Learn about pitfalls and how to avoid them.
- Artificial intelligence has been around since the early 1950s. Since then, research has gone up and down. In between these bursts of interests, there have been long cooling-off periods. Sometimes these are called the AI winters. Right now, machine learning and artificial neural networks have been in an upswing. This has led to a lot of activity around deep learning with large datasets. But not everyone thinks of data and pattern matching as the surest way to intelligence.
In fact, many people see this as an improvement of weak artificial intelligence. Machines are getting better at very narrow tasks. They can identify pictures and look for other patterns in massive datasets. That's why it's important to really understand the benefits of different AI approaches. Don't assume that the most popular AI approaches like deep learning and artificial neural networks will be the best fit for your organization. Remember, there was only a few decades ago when these approaches were considered outside the mainstream.
It's likely that in the next few years some of the most celebrated innovations in AI will be using deep learning or other data-centered approaches. But it's also likely that there'll be just as many smaller projects that use a different approach to AI. You shouldn't assume that because some of the largest AI companies are using one approach, that that approach will work the best for your programs. Imagine if you want to create a program that would recommend meals at a fast food restaurant.
You could certainly go through and process millions of different customer orders. Then you could use machine learning to find patterns in these orders. That way, when customer makes a request, the machine could match it to one of the other millions of orders. That's likely the approach that a company like Google, Microsoft, or IBM would have for the problem. If you're a smaller organization, then you might be able to build a very similar program using symbolic reasoning. You just have a few cashiers act as experts to build up a series of possible orders.
You wouldn't have to crunch all that data to come up with a workable system. Still, symbolic reasoning has challenges of its own. Much of what humans are good at cannot easily be described. Imagine if you wanted to create an AI program that could juggle three balls. It would be very difficult to find experts that could describe juggling in such detail that it could be programmed into a machine. Jugglers don't usually think in terms of dependent and independent variables. In these cases, machine learning is really the only option.
In general, when you're starting an AI project you can think of it in two ways. Does the program require abstract reasoning or does it really just need detailed pattern matching? For abstract reasoning and problems that can be easily described, you should stick with symbolic reasoning. These could be challenges like building software to help with taxes, looking over a legal contract, or checking medical prescriptions for errors. These are all challenges that can be broken down into decisions, actions, and symbols.
For problems that can not be easily described or for patterns that are difficult to see, you should stick with machine learning. These are some of the common challenges where you have a neural network classify massive amounts of data or find key relationships. The main thing you want to watch out for is that you apply the right approach to the right problem. If you use deep learning for an AI program that helps with your taxes, then you might run into some difficult challenges. For one, most of the decision-making will be done in a black box.
You won't necessarily know the machine's reasoning when it works on your tax returns. Also, a system like this would be very difficult to update. When you have the machine create its own models and algorithms, it's very difficult for a human to go into the system to make changes. The more you think about the strengths and weaknesses of these different approaches, the more likely you'll find an approach that works well. Don't assume that the current trends in AI will give you the most powerful approach or even the most valuable.
As of right now, weak AI is a very powerful but very narrow tool. The more time you spend figuring out what you want from this tool, the happier you'll be with the result.
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