What Neural Networks, Artificial Intelligence, and Machine Learning Really Do
When an app claims to be powered by “artificial intelligence”, it feels like you are in the future. But what does this actually mean? We’re exploring what buzzwords like AI, machine learning, and neural networks really mean, and if they help improve your apps.
More recently, Google and Microsoft have added neural network training to their translation apps . Google said it is using machine learning to suggest playlists . Todoist says it uses AI to tell you when you should finish a task . Any.do claims that its AI bot can do some of the tasks for you . All of this just from last week. Some of these are marketing bullshit to make the new features look impressive, but sometimes the changes are really helpful. Artificial intelligence , machine learning and neural networks describe the ways in which computers can perform more complex tasks and learn in their environment. While you may hear they are used interchangeably by application developers, in practice they can be very different.
Neural networks analyze complex data to simulate the human brain
Artificial neural networks (ANNs, or simply “neural networks” for short) refer to a specific type of learning model that mimics how the synapses in your brain work. Traditional computing uses a series of logical operators to accomplish a task. Neural networks, on the other hand, use a network of nodes (which act like neurons) and edges (which act as synapses) to process data. The input is then passed through the system and a series of outputs are generated.
This output is then compared with known data. For example, let’s say you want to teach a computer to recognize an image of a dog. You browse millions of dog images online to see which images look like dogs. The person will then confirm which images are actually dogs. The system then gives preference to paths through the neural network that led to the correct answer. Over time and millions of iterations, the network will eventually improve the accuracy of its results.
To see how this works in action, you can try Google Quick, Draw! experiment here . In this case, Google trains the network to recognize the scribbles. He compares the drawing you draw with examples drawn by other people. The networks tell what doodles are and then teach them to recognize future doodles based on what the past looks like. Even if you have lame drawing skills (like mine), the net is pretty good at recognizing basic shapes like submarines, houseplants, and ducks.
Neural networks aren’t good for everything, but they are great at handling complex data. Google and Microsoft using neural networks in their translation applications are really interesting because translating into languages is difficult . We’ve all seen broken translations, but training the neural network can allow the system to learn from the correct translations to get better over time. We’ve seen this happen with voice transcriptions. Since the introduction of neural network training in Google Voice, the number of transcription errors has decreased by 49% . You might not notice it right away and it won’t be perfect, but this type of training does improve complex data analysis, which can lead to more natural functionality in your applications.
Machine learning teaches computers to improve with practice
Machine learning is a broad term that encompasses everything related to teaching a machine to improve a problem on its own. In particular, this applies to any system in which the performance of a machine on a task is improved solely by more experience with the task . Neural networks are an example of machine learning, but they are not the only way to learn a machine.
For example, one alternative machine learning technique is called reinforcement learning . In this method, the computer performs a task and then evaluates the result. The above video from Android Authority uses a chess game as an example. The computer plays chess completely and then either wins or loses. If he wins, then he assigns the winnings to the series of moves used during this game. After playing millions of games, the system can determine which moves are most likely to win based on the results of those games.
While neural networks are good for things like pattern recognition in images, other types of machine learning can be more useful for different tasks, such as determining what kind of music you like. Namely, Google claims that its music app will find you the music you want, when you want it . It does this by choosing playlists for you based on your past behavior. If you ignore his suggestions, it will (presumably) be considered a failure. However, if you choose one of the proposals, the process he used for that proposal will be marked as successful, so it reinforces the process that led to this proposal.
In such cases, you might not get the full benefits of machine learning if you don’t use this feature frequently. The first time you open the Google music app, your recommendations are likely to be rather scattered. The more you use it, the more accurate your suggestions will be. In any case, theoretically. Machine learning is not a panacea, so you can still get unnecessary recommendations. However, you are bound to get unwanted recommendations if you only open the music app every six months. Without regular use to help him learn, machine learning suggestions are not much better than regular smart suggestions. The buzzword “machine learning” is vague than neural networks, but it still implies that the software you use will use your feedback to improve its performance.
Artificial intelligence means everything that’s smart
Just like neural networks are a form of machine learning, machine learning is a form of artificial intelligence. However, the category of what is still considered “artificial intelligence” is so poorly defined that it is almost meaningless. While this creates a mental image of futuristic science fiction, in fact, we have already reached the milestones that were previously considered the realm of the future of AI. For example, it was once thought that OCR was too difficult for a machine, but now an app on your phone can scan documents and turn them into text . The description of such a now basic task as AI sounds more impressive than it actually is.
The reason basic telephony tasks can be thought of as AI is because there are actually two very different categories of AI. Weak or narrow AI describes any system designed for a narrow task or set of tasks. For example, Google Assistant and Siri, while powerful, are designed to perform a very narrow set of tasks. Namely, take a specific series of voice commands and return responses or run applications. These features are used in artificial intelligence research, but it is still considered “weak”.
In contrast, strong AI – otherwise known as general artificial intelligence or “complete AI” – is a system that can perform any task that a human can perform. It doesn’t exist either. If you were hoping your to-do list app would runon a cute robot voiced by Alan Tudik , it’s still a long way off. Since almost any AI that you would actually use is considered weak AI, the phrase “artificial intelligence” in the application description actually means “this is a smart application”. You may get some interesting suggestions, but don’t expect them to rival human intelligence.
While the semantics may be fuzzy, the field of AI research is so rewarding that you’ve probably already incorporated it into your daily life. Every time your phone automatically remembers where you parked , recognizes faces in your photos, gets tips for finding and automatically groups all of your vacation photos , you are directly or indirectly receive the benefits of research AI. To a certain extent, “artificial intelligence” actually means apps are getting smarter, which is to be expected. However, machine learning and neural networks are uniquely suited to improve certain kinds of problems. If an application simply says it uses “AI,” that’s less meaningful than any type of machine learning.
It’s also worth noting that neural networks and machine learning are not created equal. To say that an application uses machine learning to do something better is like saying that the camera is better because it is “digital”. Yes, digital cameras can do what film cameras cannot, but that doesn’t mean that every digital photo is better than any film photo. It’s all about how you use it. Some companies will be able to develop powerful neural networks that do really complex things that make your life better. Others will stick a machine learning label on a feature that already offered smart suggestions, and you will ignore it anyway.
Behind the scenes, machine learning and neural networks are very interesting. However, if you read the description of an application that uses these phrases, you can simply read it as “This feature is probably a little smarter” and continue doing what you have always done: rate applications for their usefulness to you.