence systems looking through your data and letting your organization see what it finds.