What does it mean to think? Learn about having a machine sense, reason, and act.
- In the very beginning, artificial intelligence used something called the symbolic systems approach. This approach allowed machines to act in a way that seemed intelligent, but in reality it was just complex pattern matching. The symbolic systems approach allowed early programmers to create expert systems. These machines could do things that were normally left to experts. They could diagnose illnesses, give you a credit score, and even help you with your taxes.
The problem with these systems is that they created long lists of matching patterns. They also required experts to create these patterns. You needed someone who knew Chinese to match every possible phrase. Maybe you'd need a doctor who would come up with an answer to every one of a patient's questions. You had to create these long lists of matching patterns. Sometimes this is called combinatorial explosion. This type of effort made it extremely difficult to create robust expert systems.
There was also the very real possibility that you wouldn't find a match. A nurse might go through a dozen questions and only at the end find out that there's no match. Or a Chinese speaker might write a statement that was not included in any of the long lists of responses. These challenges made it really difficult for expert systems. Because of that, they started to disappear in the late 1980s. Still, the disappearance of expert systems didn't entirely eliminate the symbolic systems approach.
Instead, you can still commonly find this approach in what's called planning artificial intelligence. It's called planning artificial intelligence because you still have to plan out how to match your symbols and patterns. It tries to solve the problem of the long lists that you get with the expert systems. Instead, it uses something called heuristic reasoning. This reasoning gives artificial intelligence a form of common sense. It tries to limit what patterns the program has to match at any one time.
This is sometimes called limiting the search space. Think about how heuristic reasoning might work for the Chinese room argument. If you're trying to create a program with planning AI, then you might use heuristic reasoning to limit the possibilities of the first note. The program might expect something like "Hello" or "How are you?" Maybe even something like "Do you speak Chinese?" If the program is expecting these as a first note, then it limits how far it has to search to match the pattern.
One common place that you'll see AI planning is with navigation systems like Google Maps. Here you'll put in your starting location and your destination, and then the system will step through all the possibilities to find you the best route. It's still using a symbolic systems approach. All the roads and street names have been carefully mapped out. The system matches your destination with the patterns that it has in its list. You may have even seen one of the Google mapping cars driving on the road, collecting the street names of all the best routes.
These Google cars were creating their own lists. If you turn right at the stop sign, then you want to turn left when you reach the end of the road. The big difference is that a planning AI system like Google Maps uses this heuristic reasoning. Google Maps knows that there are only a few possibilities each time you make a turn. You won't be able to turn right onto a street that's hundreds of miles away. Even though it's considered old-fashioned AI, the symbolic systems approach and planning AI are still used in many new projects.
Planning AI performs really well in systems that have predefined symbols and patterns. You can see this with driving directions, but it also works well with legal contracts, logistics, and even video games. The trick is to make sure that your heuristic reasoning is accurate. Like most common sense, if you turn out to be wrong, then pretty much everything you assume aftewards will also be incorrect. Think about how your mapping software might react if early on you turn onto a road that hasn't been added to its list.
Even though you might use your mapping software every day, it still comes from symbolic system ideas that came out in the 1950s. A lot of newer AI projects just assume that they should use newer approaches. The newer approaches might not be the best fit for your organization. If you have a new AI project, don't be quick to dismiss the benefits of good old-fashioned AI.
This course will introduce you to some of the key concepts behind artificial intelligence, including the differences between "strong" and "weak" AI. You'll see how AI has created questions around what it means to be intelligent and how much trust we should put in machines. Instructor Doug Rose explains the different approaches to AI, including machine learning and deep learning, and the practical uses for new AI-enhanced technologies. Plus, learn how to integrate AI with other technology, such as big data, and avoid some common pitfalls associated with programming AI.
- The history of AI
- Machine learning
- Technical approaches to AI
- AI in robotics
- Integrating AI with big data
- Avoiding pitfalls