Everyone is talking about "AI" these days. But if you look at Siri, Alexa, or just the auto-correction features available on your smart keyboard, we do not create general intelligence for general purposes. We create programs that can perform specific, narrow tasks.
Computers Can not "Think"
When a company says it comes out with a new "AI" feature, it usually means that the company using machine learns to build a neural network. "Machine learning" is a technique that allows a machine to "learn" how to better perform a particular task.
We do not attack machine learning here! Machine learning is a great technique with very powerful applications. But it's not artificial intelligence for general use, and understanding the limitations of machine learning helps you understand why our current AI technology is so limited.
The "artificial intelligence" of sci-fi dreams is a computerized or robotic kind of brain that thinks of things and understands them as people do. Such artificial intelligence would be an artificial general intelligence (AGI), which means that it can think of several different things and apply that intelligence to several different domains. A related concept is "strong AI", which would be a machine that can experience human awareness.
We do not have that type of AI yet. We are not anywhere near it. A computer device like Siri, Alexa or Cortana does not understand and think like we humans do. It does not really understand "understand" things.
The artificial intelligence we have been trained to do a specific task very well provided that people can provide the information to help them learn. They learn to do something but still do not understand.
Computers Do not Understand
Gmail has a new "smart response" feature that suggests replies to emails. The Smart Response feature identified "Sent from my iPhone" as a regular response. It would also suggest "I love you" as an answer to many different types of emails, including work mail.
This is because the computer does not understand what these answers mean. It has only been learned that many people send these phrases in emails. It does not know if you want to say "I love you" to your boss or not.
As another example, Google Photos puts a collage of accidental images of the carpet in one of our homes. It then identified the collagen as a new highlight of a Google Home Hub. Google Photos knew the images were similar but did not understand how important they were.
Machines often learn to play the system
Machine learning is about assigning a task and letting a computer determine the most effective way to do it. Because they do not understand, it's easy to end up with a computer that learns how to solve a different problem than you would like.
Here is a list of fun examples where "artificial intelligences" were created to play games and assigned goals just learned to play the system. These examples are all from this excellent spreadsheet:
- "Creatures grown for speed grow very long and generate high speeds by falling over."
- "Agent kills at the end of level 1 to avoid losing in level 2."
- "Agent pauses the game endlessly to avoid losing."
- "In an artificial life simulation where survival demanded energy but the birth had no energy cost, a species developed a sedentary lifestyle that consisted most of mating to produce new children that could be eaten (or used as friends to produce more edible children) . "
- " Because AIs were more likely to be "killed" if they lost a game, the crash could be an advantage for the genetic selection process. Therefore, several AI's ways of crashing the game developed. "
- " Neurala networks developed to classify edible and poisonous fungi utilized data presented alternately and did not actually learn any characteristics of the submitted images. " Some of these solutions may sound smart, but none of these neural networks understood what they did. They were assigned a goal and learned a way to achieve it. If the goal is to avoid losing in a computer game, it's easiest to find the fastest solution they can find.
Machine Learning and Neural Networks
With machine learning, a computer is not programmed to perform a specific task. Instead, data is fed and evaluated on its performance at the task.
An elementary example of machine learning is image recognition. Let's say we want to train a computer program to identify images that have a dog in them. We can give a computer millions of pictures, some of which have dogs in them and some do not. The pictures are marked if they have a dog in them or not. The computer program "trains" itself to identify which dogs are looking based on that data set.
The machine learning process is used to train a neural network, which is a multi-layer computer program that passes through each data entry, and each layer assigns different weights and probabilities to them before they finally decide. It is modeled on how we think the brain can function, with different layers of neurons involved in thinking about a task. "Deep Learning" generally refers to neural networks with many layers stacked between input and output.
Since we know which photos in the dataset contain dogs and who can not, we can run the images via the Neural Network and see if they result in the correct answer. If the network determines a particular photo, it does not have a dog when it does, for example, there is a mechanism to tell the network that it was wrong, adjust some things and try again. The computer continues to better identify if pictures contain a dog.
All this happens automatically. With the right software and a lot of structured data for the computer to work out, the computer can set up its neural network to identify dogs in pictures. We call this "AI."
But at the end of the day you do not have an intelligent computer program that understands what a dog is. You have a computer that teaches you to decide if a dog is in a photo or not. It's still quite impressive, but that's all it can do.
And depending on the input you gave it, the neural network can not be as smart as it looks. For example, if there were no photos of cats in your dataset, the neural network may not see the difference between cats and dogs and can label all cats as dogs when you release it on people's correct photos.
What is Machine Learning used for?
Machine learning is used for all types of tasks, including speech recognition. Voice assistants like Google, Alexa and Siri are so good to understand human voices because of machine learning techniques who have trained them to understand human speech. They have trained themselves on a massive amount of human speech samples and become better and better at understanding what sounds that match those words.
Self-propelled cars use machine learning techniques that train your computer to identify items along the way and how to respond correctly. Google Photos is full of features like Live Albums that automatically identify people and animals on photos using machine learning.
Alphabet's DeepMind machine that teaches to create AlphaGo, a computer program that can play the complex board game Go and beat the best people in the world. Machine learning has also been used to create computers that are good for playing other games, from chess to DOTA 2.
Machine learning is even used for Face ID on the latest iPhones. Your iPhone designs a neural network that teaches you to identify your face, and Apple includes a dedicated "Neural Motor" chip that performs all the number requisites for this and other machine learning tasks.
Machine learning can be used for many other things, from identifying credit card fraud to personal product recommendations on shopping websites.
But the neural networks created with machine learning do not understand anything. They are beneficial programs that can perform the narrow tasks they were trained for, and that is it.
Image Credit: Phonlamai Photo / Shutterstock.com, Tatiana Shepeleva / Shutterstock.com, Miscellaneous Photography / Shutterstock.com.