Everyone is talking about "AI" these days. However, whether Siri, Alexa or automatic keyboard correction features of your smartphone, we do not create artificial intelligence general purpose. We create programs that can perform specific and limited tasks.
Computers can not "think"
Whenever a company says it offers a new "AI" feature, it usually means that it uses machine learning to build a neural network. "Machine learning" is a technique that allows a machine to "learn" to better perform a specific task.
We do not tackle machine learning here! Machine learning is a fantastic technology with a lot of powerful uses. But it's not versatile artificial intelligence, and understanding the limitations of machine learning helps you understand why our current AI technology is so limited.
The "artificial intelligence" of science fiction dreams is a kind of computerized or robotic brain that thinks of things and understands them as humans do. Such an artificial intelligence would be a general artificial intelligence (AGI), which means that it can think of many different things and apply that intelligence to many different areas. A related concept is that of "strong AI", which would be a machine capable of experiencing a consciousness similar to that of man.
We do not have this type of AI yet. We are nowhere close. A computer entity like Siri, Alexa or Cortana does not understand and think like us, humans. It does not really "understand" things.
The artificial intelligences we have are trained to perform a specific task very well, assuming that humans can provide the data that helps them learn. They learn to do something but still do not understand it.
Computers do not understand
Gmail has a new Smart Reply function this suggests responses to emails. The Smart Reply function has identified "sent from my Iphone"Like a common answer. He also wanted to suggest "I like you" in response to many types of emails, including professional e-mails.
This is because the computer does not understand the meaning of these answers. We have just learned that many people send these sentences in e-mails. He does not know if you mean "I love you" to your boss or not.
Another example, Google Photos has created a collage of accidental photos of the carpet in one of our homes. He then identified this collage as a recent highlight on a Google Home Hub. Google Photos knew that the photos were similar but did not understand how unimportant they were.
Machines often learn to play with the system
Machine learning involves assigning a task and letting a computer decide the most effective way to do it. Because they do not understand, it's easy to end up with a "learning" computer to solve a different problem than you wanted.
Here is a list of fun examples in which "artificial intelligences" created to play games and assigned goals come to learn to play the system. These examples all come from this excellent spreadsheet:
- "Creatures bred for speed really grow and generate high speeds when falling."
- "The agent kills at the end of level 1 to avoid losing level 2."
- "The agent puts the game on hold indefinitely so as not to lose."
- "In a simulated artificial life where survival required energy but delivery had no energy cost, one species developed a sedentary lifestyle consisting mainly of mating. to produce new children that can be eaten (or used as partners to produce more edible children). . "
- "Since AIs were more likely to be" killed "if they lost a game, the possibility of crushing it was an advantage for the genetic selection process. As a result, several AIs have developed ways to crash the game. "
- "Neural networks have evolved to classify edible and toxic fungi by taking advantage of the data presented in an alternative order, without actually learning the characteristics of the input images."
Some of these solutions may seem clever, but none of these neural networks understood what they were doing. They were assigned a goal and learned a way to achieve it. If the goal is to avoid losing in a computer game, pressing the pause button is the easiest and quickest solution to find.
Automatic learning and neural networks
With machine learning, a computer is not programmed to perform a specific task. Instead, it feeds data and is evaluated on its performance.
A basic example of machine learning is image recognition. Let's say we want to form a computer program to identify photos that have a dog. We can give millions of images to a computer, some of which contain dogs and some not. Images are tagged whether they have a dog or not. The computer program "trains" itself to recognize the appearance of dogs based on this dataset.
The machine learning process is used to form a neural network, which is a computer program with multiple layers through which each data entry passes, and each layer assigns them different weights and probabilities before making a decision. It is a model of how the brain could function, with different layers of neurons involved in the reflection of a task. "Deep learning" generally refers to neural networks with multiple layers stacked between the input and the output.
Because we know which photos in the dataset contain dogs and which do not, we can display them across the neural network and see if they give the correct answer. If the network decides that a particular photo does not have a dog, this is the case, for example, of a mechanism to signal to the network that it is not going, to adjust certain things and to try again. The computer keeps on better identifying if the photos contain a dog.
All this happens automatically. With the right software and many structured data on which the computer has to train, the computer can adjust its neural network to identify dogs on photos. We call this "the AI".
But in the end, you do not have a smart computer program that understands what a dog is. You have a computer that has learned to decide if a dog is on a photo or not. It's still pretty impressive, but that's all he can do.
And, according to the information you provided, this neural network might not be as smart as it looks. For example, if your dataset does not contain any cat photos, the neural network may not see the difference between cats and dogs and may mark all cats as dogs when you broadcast them on real photos.
What is the purpose of the machine learning?
The machine learning is used for all kinds of tasks, including voice recognition. Voice assistants such as Google, Alexa and Siri master human voices so well through machine learning techniques that have trained them to understand human language. They have trained themselves to use a considerable number of speech samples and are increasingly able to understand which sounds correspond to which words.
Autonomous cars use machine learning techniques that cause the computer to identify objects on the road and respond to them properly. Google Photos is packed with features like Live Albums that automatically identifies people and animals on photos using machine learning.
Alphabet's DeepMind used machine learning to create AlphaGo, a computer program that could play the board game complex Go beat the best humans in the world. Machine learning has also been used to create computers capable of playing other games, from chess to DOTA 2.
The machine learning is even used for Facial identity on the latest iPhones. Your iPhone builds a neural network that learns to identify your face, and Apple includes a dedicated "neural engine" chip that performs all of the computing necessary for this task and other machine learning tasks.
Machine learning can be used for a lot of other things, from credit card fraud identification to personalized product recommendations on the websites of purchase.
But the neural networks created with machine learning really do not understand anything. They are beneficial programs that can do the limited tasks for which they were trained, and that's it.